remodnav/paper
2
0
Fork 0
Code Issues 1 Pull requests Releases Packages Activity Actions

Revision 2 #11

Merged
AsimHDar merged 124 commits from revision_2 into master 2025-08-19 08:44:09 +00:00
Conversation 5 Commits 124 Files changed 18 +24509 -709
18 changed files with 24509 additions and 709 deletions
Download patch file Download diff file Expand all files Collapse all files

411
EyeGaze.bib
View file

@ -1,411 +0,0 @@
Automatically generated by Mendeley Desktop 1.19.2
Any changes to this file will be lost if it is regenerated by Mendeley.
BibTeX export options can be customized via Options -> BibTeX in Mendeley Desktop
@article{Hannula2010,
abstract = {Results of several investigations indicate that eye movements can reveal memory for elements of previous experience. These effects of memory on eye movement behavior can emerge very rapidly, changing the efficiency and even the nature of visual processing without appealing to verbal reports and without requiring conscious recollection. This aspect of eye-movement based memory investigations is particularly useful when eye movement methods are used with special populations (e.g., young children, elderly individuals, and patients with severe amnesia), and also permits use of comparable paradigms in animals and humans, helping to bridge different memory literatures and permitting cross-species generalizations. Unique characteristics of eye movement methods have produced findings that challenge long-held views about the nature of memory, its organization in the brain, and its failures in special populations. Recently, eye movement methods have been successfully combined with neuroimaging techniques such as fMRI, single-unit recording, and MEG, permitting more sophisticated investigations of memory. Ultimately, combined use of eye-tracking with neuropsychological and neuroimaging methods promises to provide a more comprehensive account of brain-behavior relationships and adheres to the {\&}{\#}8220;converging evidence{\&}{\#}8221; approach to cognitive neuroscience.},
author = {Hannula, Deborah E. and Althoff, Robert R and Warren, David E and Riggs, Lily and Cohen, Neal J and Ryan, Jennifer D},
doi = {10.3389/fnhum.2010.00166},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Hannula et al. - 2010 - Worth a glance using eye movements to investigate the cognitive neuroscience of memory.pdf:pdf},
issn = {16625161},
journal = {Frontiers in Human Neuroscience},
keywords = {Amnesia,Eye Movements,Hippocampus,MEG,Memory,fMRI},
month = {oct},
pages = {166},
publisher = {Frontiers},
title = {{Worth a glance: using eye movements to investigate the cognitive neuroscience of memory}},
volume = {4},
year = {2010}
}
@article{Andersson2017,
abstract = {Almost all eye-movement researchers use algorithms to parse raw data and detect distinct types of eye movement events, such as fixations, saccades, and pursuit, and then base their results on these. Surprisingly, these algorithms are rarely evaluated. We evaluated the classifications of ten eye-movement event detection algorithms, on data from an SMI HiSpeed 1250 system, and compared them to manual ratings of two human experts. The evaluation focused on fixations, saccades, and post-saccadic oscillations. The evaluation used both event duration parameters, and sample-by-sample comparisons to rank the algorithms. The resulting event durations varied substantially as a function of what algorithm was used. This evaluation differed from previous evaluations by considering a relatively large set of algorithms, multiple events, and data from both static and dynamic stimuli. The main conclusion is that current detectors of only fixations and saccades work reasonably well for static stimuli, but barely better than chance for dynamic stimuli. Differing results across evaluation methods make it difficult to select one winner for fixation detection. For saccade detection, however, the algorithm by Larsson, Nystr{\"{o}}m and Stridh (IEEE Transaction on Biomedical Engineering, 60(9):2484-2493,2013) outperforms all algorithms in data from both static and dynamic stimuli. The data also show how improperly selected algorithms applied to dynamic data misestimate fixation and saccade properties.},
author = {Andersson, Richard and Larsson, Linnea and Holmqvist, Kenneth and Stridh, Martin and Nystr{\"{o}}m, Marcus},
doi = {10.3758/s13428-016-0738-9},
issn = {1554-3528},
journal = {Behavior Research Methods},
keywords = {Eye-tracking,Inter-rater reliability,Parsing},
month = {apr},
number = {2},
pages = {616--637},
pmid = {27193160},
title = {{One algorithm to rule them all? An evaluation and discussion of ten eye movement event-detection algorithms}},
volume = {49},
year = {2017}
}
@article{Friedman2018,
author = {Friedman, Lee and Rigas, Ioannis and Abdulin, Evgeny and Komogortsev, Oleg V.},
doi = {10.3758/s13428-018-1050-7},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Friedman et al. - 2018 - A novel evaluation of two related and two independent algorithms for eye movement classification during read(2).pdf:pdf},
issn = {1554-3528},
journal = {Behavior Research Methods},
month = {aug},
number = {4},
pages = {1374--1397},
publisher = {Springer US},
title = {{A novel evaluation of two related and two independent algorithms for eye movement classification during reading}},
volume = {50},
year = {2018}
}
@article{Hanke2016,
abstract = {A {\textless}i{\textgreater}studyforrest{\textless}/i{\textgreater} extension, simultaneous fMRI and eye gaze recordings during prolonged natural stimulation},
author = {Hanke, Michael and Adelh{\"{o}}fer, Nico and Kottke, Daniel and Iacovella, Vittorio and Sengupta, Ayan and Kaule, Falko R. and Nigbur, Roland and Waite, Alexander Q. and Baumgartner, Florian and Stadler, J{\"{o}}rg},
doi = {10.1038/sdata.2016.92},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Hanke et al. - 2016 - A studyforrest extension, simultaneous fMRI and eye gaze recordings during prolonged natural stimulation.pdf:pdf},
issn = {2052-4463},
journal = {Scientific Data},
keywords = {Attention,Cortex,Language,Neural encoding,Visual system},
month = {oct},
pages = {160092},
publisher = {Nature Publishing Group},
title = {{A studyforrest extension, simultaneous fMRI and eye gaze recordings during prolonged natural stimulation}},
volume = {3},
year = {2016}
}
@inproceedings{Holmqvist2012,
address = {New York, New York, USA},
author = {Holmqvist, Kenneth and Nystr{\"{o}}m, Marcus and Mulvey, Fiona},
booktitle = {Proceedings of the Symposium on Eye Tracking Research and Applications - ETRA '12},
doi = {10.1145/2168556.2168563},
isbn = {9781450312219},
keywords = {accuracy,data quality,eye movements,eye tracker,latency,precision},
pages = {45},
publisher = {ACM Press},
title = {{Eye tracker data quality}},
year = {2012}
}
@article{HantaoLiu2011,
abstract = {Since the human visual system (HVS) is the ultimate assessor of image quality, current research on the design of objective image quality metrics tends to include an important feature of the HVS, namely, visual attention. Different metrics for image quality prediction have been extended with a computational model of visual attention, but the resulting gain in reliability of the metrics so far was variable. To better understand the basic added value of including visual attention in the design of objective metrics, we used measured data of visual attention. To this end, we performed two eye-tracking experiments: one with a free-looking task and one with a quality assessment task. In the first experiment, 20 observers looked freely to 29 unimpaired original images, yielding us so-called natural scene saliency (NSS). In the second experiment, 20 different observers assessed the quality of distorted versions of the original images. The resulting saliency maps showed some differences with the NSS, and therefore, we applied both types of saliency to four different objective metrics predicting the quality of JPEG compressed images. For both types of saliency the performance gain of the metrics improved, but to a larger extent when adding the NSS. As a consequence, we further integrated NSS in several state-of-the-art quality metrics, including three full-reference metrics and two no-reference metrics, and evaluated their prediction performance for a larger set of distortions. By doing so, we evaluated whether and to what extent the addition of NSS is beneficial to objective quality prediction in general terms. In addition, we address some practical issues in the design of an attention-based metric. The eye-tracking data are made available to the research community {\textless}citerefgrp{\textgreater}{\textless}citeref refid="ref1"/{\textgreater}{\textless}/citerefgrp{\textgreater}.},
author = {Liu, Hantao and Heynderickx, Ingrid},
doi = {10.1109/TCSVT.2011.2133770},
isbn = {1051-8215},
issn = {10518215},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
keywords = {Eye tracking,image quality assessment,objective metric,saliency map,visual attention},
month = {jul},
number = {7},
pages = {971--982},
title = {{Visual attention in objective image quality assessment: Based on eye-tracking data}},
volume = {21},
year = {2011}
}
@article{Larsson2013,
abstract = {A novel algorithm for detection of saccades and postsaccadic oscillations in the presence of smooth pursuit movements is proposed. The method combines saccade detection in the acceleration domain with specialized on- and offset criteria for saccades and postsaccadic oscillations. The performance of the algorithm is evaluated by comparing the detection results to those of an existing velocity-based adaptive algorithm and a manually annotated database. The results show that there is a good agreement between the events detected by the proposed algorithm and those in the annotated database with Cohen's kappa around 0.8 for both a development and a test database. In conclusion, the proposed algorithm accurately detects saccades and postsaccadic oscillations as well as intervals of disturbances.},
author = {Larsson, Linnea and Nystr{\"{o}}m, Marcus and Stridh, Martin},
doi = {10.1109/TBME.2013.2258918},
isbn = {1558-2531 (Electronic)
0018-9294 (Linking)},
issn = {15582531},
journal = {IEEE Transactions on Biomedical Engineering},
keywords = {Eye-tracking,signal processing,smooth pursuit},
month = {sep},
number = {9},
pages = {2484--2493},
pmid = {23625350},
title = {{Detection of saccades and postsaccadic oscillations in the presence of smooth pursuit}},
volume = {60},
year = {2013}
}
@article{Hanke2014,
abstract = {A high-resolution 7-Tesla fMRI dataset from complex natural stimulation with an audio movie},
author = {Hanke, Michael and Baumgartner, Florian J. and Ibe, Pierre and Kaule, Falko R. and Pollmann, Stefan and Speck, Oliver and Zinke, Wolf and Stadler, J{\"{o}}rg},
doi = {10.1038/sdata.2014.3},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Hanke et al. - 2014 - A high-resolution 7-Tesla fMRI dataset from complex natural stimulation with an audio movie.pdf:pdf},
issn = {2052-4463},
journal = {Scientific Data},
keywords = {Auditory system,Functional magnetic resonance imaging,Language,Perception},
month = {may},
pages = {140003},
publisher = {Nature Publishing Group},
title = {{A high-resolution 7-Tesla fMRI dataset from complex natural stimulation with an audio movie}},
volume = {1},
year = {2014}
}
@article{Harris2014,
abstract = {Face-selective regions in the amygdala and posterior superior temporal sulcus (pSTS) are strongly implicated in the processing of transient facial signals, such as expression. Here, we measured neural responses in participants while they viewed dynamic changes in facial expression. Our aim was to explore how facial expression is represented in different face-selective regions. Short movies were generated by morphing between faces posing a neutral expression and a prototypical expression of a basic emotion (either anger, disgust, fear, happiness or sadness). These dynamic stimuli were presented in block design in the following four stimulus conditions: (1) same-expression change, same-identity, (2) same-expression change, different-identity, (3) different-expression change, same-identity, and (4) different-expression change, different-identity. So, within a same-expression change condition the movies would show the same change in expression whereas in the different-expression change conditions each movie would have a different change in expression. Facial identity remained constant during each movie but in the different identity conditions the facial identity varied between each movie in a block. The amygdala, but not the posterior STS, demonstrated a greater response to blocks in which each movie morphed from neutral to a different emotion category compared to blocks in which each movie morphed to the same emotion category. Neural adaptation in the amygdala was not affected by changes in facial identity. These results are consistent with a role of the amygdala in category-based representation of facial expressions of emotion.},
author = {Harris, Richard J and Young, Andrew W and Andrews, Timothy J},
doi = {10.1016/j.neuropsychologia.2014.01.005},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Harris, Young, Andrews - 2014 - Dynamic stimuli demonstrate a categorical representation of facial expression in the amygdala.pdf:pdf},
issn = {1873-3514},
journal = {Neuropsychologia},
keywords = {Emotion,Expression,Face,fMRI},
month = {apr},
number = {100},
pages = {47--52},
pmid = {24447769},
publisher = {Elsevier},
title = {{Dynamic stimuli demonstrate a categorical representation of facial expression in the amygdala.}},
volume = {56},
year = {2014}
}
@article{Toiviainen2014,
abstract = {We investigated neural correlates of musical feature processing with a decoding approach. To this end, we used a method that combines computational extraction of musical features with regularized multiple regression (LASSO). Optimal model parameters were determined by maximizing the decoding accuracy using a leave-one-out cross-validation scheme. The method was applied to functional magnetic resonance imaging (fMRI) data that were collected using a naturalistic paradigm, in which participants' brain responses were recorded while they were continuously listening to pieces of real music. The dependent variables comprised musical feature time series that were computationally extracted from the stimulus. We expected timbral features to obtain a higher prediction accuracy than rhythmic and tonal ones. Moreover, we expected the areas significantly contributing to the decoding models to be consistent with areas of significant activation observed in previous research using a naturalistic paradigm with fMRI. Of the six musical features considered, five could be significantly predicted for the majority of participants. The areas significantly contributing to the optimal decoding models agreed to a great extent with results obtained in previous studies. In particular, areas in the superior temporal gyrus, Heschl's gyrus, Rolandic operculum, and cerebellum contributed to the decoding of timbral features. For the decoding of the rhythmic feature, we found the bilateral superior temporal gyrus, right Heschl's gyrus, and hippocampus to contribute most. The tonal feature, however, could not be significantly predicted, suggesting a higher inter-participant variability in its neural processing. A subsequent classification experiment revealed that segments of the stimulus could be classified from the fMRI data with significant accuracy. The present findings provide compelling evidence for the involvement of the auditory cortex, the cerebellum and the hippocampus in the processing of musical features during continuous listening to music.},
author = {Toiviainen, Petri and Alluri, Vinoo and Brattico, Elvira and Wallentin, Mikkel and Vuust, Peter},
doi = {10.1016/J.NEUROIMAGE.2013.11.017},
issn = {1053-8119},
journal = {NeuroImage},
month = {mar},
pages = {170--180},
publisher = {Academic Press},
title = {{Capturing the musical brain with Lasso: Dynamic decoding of musical features from fMRI data}},
volume = {88},
year = {2014}
}
@article{Holsanova2006,
abstract = {The aim of this article is to compare general assumptions about newspaper reading with eye-tracking data from readers' actual interaction with a newspaper. First, we extract assumptions about the way people read newspapers from socio-semiotic research. Second, we apply these assumptions by analysing a newspaper spread; this is done without any previous knowledge of actual reading behaviour. Finally, we use eye-tracking to empirically examine so-called entry points and reading paths. Eye movement data on reading newspaper spreads are analysed in three different ways: the time sequence in which different areas attract attention is calculated in order to determine reading priorities; the amount of time spent on different areas is calculated in order to determine which areas have been read most; the depth of attention is calculated in order to determine how carefully those areas have been read. General assumptions extracted from the socio-semiotic framework are compared to the results of the actual behaviour of subjects reading the newspaper spread. The results show that the empirical data confirm some of the extracted assumptions. The reading paths of the five subjects participating in the eye-tracking tests suggest that there are three main categories of readers: editorial readers, overview readers and focused readers.},
author = {Holsanova, Jana and Rahm, Henrik and Holmqvist, Kenneth},
doi = {10.1177/1470357206061005},
issn = {1470-3572},
journal = {Visual Communication},
month = {feb},
number = {1},
pages = {65--93},
publisher = {Sage PublicationsSage CA: Thousand Oaks, CA},
title = {{Entry points and reading paths on newspaper spreads: comparing a semiotic analysis with eye-tracking measurements}},
volume = {5},
year = {2006}
}
@article{Gordon2006,
author = {Gordon, Peter C. and Hendrick, Randall and Johnson, Marcus and Lee, Yoonhyoung},
doi = {10.1037/0278-7393.32.6.1304},
issn = {1939-1285},
journal = {Journal of Experimental Psychology: Learning, Memory, and Cognition},
number = {6},
pages = {1304--1321},
title = {{Similarity-based interference during language comprehension: Evidence from eye tracking during reading.}},
volume = {32},
year = {2006}
}
@article{Tikka2012,
abstract = {We outline general theoretical and practical implications of what we promote as enactive cinema for the neuroscientific study of online socio-emotional interaction. In a real-time functional magnetic resonance imaging (rt-fMRI) setting, participants are immersed in cinematic experiences that simulate social situations. While viewing, their physiological reactions - including brain responses - are tracked, representing implicit and unconscious experiences of the on-going social situations. These reactions, in turn, are analysed in real-time and fed back to modify the cinematic sequences they are viewing while being scanned. Due to the engaging cinematic content, the proposed setting focuses on living-by in terms of shared psycho-physiological epiphenomena of experience rather than active coping in terms of goal-oriented motor actions. It constitutes a means to parametrically modify stimuli that depict social situations and their broader environmental contexts. As an alternative to studying the variation of brain responses as a function of a priori fixed stimuli, this method can be applied to survey the range of stimuli that evoke similar responses across participants at particular brain regions of interest.},
author = {Tikka, Pia and V{\"{a}}ljam{\"{a}}e, Aleksander and de Borst, Aline W. and Pugliese, Roberto and Ravaja, Niklas and Kaipainen, Mauri and Takala, Tapio},
doi = {10.3389/fnhum.2012.00298},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Tikka et al. - 2012 - Enactive cinema paves way for understanding complex real-time social interaction in neuroimaging experiments.pdf:pdf},
issn = {1662-5161},
journal = {Frontiers in Human Neuroscience},
keywords = {Brain-Computer-Interfaces,enactive cinema,generative storytelling,implicit interaction,real-time fMRI,social neuroscience,two-way feedback},
month = {nov},
pages = {298},
publisher = {Frontiers},
title = {{Enactive cinema paves way for understanding complex real-time social interaction in neuroimaging experiments}},
volume = {6},
year = {2012}
}
@techreport{Larsson2016,
author = {Larsson and Linn{\'{e}}a},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Larsson, Linn{\'{e}}a - 2016 - P O B o x 1 1 7 2 2 1 0 0 L u n d 4 6 4 6-2 2 2 0 0 0 0 Event Detection in Eye-Tracking Data for Use in Applic.pdf:pdf},
title = {{P O B o x 1 1 7 2 2 1 0 0 L u n d + 4 6 4 6-2 2 2 0 0 0 0 Event Detection in Eye-Tracking Data for Use in Applications with Dynamic Stimuli}},
url = {http://portal.research.lu.se/portal/files/6192499/8600514.pdf},
year = {2016}
}
@article{Nystrom2010AnData,
title = {{An adaptive algorithm for fixation, saccade, and glissade detection in eyetracking data}},
year = {2010},
journal = {Behavior Research Methods},
author = {Nystr{\"{o}}m, Marcus and Holmqvist, Kenneth},
number = {1},
month = {2},
pages = {188--204},
volume = {42},
publisher = {Springer-Verlag},
doi = {10.3758/BRM.42.1.188},
issn = {1554-351X}
}
@Article{Stampe1993,
author="Stampe, Dave M.",
title="Heuristic filtering and reliable calibration methods for video-based pupil-tracking systems",
journal="Behavior Research Methods, Instruments, {\&} Computers",
year="1993",
month="Jun",
day="01",
volume="25",
number="2",
pages="137--142",
abstract="Methods for enhancing the accuracy of fixation and saccade detection and the reliability of calibration in video gaze-tracking systems are discussed. The unique aspects of the present approach include effective low-delay noise reduction prior to the detection of fixation changes, monitoring of gaze position in real time by the operator, identification of saccades as small as 0.5{\textdegree} while eliminating false fixations, and a quick, high-precision, semiautomated calibration procedure.",
issn="1532-5970",
doi="10.3758/BF03204486"
}
@article{dorr2010variability,
title={Variability of eye movements when viewing dynamic natural scenes},
author={Dorr, Michael and Martinetz, Thomas and Gegenfurtner, Karl R and Barth, Erhardt},
journal={Journal of vision},
volume={10},
number={10},
pages={28--28},
year={2010},
publisher={The Association for Research in Vision and Ophthalmology},
doi={10.1167/10.10.28}
}
@Article{Duchowski2002,
author="Duchowski, Andrew T.",
title="A breadth-first survey of eye-tracking applications",
journal="Behavior Research Methods, Instruments, {\&} Computers",
year="2002",
month="Nov",
day="01",
volume="34",
number="4",
pages="455--470",
abstract="Eye-tracking applications are surveyed in a breadth-first manner, reporting on work from the following domains: neuroscience, psychology, industrial engineering and human factors, marketing/advertising, and computer science. Following a review of traditionally diagnostic uses, emphasis is placed on interactive applications, differentiating between selective and gaze-contingent approaches.",
issn="1532-5970",
doi="10.3758/BF03195475"
}
@book{holmqvist2011eye,
title={Eye tracking: A comprehensive guide to methods and measures},
author={Holmqvist, Kenneth and Nystr{\"o}m, Marcus and Andersson, Richard and Dewhurst, Richard and Jarodzka, Halszka and Van de Weijer, Joost},
year={2011},
publisher={OUP Oxford}
}
@inproceedings{munn2008fixation,
title={Fixation-identification in dynamic scenes: Comparing an automated algorithm to manual coding},
author={Munn, Susan M and Stefano, Leanne and Pelz, Jeff B},
booktitle={Proceedings of the 5th symposium on Applied perception in graphics and visualization},
pages={33--42},
year={2008},
doi={10.1145/1394281.1394287},
organization={ACM}
}
@article{LARSSON2015145,
title = "Detection of fixations and smooth pursuit movements in high-speed eye-tracking data",
journal = "Biomedical Signal Processing and Control",
volume = "18",
pages = "145 - 152",
year = "2015",
issn = "1746-8094",
doi = "https://doi.org/10.1016/j.bspc.2014.12.008",
author = "Linnéa Larsson and Marcus Nystr{\"{o}}m and Richard Andersson and Martin Stridh",
keywords = "Signal processing, Eye-tracking, Smooth pursuit",
abstract = "A novel algorithm for the detection of fixations and smooth pursuit movements in high-speed eye-tracking data is proposed, which uses a three-stage procedure to divide the intersaccadic intervals into a sequence of fixation and smooth pursuit events. The first stage performs a preliminary segmentation while the latter two stages evaluate the characteristics of each such segment and reorganize the preliminary segments into fixations and smooth pursuit events. Five different performance measures are calculated to investigate different aspects of the algorithm's behavior. The algorithm is compared to the current state-of-the-art (I-VDT and the algorithm in [11]), as well as to annotations by two experts. The proposed algorithm performs considerably better (average Cohen's kappa 0.42) than the I-VDT algorithm (average Cohen's kappa 0.20) and the algorithm in [11] (average Cohen's kappa 0.16), when compared to the experts’ annotations."
}
@Article{Komogortsev2013,
author="Komogortsev, Oleg V.
and Karpov, Alex",
title="Automated classification and scoring of smooth pursuit eye movements in the presence of fixations and saccades",
journal="Behavior Research Methods",
year="2013",
month="Mar",
day="01",
volume="45",
number="1",
pages="203--215",
abstract="Ternary eye movement classification, which separates fixations, saccades, and smooth pursuit from the raw eye positional data, is extremely challenging. This article develops new and modifies existing eye-tracking algorithms for the purpose of conducting meaningful ternary classification. To this end, a set of qualitative and quantitative behavior scores is introduced to facilitate the assessment of classification performance and to provide means for automated threshold selection. Experimental evaluation of the proposed methods is conducted using eye movement records obtained from 11 subjects at 1000 Hz in response to a step-ramp stimulus eliciting fixations, saccades, and smooth pursuits. Results indicate that a simple hybrid method that incorporates velocity and dispersion thresholding allows producing robust classification performance. It is concluded that behavior scores are able to aid automated threshold selection for the algorithms capable of successful classification.",
issn="1554-3528",
doi="10.3758/s13428-012-0234-9"
}
@Article{Zemblys2018,
author="Zemblys, Raimondas
and Niehorster, Diederick C.
and Holmqvist, Kenneth",
title="gazeNet: End-to-end eye-movement event detection with deep neural networks",
journal="Behavior Research Methods",
year="2018",
month="Oct",
day="17",
abstract="Existing event detection algorithms for eye-movement data almost exclusively rely on thresholding one or more hand-crafted signal features, each computed from the stream of raw gaze data. Moreover, this thresholding is largely left for the end user. Here we present and develop gazeNet, a new framework for creating event detectors that do not require hand-crafted signal features or signal thresholding. It employs an end-to-end deep learning approach, which takes raw eye-tracking data as input and classifies it into fixations, saccades and post-saccadic oscillations. Our method thereby challenges an established tacit assumption that hand-crafted features are necessary in the design of event detection algorithms. The downside of the deep learning approach is that a large amount of training data is required. We therefore first develop a method to augment hand-coded data, so that we can strongly enlarge the data set used for training, minimizing the time spent on manual coding. Using this extended hand-coded data, we train a neural network that produces eye-movement event classification from raw eye-movement data without requiring any predefined feature extraction or post-processing steps. The resulting classification performance is at the level of expert human coders. Moreover, an evaluation of gazeNet on two other datasets showed that gazeNet generalized to data from different eye trackers and consistently outperformed several other event detection algorithms that we tested.",
issn="1554-3528",
doi="10.3758/s13428-018-1133-5"
}
@article{real_world,
author = {Matusz, Pawel J. and Dikker, Suzanne and Huth, Alexander G. and Perrodin, Catherine},
title = {Are We Ready for Real-world Neuroscience?},
journal = {Journal of Cognitive Neuroscience},
volume = {31},
number = {3},
pages = {327-338},
year = {2019},
doi = {10.1162/jocn\_e\_01276},
note ={PMID: 29916793},
abstract = { Real-world environments are typically dynamic, complex, and multisensory in nature and require the support of top–down attention and memory mechanisms for us to be able to drive a car, make a shopping list, or pour a cup of coffee. Fundamental principles of perception and functional brain organization have been established by research utilizing well-controlled but simplified paradigms with basic stimuli. The last 30 years ushered a revolution in computational power, brain mapping, and signal processing techniques. Drawing on those theoretical and methodological advances, over the years, research has departed more and more from traditional, rigorous, and well-understood paradigms to directly investigate cognitive functions and their underlying brain mechanisms in real-world environments. These investigations typically address the role of one or, more recently, multiple attributes of real-world environments. Fundamental assumptions about perception, attention, or brain functional organization have been challenged—by studies adapting the traditional paradigms to emulate, for example, the multisensory nature or varying relevance of stimulation or dynamically changing task demands. Here, we present the state of the field within the emerging heterogeneous domain of real-world neuroscience. To be precise, the aim of this Special Focus is to bring together a variety of the emerging “real-world neuroscientific” approaches. These approaches differ in their principal aims, assumptions, or even definitions of “real-world neuroscience” research. Here, we showcase the commonalities and distinctive features of the different “real-world neuroscience” approaches. To do so, four early-career researchers and the speakers of the Cognitive Neuroscience Society 2017 Meeting symposium under the same title answer questions pertaining to the added value of such approaches in bringing us closer to accurate models of functional brain organization and cognitive functions.}
}
@Article{Hooge2018,
author="Hooge, Ignace T. C.
and Niehorster, Diederick C.
and Nystr{\"o}m, Marcus
and Andersson, Richard
and Hessels, Roy S.",
title="Is human classification by experienced untrained observers a gold standard in fixation detection?",
journal="Behavior Research Methods",
year="2018",
month="Oct",
day="01",
volume="50",
number="5",
pages="1864--1881",
abstract="Manual classification is still a common method to evaluate event detection algorithms. The procedure is often as follows: Two or three human coders and the algorithm classify a significant quantity of data. In the gold standard approach, deviations from the human classifications are considered to be due to mistakes of the algorithm. However, little is known about human classification in eye tracking. To what extent do the classifications from a larger group of human coders agree? Twelve experienced but untrained human coders classified fixations in 6 min of adult and infant eye-tracking data. When using the sample-based Cohen's kappa, the classifications of the humans agreed near perfectly. However, we found substantial differences between the classifications when we examined fixation duration and number of fixations. We hypothesized that the human coders applied different (implicit) thresholds and selection rules. Indeed, when spatially close fixations were merged, most of the classification differences disappeared. On the basis of the nature of these intercoder differences, we concluded that fixation classification by experienced untrained human coders is not a gold standard. To bridge the gap between agreement measures (e.g., Cohen's kappa) and eye movement parameters (fixation duration, number of fixations), we suggest the use of the event-based F1 score and two new measures: the relative timing offset (RTO) and the relative timing deviation (RTD).",
issn="1554-3528",
doi="10.3758/s13428-017-0955-x"
}
@ARTICLE{5523936,
author={O. V. {Komogortsev} and D. V. {Gobert} and S. {Jayarathna} and D. H. {Koh} and S. M. {Gowda}},
journal={IEEE Transactions on Biomedical Engineering},
title={Standardization of Automated Analyses of Oculomotor Fixation and Saccadic Behaviors},
year={2010},
volume={57},
number={11},
pages={2635-2645},
keywords={biomechanics;biomedical optical imaging;eye;image classification;medical image processing;oculomotor fixation;saccadic behaviors;standardization;eye movement classification algorithms;stimulus-evoked task;threshold-value selection;Standardization;Classification algorithms;Computer science;Logic;Humans;Visual system;Psychology;Permission;Brain injuries;Alzheimer's disease;Analysis;baseline;eye-movement classification;oculomotor behavior;Adolescent;Adult;Algorithms;Female;Fixation, Ocular;Humans;Male;Saccades;Young Adult},
doi={10.1109/TBME.2010.2057429},
ISSN={0018-9294},
month={Nov}}
@article{gorgolewski2016brain,
title={The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments},
author={Gorgolewski, Krzysztof J and Auer, Tibor and Calhoun, Vince D and Craddock, R Cameron and Das, Samir and Duff, Eugene P and Flandin, Guillaume and Ghosh, Satrajit S and Glatard, Tristan and Halchenko, Yaroslav O and others},
journal={Scientific Data},
volume={3},
pages={160044},
year={2016},
doi={10.1038/sdata.2016.44},
publisher={Nature Publishing Group}
}
@article{carl1987pursuits,
author = {Carl, J. R. and Gellman, R. S.},
title = {Human smooth pursuit: stimulus-dependent responses},
journal = {Journal of Neurophysiology},
volume = {57},
number = {5},
pages = {1446-1463},
year = {1987},
doi = {10.1152/jn.1987.57.5.1446},
note ={PMID: 3585475},
abstract = { We studied pursuit eye movements in seven normal human subjects with the scleral search-coil technique. The initial eye movements in response to unpredictable changes in target motion were analyzed to determine the effect of target velocity and position on the latency and acceleration of the response. By restricting our analysis to the presaccadic portion of the response we were able to eliminate any saccadic interactions, and the randomized stimulus presentation minimized anticipatory responses. This approach has allowed us to characterize a part of the smooth-pursuit system that is dependent primarily on retinal image properties. The latency of the smooth-pursuit response was very consistent, with a mean of 100 +/- 5 ms to targets moving 5 degrees/s or faster. The responses were the same whether the velocity step was presented when the target was initially stationary or after tracking was established. The latency did increase for lower velocity targets; this increase was well described by a latency model requiring a minimum target movement of 0.028 degrees, in addition to a fixed processing time of 98 ms. The presaccadic accelerations were fairly low, and increased with target velocity until an acceleration of about 50 degrees/s2 was reached for target velocities of 10 degrees/s. Higher velocities produced only a slight increase in eye acceleration. When the target motion was adjusted so that the retinal image slip occurred at increasing distances from the fovea, the accelerations declined until no presaccadic response was measurable when the image slip started 15 degrees from the fovea. The smooth-pursuit response to a step of target position was a brief acceleration; this response occurred even when an oppositely directed velocity stimulus was present. The latency of the pursuit response to such a step was also approximately 100 ms. This result seems consistent with the idea that sensory pathways act as a low-pass spatiotemporal filter of the retinal input, effectively converting position steps into briefly moving stimuli. There was a large asymmetry in the responses to position steps: the accelerations were much greater when the position step of the target was away from the direction of tracking, compared with steps in the direction of tracking. The asymmetry may be due to the addition of a fixed slowing of the eyes whenever the target image disappears from the foveal region. When saccades were delayed by step-ramp stimuli, eye accelerations increased markedly approximately 200 ms after stimulus onset.(ABSTRACT TRUNCATED AT 400 WORDS)}
}
@Article{Startsev2018,
author="Startsev, Mikhail
and Agtzidis, Ioannis
and Dorr, Michael",
title="1D CNN with BLSTM for automated classification of fixations, saccades, and smooth pursuits",
journal="Behavior Research Methods",
year="2018",
month="Nov",
day="08",
abstract="Deep learning approaches have achieved breakthrough performance in various domains. However, the segmentation of raw eye-movement data into discrete events is still done predominantly either by hand or by algorithms that use hand-picked parameters and thresholds. We propose and make publicly available a small 1D-CNN in conjunction with a bidirectional long short-term memory network that classifies gaze samples as fixations, saccades, smooth pursuit, or noise, simultaneously assigning labels in windows of up to 1 s. In addition to unprocessed gaze coordinates, our approach uses different combinations of the speed of gaze, its direction, and acceleration, all computed at different temporal scales, as input features. Its performance was evaluated on a large-scale hand-labeled ground truth data set (GazeCom) and against 12 reference algorithms. Furthermore, we introduced a novel pipeline and metric for event detection in eye-tracking recordings, which enforce stricter criteria on the algorithmically produced events in order to consider them as potentially correct detections. Results show that our deep approach outperforms all others, including the state-of-the-art multi-observer smooth pursuit detector. We additionally test our best model on an independent set of recordings, where our approach stays highly competitive compared to literature methods.",
issn="1554-3528",
doi="10.3758/s13428-018-1144-2"
}
@article{Schutz2011,
author = {Schutz, A. C. and Braun, D. I. and Gegenfurtner, K. R.},
doi = {10.1167/11.5.9},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Schutz, Braun, Gegenfurtner - 2011 - Eye movements and perception A selective review.pdf:pdf},
issn = {1534-7362},
journal = {Journal of Vision},
keywords = {eye,eye movement,motion,motion perception,noise,object recognition,perception,pursuit, smooth,saccades},
mendeley-groups = {EyeGaze},
month = {sep},
number = {5},
pages = {9--9},
publisher = {The Association for Research in Vision and Ophthalmology},
title = {{Eye movements and perception: A selective review}},
volume = {11},
year = {2011}
}

2
Makefile
View file

@ -1,6 +1,6 @@
all: main.pdf
main.pdf: main.tex tools.bib EyeGaze.bib results_def.tex figures
main.pdf: main.tex references.bib results_def.tex figures
latexmk -pdf -g $<
results_def.tex: code/mk_figuresnstats.py

378
code/mk_figuresnstats.py
View file

@ -136,7 +136,10 @@ label_map = {
# we need the distribution parameters of all algorithms and human coders
# in tables 3, 4, 5, 6 from Andersson et al., 2017. Well worth double-checking,
# I needed to hand-copy-paste from the paper.
# I needed to hand-copy-paste from the paper. The summary statistics were made
# publicly available in the file # matlab_analysis_code/20150807.mat in the
# original authors GitHub repository
# (https://github.com/richardandersson/EyeMovementDetectorEvaluation/blob/0e6f82708e10b48039763aa1078696e802260674/matlab_analysis_code/20150807.mat).
# The first two entries within each value-list belong to human coders
image_params = {
@ -227,6 +230,16 @@ lab_ids = ['22', '23', '24', '25', '26', '27', '28', '29', '30', '31',
'32', '33', '34', '35', '36']
# this used to be within confusion(), is global now because we also need it for Kappa()
# --> defines mapping between remodnav labels (strings) and andersson labels (ints)
anderson_remap = {
'FIX': 1,
'SAC': 2,
'PSO': 3,
'PUR': 4,
}
def get_durations(events, evcodes):
events = [e for e in events if e['label'] in evcodes]
# TODO minus one sample at the end?
@ -240,12 +253,6 @@ def confusion(refcoder,
stats):
conditions = ['FIX', 'SAC', 'PSO', 'PUR']
#conditions = ['FIX', 'SAC', 'PSO']
anderson_remap = {
'FIX': 1,
'SAC': 2,
'PSO': 3,
'PUR': 4,
}
plotter = 1
# initialize a maximum misclassification rate, to later automatically reference,
max_mclf = 0
@ -420,6 +427,263 @@ def savefigs(fig,
% ('%.1f' % max_mclf))
def quality_stats():
"""
Computes the percent of signal loss in raw data
Note: takes a while to run (30 subj. x 8 runs x 15 min rec with 1000Hz sr),
therefore, I'm just adding results here for now
and save the resulting histogram to the repository.
\newcommand{\avglosslab}{0.041005777923189775}
\newcommand{\avglossmri}{0.1507901497174581}
To include this computation in a run with Make, add this function to the
list of command invocations if the script is ran from the command line at the
end of the script.
"""
import datalad.api as dl
import matplotlib.pyplot as plt
datapath_mri = op.join('data', 'raw_eyegaze', 'sub-*', 'ses-movie', 'func',
'sub-*_ses-movie_task-movie_run-*_recording-eyegaze_physio.tsv.gz')
datapath_lab = op.join('data', 'raw_eyegaze', 'sub-*', 'beh',
'sub-*_task-movie_run-*_recording-eyegaze_physio.tsv.gz')
for (data, assoc) in [(datapath_lab, 'lab'),
(datapath_mri, 'mri')]:
infiles = glob(data)
for f in infiles:
dl.get(f)
# make sure we have 15 subjects' data
assert len(infiles) == 120
print("Currently processing data from {} sample".format(assoc))
# set sampling rate and px2deg
px2deg = 0.0266711972026 if assoc == 'lab' else 0.0185581232561
sr = 1000
# calculate percent signal loss across subjects and runs
losses = []
vels = []
for f in infiles:
data = np.recfromcsv(f,
delimiter='\t',
names=['x', 'y', 'pupil', 'frame'])
# all periods of signal loss are marked as nan in the data
signal_loss = np.sum(np.isnan(data['x'])) / len(data['x'])
losses.append(signal_loss)
velocities = cal_velocities(data=data,
px2deg=px2deg,
sr=sr)
vels.append(velocities)
print("Calculated velocities and losses for {} sample".format(assoc))
# average across signal losses in sample (mri or lab)
loss = np.nanmean(losses)
# print results as Latex command using 'assoc' as sample identifier in name
label_loss = 'avgloss{}'.format(assoc)
print('\\newcommand{\\%s}{%s}'
% (label_loss, loss))
# vels is a list of arrays atm
v = np.concatenate(vels).ravel()
if assoc == 'lab':
v_lab = v
elif assoc == 'mri':
v_mri = v
# plot velocities in a histogram on logscale
# create non-linear non-equal bin sizes, as x axis will be log
hist, bins, _ = plt.hist(v[~np.isnan(v)], bins=40)
plt.close()
logbins = np.logspace(1, # don't start with 0, does not make sense in logspace
np.log10(bins[-1]),
len(bins))
fig, ax = plt.subplots()
fig.set_figheight(3)
fig.set_figwidth(5)
ax.set_ylabel('frequency')
ax.set_xlabel('velocities (deg/s)')
plt.hist(v_mri[~np.isnan(v_mri)],
weights=np.zeros_like(v_mri[~np.isnan(v_mri)]) + 1. / (v_mri[~np.isnan(v_mri)]).size,
bins=logbins,
histtype='bar',
color='orangered',
alpha=0.5,
label='mri')
plt.hist(v_lab[~np.isnan(v_lab)],
weights=np.zeros_like(v_lab[~np.isnan(v_lab)]) + 1. / (v_lab[~np.isnan(v_lab)]).size,
bins=logbins,
histtype='bar',
color='darkslategrey',
alpha=0.5,
label='lab')
plt.legend(loc='upper right')
plt.xscale('log')
plt.savefig(op.join('img', 'velhist.svg'),
transparent=True,
bbox_inches="tight")
def flowchart_figs():
"""
Just for future reference: This is the subset of preprocessed and raw data
used for the flowchart of the algorithm. Not to be executed.
"""
import matplotlib.pyplot as plt
from scipy import signal
datapath = op.join('data', 'raw_eyegaze', 'sub-32', 'beh',
'sub-32_task-movie_run-1_recording-eyegaze_physio.tsv.gz')
data = np.recfromcsv(datapath,
delimiter='\t',
names=['x', 'y', 'pupil', 'frame'])
clf = EyegazeClassifier(
px2deg=0.0266711972026,
sampling_rate=1000.0)
velocities = cal_velocities(data=data, sr=1000, px2deg=0.0266711972026)
vel_subset_unfiltered = velocities[15200:17500]
p = clf.preproc(data)
# this is to illustrate PTn estimation and chunking
vel_subset = p['vel'][15200:17500]
fig, ax1 = plt.subplots()
fig.set_figheight(2)
fig.set_figwidth(7)
fig.set_dpi(120)
ax1.plot(
vel_subset,
color='black', lw=0.5)
plt.close()
# this is to illustrate preprocessing
fig, ax1 = plt.subplots()
fig.set_figheight(2)
fig.set_figwidth(7)
fig.set_dpi(120)
ax1.plot(
vel_subset,
color='black', lw=0.5)
ax1.plot(
vel_subset_unfiltered,
color='darkorange', ls='dotted', lw=0.5)
plt.close()
def _butter_lowpass(cutoff, fs, order=5):
nyq = 0.5 * fs
normal_cutoff = cutoff / nyq
b, a = signal.butter(
order,
normal_cutoff,
btype='low',
analog=False)
return b,
# Here is fixation and pursuit detection on Butterworth filtered subsets
lp_cutoff_freq = 4.0
sr=1000
# let's get a data sample with no saccade
win_data = p[16600:17000]
b, a = _butter_lowpass(lp_cutoff_freq, sr)
win_data['x'] = signal.filtfilt(b, a, win_data['x'], method='gust')
win_data['y'] = signal.filtfilt(b, a, win_data['y'], method='gust')
filtered_vels = cal_velocities(data=win_data, sr=1000, px2deg=0.0266711972026)
fig, ax1 = plt.subplots()
fig.set_figheight(2)
fig.set_figwidth(7)
fig.set_dpi(120)
ax1.plot(
filtered_vels,
color='black', lw=0.5)
plt.close()
def cal_velocities(data, sr, px2deg):
"""Helper to calculate velocities
sr: sampling rate
px2deg: conversion factor from pixel to degree
"""
velocities = (np.diff(data['x']) ** 2 + np.diff(
data['y']) ** 2) ** 0.5
velocities *= px2deg * sr
return velocities
def plot_raw_vel_trace():
"""
Small helper function to plot raw velocity traces, as requested by reviewer 2
in the second revision.
"""
import matplotlib.pyplot as plt
# use the same data as in savegaze() (no need for file retrieval, should be there)
infiles = [
op.join(
'data',
'raw_eyegaze', 'sub-32', 'beh',
'sub-32_task-movie_run-5_recording-eyegaze_physio.tsv.gz'),
op.join(
'data',
'raw_eyegaze', 'sub-02', 'ses-movie', 'func',
'sub-02_ses-movie_task-movie_run-5_recording-eyegaze_physio.tsv.gz'
),
]
# we need the sampling rate for plotting in seconds and velocity calculation
sr = 1000
# load data
for i, f in enumerate(infiles):
# read data
data = np.recfromcsv(f,
delimiter='\t',
names=['x', 'y', 'pupil', 'frame'])
# subset data. Hessels et al., 2017 display different noise levels on 4
# second time series (ref. Fig 10). That still looks a bit dense, so we
# go with 2 seconds, from start of 10sec excerpt to make it easier to
# associate the 2 sec excerpt in to its place in the 10 sec excerpt
# above
data_subset = data[15000:17000]
px2deg, ext = (0.0266711972026, 'lab') if '32' in f \
else (0.0185581232561, 'mri')
# take raw data and convert it to velocity: euclidean distance between
# successive coordinate samples. Note: no entry for first datapoint!
# Will plot all but first data point in other time series
velocities = cal_velocities(data_subset, sr, px2deg)
vel_color = 'xkcd:gunmetal'
# prepare plotting - much manual setup, quite ugly - sorry
fig, ax1 = plt.subplots()
fig.set_figheight(2)
fig.set_figwidth(7)
fig.set_dpi(120)
time_idx = np.linspace(0, len(data_subset) / sr, len(data_subset))[1:]
max_x = float(len(data_subset) / sr)
ax1.set_xlim(0, max_x)
ax1.set_xlabel('time (seconds)')
ax1.set_ylabel('coordinates')
# left y axis set to max screensize in px
ax1.set_ylim(0, 1280)
# plot gaze trajectories (not preprocessed)
ax1.plot(time_idx,
data_subset['x'][1:],
color='black', lw=1)
ax1.plot(
time_idx,
data_subset['y'][1:],
color='black', lw=1)
# right y axis shows velocity "as is" (not preprocessed)
ax2 = ax1.twinx()
ax2.set_ylabel('velocity (deg/sec)', color=vel_color)
ax2.tick_params(axis='y', labelcolor=vel_color)
#ax2.set_yscale('log') ## TODO: Log scale or not?
ax2.set_ylim(1, 2000)
ax2.plot(time_idx,
velocities,
color=vel_color, lw=1)
pl.savefig(
op.join('img', 'rawtrace_{}.svg'.format(ext)),
transparent=True,
bbox_inches="tight")
pl.close()
def savegaze():
"""
small function to generate and save remodnav classification figures
@ -461,6 +725,10 @@ def savegaze():
# window is within the originally plotted 50s and contains missing data
# for both data types (lab & mri)
events = clf(p[15000:25000])
# we remove plotting of details in favor of plotting raw gaze and
# velocity traces with plot_raw_vel_trace() as requested by reviewer 2
# in the second round of revision
#events_detail = clf(p[24500:24750])
fig = pl.figure(
# fake size to get the font size down in relation
@ -479,6 +747,24 @@ def savegaze():
transparent=True,
bbox_inches="tight")
pl.close()
# plot details
fig = pl.figure(
# fake size to get the font size down in relation
figsize=(7, 2),
dpi=120,
frameon=False)
#ut.show_gaze(
# pp=p[24500:24750],
# events=events_detail,
# sampling_rate=1000.0,
# show_vels=True,
# coord_lim=(0, 1280),
# vel_lim=(0, 1000))
#pl.savefig(
# op.join('img', 'remodnav_{}_detail.svg'.format(ext)),
# transparent=True,
# bbox_inches="tight")
#pl.close()
def mainseq(s_mri,
@ -776,6 +1062,82 @@ def plot_dist(figures):
pl.close()
def kappa():
"""
During the review process, reviewer 2 requested Cohens Kappa computation.
We have not implemented the measure before because we felt it did not add
information beyond the confusion and RMSD computations.
"""
px2deg = None
sr = None
from sklearn.metrics import cohen_kappa_score
# for every stimulus type
for stim in ['img', 'dots', 'video']:
# for every eye movement label used in Anderson et al. (2017)
for (ev, i) in [('Fix', 1), ('Sac', 2), ('PSO', 3)]:
# initialize lists to store classification results in
RA_res = []
MN_res = []
AL_res = []
# aggregate the target_labels of all files per coder + stim_type
for idx, fname in enumerate(labeled_files[stim]):
for coder in ['MN', 'RA', 'AL']:
if coder in ['MN', 'RA']:
data, target_labels, target_events, px2deg, sr = \
load_anderson(stim, fname.format(coder))
# dichotomize classification based on event type
labels = [1 if j == i else 0 for j in target_labels]
if coder == 'MN':
MN_res.append(labels)
elif coder == 'RA':
RA_res.append(labels)
else:
# get REMoDNaV classification
clf = EyegazeClassifier(
px2deg=px2deg,
sampling_rate=sr,
)
p = clf.preproc(data)
events = clf(p)
# convert event list into anderson-style label array
l = np.zeros(target_labels.shape, target_labels.dtype)
for e in events:
l[int(e['start_time'] * sr):int((e['end_time']) * sr)] = \
anderson_remap[label_map[e['label']]]
# dichotomize REMoDNaV classification results as well
labels = [1 if j == i else 0 for j in l]
AL_res.append(labels)
if len(MN_res[idx]) != len(RA_res[idx]):
print(
"% #\n% # %INCONSISTENCY Found label length mismatch "
"between coders for: {}\n% #\n".format(fname))
shorter = min([len(RA_res[idx]), len(MN_res[idx])])
print('% Truncate labels to shorter sample: {}'.format(
shorter))
# truncate the labels by indexing up to the highest index
# in the shorter list of labels
MN_res[idx] = MN_res[idx][:shorter]
RA_res[idx] = RA_res[idx][:shorter]
AL_res[idx] = AL_res[idx][:shorter]
# dummy check whether we really have the same number of files per coder
assert len(RA_res) == len(MN_res)
# flatten the list of lists
RA_res_flat = [item for sublist in RA_res for item in sublist]
MN_res_flat = [item for sublist in MN_res for item in sublist]
AL_res_flat = [item for sublist in AL_res for item in sublist]
#print(sum(RA_res_flat), sum(MN_res_flat))
assert len(RA_res_flat) == len(MN_res_flat) == len(AL_res_flat)
# compute Cohens Kappa
for rating, comb in [('RAMN', [RA_res_flat, MN_res_flat]),
('ALRA', [RA_res_flat, AL_res_flat]),
('ALMN', [MN_res_flat, AL_res_flat])]:
kappa = cohen_kappa_score(comb[0], comb[1])
label = 'kappa{}{}{}'.format(rating, stim, ev)
print('\\newcommand{\\%s}{%s}' % (label, '%.2f' % kappa))
if __name__ == '__main__':
import argparse
@ -811,6 +1173,8 @@ if __name__ == '__main__':
savefigs(args.figure, args.stats)
print_RMSD()
plot_dist(args.figure)
kappa()
plot_raw_vel_trace()
if args.mainseq:
mainseq(args.submri, args.sublab)
if args.remodnav:

16
img/.gitignore vendored
View file

@ -34,3 +34,19 @@ mainseq_mri.pdf
mainseq_sub_lab.pdf
mainseq_lab.pdf
mainseq_sub_mri.pdf
remodnav_lab_detail.*
remodnav_mri_detail.*
vel_est_1.pdf
preproc.pdf
pseudocode.pdf
vel_est_2.pdf
remodnav_lab_detail.pdf
remodnav_mri_detail.pdf
rawtrace_lab.pdf
flowchart.pdf
rawtrace_mri.pdf
chunking.pdf
preproc_new.pdf
flowchart_2.pdf
velhist.pdf
butterworth_filter.pdf

521
img/butterworth_filter.svg Normal file
View file

@ -0,0 +1,521 @@
<?xml version="1.0" encoding="utf-8" standalone="no"?>
<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN"
"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
<!-- Created with matplotlib (https://matplotlib.org/) -->
<svg height="123.6pt" version="1.1" viewBox="0 0 504 123.6" width="504pt" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
<defs>
<style type="text/css">
*{stroke-linecap:butt;stroke-linejoin:round;}
</style>
</defs>
<g id="figure_1">
<g id="patch_1">
<path d="M 0 123.6
L 504 123.6
L 504 0
L 0 0
z
" style="fill:#ffffff;"/>
</g>
<g id="axes_1">
<g id="patch_2">
<path d="M 63 110.004
L 453.6 110.004
L 453.6 14.832
L 63 14.832
z
" style="fill:#ffffff;"/>
</g>
<g id="matplotlib.axis_1">
<g id="xtick_1">
<g id="line2d_1">
<defs>
<path d="M 0 0
L 0 3.5
" id="mcccbe7fc8e" style="stroke:#000000;stroke-width:0.8;"/>
</defs>
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="80.754545" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_1">
<!-- 0 -->
<defs>
<path d="M 31.78125 66.40625
Q 24.171875 66.40625 20.328125 58.90625
Q 16.5 51.421875 16.5 36.375
Q 16.5 21.390625 20.328125 13.890625
Q 24.171875 6.390625 31.78125 6.390625
Q 39.453125 6.390625 43.28125 13.890625
Q 47.125 21.390625 47.125 36.375
Q 47.125 51.421875 43.28125 58.90625
Q 39.453125 66.40625 31.78125 66.40625
z
M 31.78125 74.21875
Q 44.046875 74.21875 50.515625 64.515625
Q 56.984375 54.828125 56.984375 36.375
Q 56.984375 17.96875 50.515625 8.265625
Q 44.046875 -1.421875 31.78125 -1.421875
Q 19.53125 -1.421875 13.0625 8.265625
Q 6.59375 17.96875 6.59375 36.375
Q 6.59375 54.828125 13.0625 64.515625
Q 19.53125 74.21875 31.78125 74.21875
z
" id="DejaVuSans-48"/>
</defs>
<g transform="translate(77.573295 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_2">
<g id="line2d_2">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="125.363956" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_2">
<!-- 50 -->
<defs>
<path d="M 10.796875 72.90625
L 49.515625 72.90625
L 49.515625 64.59375
L 19.828125 64.59375
L 19.828125 46.734375
Q 21.96875 47.46875 24.109375 47.828125
Q 26.265625 48.1875 28.421875 48.1875
Q 40.625 48.1875 47.75 41.5
Q 54.890625 34.8125 54.890625 23.390625
Q 54.890625 11.625 47.5625 5.09375
Q 40.234375 -1.421875 26.90625 -1.421875
Q 22.3125 -1.421875 17.546875 -0.640625
Q 12.796875 0.140625 7.71875 1.703125
L 7.71875 11.625
Q 12.109375 9.234375 16.796875 8.0625
Q 21.484375 6.890625 26.703125 6.890625
Q 35.15625 6.890625 40.078125 11.328125
Q 45.015625 15.765625 45.015625 23.390625
Q 45.015625 31 40.078125 35.4375
Q 35.15625 39.890625 26.703125 39.890625
Q 22.75 39.890625 18.8125 39.015625
Q 14.890625 38.140625 10.796875 36.28125
z
" id="DejaVuSans-53"/>
</defs>
<g transform="translate(119.001456 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-53"/>
<use x="63.623047" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_3">
<g id="line2d_3">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="169.973367" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_3">
<!-- 100 -->
<defs>
<path d="M 12.40625 8.296875
L 28.515625 8.296875
L 28.515625 63.921875
L 10.984375 60.40625
L 10.984375 69.390625
L 28.421875 72.90625
L 38.28125 72.90625
L 38.28125 8.296875
L 54.390625 8.296875
L 54.390625 0
L 12.40625 0
z
" id="DejaVuSans-49"/>
</defs>
<g transform="translate(160.429617 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-49"/>
<use x="63.623047" xlink:href="#DejaVuSans-48"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_4">
<g id="line2d_4">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="214.582778" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_4">
<!-- 150 -->
<g transform="translate(205.039028 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-49"/>
<use x="63.623047" xlink:href="#DejaVuSans-53"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_5">
<g id="line2d_5">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="259.192188" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_5">
<!-- 200 -->
<defs>
<path d="M 19.1875 8.296875
L 53.609375 8.296875
L 53.609375 0
L 7.328125 0
L 7.328125 8.296875
Q 12.9375 14.109375 22.625 23.890625
Q 32.328125 33.6875 34.8125 36.53125
Q 39.546875 41.84375 41.421875 45.53125
Q 43.3125 49.21875 43.3125 52.78125
Q 43.3125 58.59375 39.234375 62.25
Q 35.15625 65.921875 28.609375 65.921875
Q 23.96875 65.921875 18.8125 64.3125
Q 13.671875 62.703125 7.8125 59.421875
L 7.8125 69.390625
Q 13.765625 71.78125 18.9375 73
Q 24.125 74.21875 28.421875 74.21875
Q 39.75 74.21875 46.484375 68.546875
Q 53.21875 62.890625 53.21875 53.421875
Q 53.21875 48.921875 51.53125 44.890625
Q 49.859375 40.875 45.40625 35.40625
Q 44.1875 33.984375 37.640625 27.21875
Q 31.109375 20.453125 19.1875 8.296875
z
" id="DejaVuSans-50"/>
</defs>
<g transform="translate(249.648438 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-50"/>
<use x="63.623047" xlink:href="#DejaVuSans-48"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_6">
<g id="line2d_6">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="303.801599" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_6">
<!-- 250 -->
<g transform="translate(294.257849 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-50"/>
<use x="63.623047" xlink:href="#DejaVuSans-53"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_7">
<g id="line2d_7">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="348.41101" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_7">
<!-- 300 -->
<defs>
<path d="M 40.578125 39.3125
Q 47.65625 37.796875 51.625 33
Q 55.609375 28.21875 55.609375 21.1875
Q 55.609375 10.40625 48.1875 4.484375
Q 40.765625 -1.421875 27.09375 -1.421875
Q 22.515625 -1.421875 17.65625 -0.515625
Q 12.796875 0.390625 7.625 2.203125
L 7.625 11.71875
Q 11.71875 9.328125 16.59375 8.109375
Q 21.484375 6.890625 26.8125 6.890625
Q 36.078125 6.890625 40.9375 10.546875
Q 45.796875 14.203125 45.796875 21.1875
Q 45.796875 27.640625 41.28125 31.265625
Q 36.765625 34.90625 28.71875 34.90625
L 20.21875 34.90625
L 20.21875 43.015625
L 29.109375 43.015625
Q 36.375 43.015625 40.234375 45.921875
Q 44.09375 48.828125 44.09375 54.296875
Q 44.09375 59.90625 40.109375 62.90625
Q 36.140625 65.921875 28.71875 65.921875
Q 24.65625 65.921875 20.015625 65.03125
Q 15.375 64.15625 9.8125 62.3125
L 9.8125 71.09375
Q 15.4375 72.65625 20.34375 73.4375
Q 25.25 74.21875 29.59375 74.21875
Q 40.828125 74.21875 47.359375 69.109375
Q 53.90625 64.015625 53.90625 55.328125
Q 53.90625 49.265625 50.4375 45.09375
Q 46.96875 40.921875 40.578125 39.3125
z
" id="DejaVuSans-51"/>
</defs>
<g transform="translate(338.86726 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-51"/>
<use x="63.623047" xlink:href="#DejaVuSans-48"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_8">
<g id="line2d_8">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="393.02042" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_8">
<!-- 350 -->
<g transform="translate(383.47667 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-51"/>
<use x="63.623047" xlink:href="#DejaVuSans-53"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="xtick_9">
<g id="line2d_9">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="437.629831" xlink:href="#mcccbe7fc8e" y="110.004"/>
</g>
</g>
<g id="text_9">
<!-- 400 -->
<defs>
<path d="M 37.796875 64.3125
L 12.890625 25.390625
L 37.796875 25.390625
z
M 35.203125 72.90625
L 47.609375 72.90625
L 47.609375 25.390625
L 58.015625 25.390625
L 58.015625 17.1875
L 47.609375 17.1875
L 47.609375 0
L 37.796875 0
L 37.796875 17.1875
L 4.890625 17.1875
L 4.890625 26.703125
z
" id="DejaVuSans-52"/>
</defs>
<g transform="translate(428.086081 124.602437)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-52"/>
<use x="63.623047" xlink:href="#DejaVuSans-48"/>
<use x="127.246094" xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
</g>
<g id="matplotlib.axis_2">
<g id="ytick_1">
<g id="line2d_10">
<defs>
<path d="M 0 0
L -3.5 0
" id="m3b237ec3d4" style="stroke:#000000;stroke-width:0.8;"/>
</defs>
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="63" xlink:href="#m3b237ec3d4" y="109.898149"/>
</g>
</g>
<g id="text_10">
<!-- 0 -->
<g transform="translate(49.6375 113.697368)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-48"/>
</g>
</g>
</g>
<g id="ytick_2">
<g id="line2d_11">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="63" xlink:href="#m3b237ec3d4" y="81.591675"/>
</g>
</g>
<g id="text_11">
<!-- 1 -->
<g transform="translate(49.6375 85.390894)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-49"/>
</g>
</g>
</g>
<g id="ytick_3">
<g id="line2d_12">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="63" xlink:href="#m3b237ec3d4" y="53.2852"/>
</g>
</g>
<g id="text_12">
<!-- 2 -->
<g transform="translate(49.6375 57.084419)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-50"/>
</g>
</g>
</g>
<g id="ytick_4">
<g id="line2d_13">
<g>
<use style="stroke:#000000;stroke-width:0.8;" x="63" xlink:href="#m3b237ec3d4" y="24.978726"/>
</g>
</g>
<g id="text_13">
<!-- 3 -->
<g transform="translate(49.6375 28.777945)scale(0.1 -0.1)">
<use xlink:href="#DejaVuSans-51"/>
</g>
</g>
</g>
</g>
<g id="line2d_14">
<path clip-path="url(#pf16420582c)" d="M 80.754545 59.760507
L 83.43111 64.808651
L 86.999863 71.004378
L 92.352992 79.744645
L 95.921745 85.273764
L 98.59831 89.11392
L 101.274874 92.548368
L 103.059251 94.530042
L 104.843627 96.230436
L 106.628004 97.652072
L 108.41238 98.824346
L 110.196757 99.775625
L 111.981133 100.520542
L 113.765509 101.081117
L 116.442074 101.62964
L 119.118639 101.853608
L 120.903015 101.820115
L 123.57958 101.5001
L 126.256144 100.928888
L 130.717085 99.67567
L 136.070215 97.891419
L 140.531156 96.490165
L 143.20772 95.823531
L 145.884285 95.363501
L 148.56085 95.178131
L 151.237414 95.295561
L 153.913979 95.683503
L 156.590544 96.279016
L 160.159296 97.307354
L 165.512426 99.161912
L 169.973367 100.631658
L 172.649931 101.266779
L 174.434308 101.477847
L 176.218684 101.453605
L 178.003061 101.157025
L 179.787437 100.580923
L 181.571814 99.746248
L 183.35619 98.692279
L 186.032755 96.799174
L 189.601508 93.897482
L 202.092143 83.248155
L 204.768707 81.293446
L 207.445272 79.573311
L 210.121836 78.134281
L 212.798401 77.018636
L 214.582778 76.469316
L 216.367154 76.079439
L 218.15153 75.85354
L 219.935907 75.799211
L 221.720283 75.923055
L 223.50466 76.229004
L 225.289036 76.714665
L 227.965601 77.748058
L 230.642165 79.093172
L 233.31873 80.687661
L 236.887483 83.089468
L 242.240612 87.041097
L 251.162494 93.724893
L 255.623435 96.727681
L 259.192188 98.82325
L 261.868753 100.161484
L 264.545317 101.267206
L 267.221882 102.104889
L 269.898447 102.643027
L 272.575011 102.86213
L 275.251576 102.781734
L 277.928141 102.457448
L 281.496894 101.775299
L 294.879717 98.885106
L 298.44847 98.458636
L 302.017222 98.277236
L 305.585975 98.347621
L 309.154728 98.656353
L 312.723481 99.18907
L 316.292234 99.927317
L 320.753175 101.117436
L 325.214116 102.54103
L 332.351622 104.866242
L 335.028186 105.4642
L 336.812563 105.65882
L 338.596939 105.640619
L 340.381316 105.402319
L 342.165692 104.970282
L 344.842257 104.057403
L 348.41101 102.565693
L 358.22508 98.3273
L 361.793833 97.056304
L 365.362586 96.041424
L 368.03915 95.480162
L 370.715715 95.109131
L 373.39228 94.942271
L 376.068844 94.987067
L 378.745409 95.246249
L 381.421974 95.714891
L 384.098538 96.380336
L 387.667291 97.535329
L 392.128232 99.291817
L 400.157926 102.554894
L 402.834491 103.403064
L 405.511055 104.004079
L 407.295432 104.223027
L 409.079808 104.248848
L 410.864185 104.01496
L 412.648561 103.429918
L 413.540749 102.973143
L 414.432937 102.389638
L 415.325126 101.667368
L 416.217314 100.795133
L 417.109502 99.762435
L 418.893878 97.17498
L 420.678255 93.822681
L 422.462631 89.616975
L 424.247008 84.459601
L 426.031384 78.241502
L 427.815761 70.842628
L 429.600137 62.131921
L 431.384513 51.967217
L 433.16889 40.19501
L 434.953266 26.65012
L 435.845455 19.158
L 435.845455 19.158
" style="fill:none;stroke:#000000;stroke-linecap:square;stroke-width:0.5;"/>
</g>
<g id="patch_3">
<path d="M 63 110.004
L 63 14.832
" style="fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;"/>
</g>
<g id="patch_4">
<path d="M 453.6 110.004
L 453.6 14.832
" style="fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;"/>
</g>
<g id="patch_5">
<path d="M 63 110.004
L 453.6 110.004
" style="fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;"/>
</g>
<g id="patch_6">
<path d="M 63 14.832
L 453.6 14.832
" style="fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;"/>
</g>
</g>
</g>
<defs>
<clipPath id="pf16420582c">
<rect height="95.172" width="390.6" x="63" y="14.832"/>
</clipPath>
</defs>
</svg>
Side by side

After

Width:  |  Height:  |  Size: 14 KiB

1504
img/chunking.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 38 KiB

1786
img/flowchart.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 92 KiB

5011
img/flowchart_2.svg Normal file
View file

File diff suppressed because one or more lines are too long
Side by side

After

Width:  |  Height:  |  Size: 453 KiB

1217
img/preproc.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 74 KiB

4365
img/preproc_new.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 108 KiB

1819
img/pseudocode.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 172 KiB

2410
img/vel_est_1.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 181 KiB

2453
img/vel_est_2.svg Normal file
View file

File diff suppressed because one or more lines are too long
Side by side

After

Width:  |  Height:  |  Size: 184 KiB

1814
img/velhist.svg Normal file
View file

File diff suppressed because it is too large Load diff
Side by side

After

Width:  |  Height:  |  Size: 50 KiB

589
main.tex
View file

@ -74,7 +74,7 @@
\input{results_def.tex}
\onecolumn
\title{REMoDNaV: Robust Eye Movement Detection for Natural Viewing } %\\ (remodnav)
\title{REMoDNaV: Robust Eye-Movement Classification for Dynamic Stimulation} %\\ (remodnav)
% \titlenote{The title should be detailed enough for someone to know whether
% the article would be of interest to them, but also concise. Please ensure the
@ -127,15 +127,15 @@ Heinrich Heine University Düsseldorf, Germany
Tracking of eye movements is an established measurement for many types of
experimental paradigms.
More complex and lengthier visual stimuli have made algorithmic approaches to
eye movement event detection the most pragmatic option.
More complex and more prolonged visual stimuli have made algorithmic approaches to
eye movement event classification the most pragmatic option.
A recent analysis revealed that many current algorithms are lackluster when it
comes to data from viewing dynamic stimuli such as video sequences.
Here we present an event detection algorithm---built on an existing
Here we present an event classification algorithm---built on an existing
velocity-based approach---that is suitable for both static and dynamic
stimulation, and is capable of detecting saccades, post-saccadic
stimulation, and is capable of classifying saccades, post-saccadic
oscillations, fixations, and smooth pursuit events.
We validated detection performance and robustness on three public datasets:
We validated classification performance and robustness on three public datasets:
1)~manually annotated, trial-based gaze trajectories for viewing static images,
moving dots, and short video sequences, 2)~lab-quality gaze recordings for a
feature length movie, and 3)~gaze recordings acquired under suboptimal lighting
@ -144,20 +144,21 @@ the same full-length movie.
We found that the proposed algorithm performs on par or better compared
to state-of-the-art alternatives for static stimulation. Moreover, it yields
eye movement events with biologically plausible characteristics on prolonged
recordings without a trial structure. Lastly, algorithm performance is robust
dynamic recordings. Lastly, algorithm performance is robust
on data acquired under suboptimal conditions that exhibit a temporally
varying noise level.
These results indicate that the proposed algorithm is a robust tool with
improved detection accuracy across a range of use cases.
A cross-platform compatible implementation in the Python programming language
is available as free and open source software.
improved classification accuracy across a range of use cases.
The algorithm is cross-platform compatible, implemented using the
Python programming language, and readily available as free and open source software
from public sources.
\keywords{%
eye tracking \and
adaptive detection algorithm \and
saccade detection algorithm \and
adaptive classification algorithm \and
saccade classification algorithm \and
statistical saccade analysis \and
glissade detection \and
glissade classification \and
adaptive threshold algorithm \and
data preprocessing
}
@ -198,32 +199,33 @@ data preprocessing
% applied an adaptive algorithm to the eye movement data to provide a more
% precise computation of saccades and fixations.
A spreading theme in cognitive neuroscience is to use dynamic and natural
stimuli as opposed to isolated and distinct imagery \citep{real_world}. Using
dynamic stimuli promises to observe the nuances of cognition in a more natural
environment. Some interesting applications include the determination of neural
A spreading theme in cognitive neuroscience is to use dynamic and naturalistic
stimuli such as video clips or movies as opposed to static and isolated
stimuli \citep{real_world}. Using dynamic stimuli promises to observe
the nuances of cognition in a more life-like environment \citep{maguire2012studying}.
Some interesting applications include the determination of neural
response to changes in facial expression \citep{Harris2014}, understanding
complex social interactions by using videos \citep{Tikka2012} and more
complex social interactions by using videos \citep{Tikka2012}, and more
untouched themes such as the underlying processing of music
\citep{Toiviainen2014}. In such studies, an unobtrusive behavioral measurement
is required to quantify the relationship between stimulus and response.
Tracking the focus of participants' gaze is a suitable, well established
measure that has been successfully employed in a variety of studies ranging
method that has been successfully employed in a variety of studies ranging
from the understanding of visual attention \citep{HantaoLiu2011}, memory
\citep{Hannula2010} and language comprehension \citep{Gordon2006}.
\citep{Hannula2010}, and language comprehension \citep{Gordon2006}.
%
Regardless of use case, the raw eye tracking data (position coordinates)
Regardless of use case, the raw eye tracking data (gaze position coordinates)
provided by eye tracking devices are rarely used ``as is". Instead, in order
to disentangle different cognitive, occulomotor, or perceptive states
to disentangle different cognitive, oculomotor, or perceptive states
associated with different types of eye movements, most research relies on the
classification of eye gaze data into distinct eye movement event categories
\citep{Schutz2011}. The most feasible approach for doing this lies in the
application of appropriate event detection algorithms.
application of appropriate event classification algorithms.
However, a recent comparison of algorithms found that while many readily
available algorithms for eye movement classification performed well on data
from static stimulation or short trial-based acquisitions with simplified
moving stimuli, none worked particularly well on data from complex natural
moving stimuli, none worked particularly well on data from complex
dynamic stimulation, such as video clips, when compared to human coders
\citep{Andersson2017}.
%
@ -232,45 +234,76 @@ the winners in the aforementioned comparison, on data from prolonged
stimulation (\unit[$\approx$15]{min}) with a feature film, we found the
average and median durations of labeled fixations to exceed literature
reports \citep[\eg][]{holmqvist2011eye,dorr2010variability} by up to a factor
of two. Additionally, and in particular for increasing levels of noise in the
of two. Additionally, and in particular for increasing levels of noise in the
data, the algorithm classified too few fixations, as also noted by
\citet{Friedman2018}, because it discarded potential fixation events that
contained artifacts such as blinks.
contained data artifacts such as signal-loss and distortion associated with
blinks.
%
However, robust performance on noisy data is of particular relevance in the
context of ``natural stimulation'', as the ultimate natural stimulation is the
actual natural environment, and data acquired outdoors or with mobile
devices typically does not match the quality achieved in dedicated lab
setups.
%However, robust performance on noisy data is of particular relevance in the
%context of ``natural stimulation'', as the ultimate natural stimulation is the
%actual natural environment, and data acquired outdoors or with mobile
%devices typically does not match the quality achieved in dedicated lab
%setups.
Therefore our objective was to improve upon the available eye movement
detection and classification algorithms, and develop a tool that performs
robustly on data from dynamic natural stimulation, without sacrificing detection
classification algorithms, and develop a tool that performs
robustly on data from dynamic, feature-rich stimulation, without sacrificing classification
accuracy for static and simplified stimulation. Importantly, we aimed for
applicability to prolonged recordings that lack any kind of trial structure,
and exhibit periods of signal-loss and non-stationary noise levels.
applicability to prolonged recordings that potentially exhibit periods of
signal-loss and non-stationary noise levels.
Finally, one of our main objectives was to keep the algorithm as accessible
and easily available as possible in order to ease the difficulties associated
with closed-source software or non-publicly available source code of published
algorithms.
% maybe work in \citep{Hooge2018} again
In addition to the event categories \textit{fixation}, \textit{saccade}, and
\textit{post-saccadic oscillation} (PSO; sometimes termed ``glissade"), the
algorithm had to support the detection of \textit{smooth pursuit} events, as
emphasized by \cite{Andersson2017}. These are slow movements of the eye during
tracking of a moving target and are routinely evoked by moving visual objects
during dynamic stimulation \citep{carl1987pursuits}. If this type of eye
movement is not properly detected and labeled, erroneous fixation and saccade
events (which smooth pursuits would be classified into instead) are introduced.
Contemporary algorithms rarely provide this functionality
\cite[but see \eg][for existing algorithms with
smooth pursuit detection]{LARSSON2015145,Komogortsev2013}.
Here we introduce \remodnav\ (robust eye movement detection for natural
viewing), a novel tool that aims to meet these objectives. It is built on the
Following the best practices proposed by \citet{hessels2018eye},
we define the different eye-movements that are supported by our algorithm
on a functional and oculomotor dimension as follows:
A \textit{fixation} is a period of time during which a part of the visual stimulus
is looked at and thereby projected to a relatively constant location on the retina.
This type of eye movement is necessary for visual intake, and characterized by a
relatively still gaze position with respect to the world (e.g., a computer screen
used for stimulus presentation) in the eye-tracker signal.
A fixation event therefore excludes periods of \textit{smooth pursuit}.
These events are eye movements during which a
part of the visual stimulus that moves with respect to the world is looked at for
visual intake (e.g., a moving dot on a computer screen). Like fixations,
the stimulus is projected to a relatively constant location on the retina
\citep{carl1987pursuits}, however, the event is characterized by steadily changing
gaze position in the eye-tracker signal.
If this type of eye movement is not properly classified,
erroneous fixation and saccade events (which smooth pursuits would be classified into
instead) are introduced \citep{Andersson2017}. Contemporary algorithms rarely provide
this functionality \cite[but see \eg][for existing algorithms with
smooth pursuit classification]{LARSSON2015145,Komogortsev2013}.
\textit{Saccades} on the other hand are also characterized by changing gaze positions,
but their velocity trace is usually higher than that of pursuit movements.
They serve to shift the position of the eye to a target region, and, unlike
during pursuit or fixation events, visual intake
is suppressed \citep{Schutz2011}. Lastly, \textit{post-saccadic oscillations} are periods of
ocular instability after a saccade \citep{Nystrom2010AnData}.
Here we introduce \remodnav\ (robust eye movement classification for dynamic
stimulation), a new tool that aims to meet our objectives and classifies the
eye movement events defined above. It is built on the
aforementioned algorithm by \citet{Nystrom2010AnData} (subsequently labeled NH)
that employs an adaptive approach to velocity based eye movement event
detection and classification. \remodnav\ enhances NH with the use of robust
classification. \remodnav\ enhances NH with the use of robust
statistics, and a compartmentalization of prolonged time series into short,
more homogeneous segments with more uniform noise levels. Furthermore, it adds
support for pursuit event detection. We evaluated \remodnav\ on three different
datasets from conventional paradigms, and natural stimulation (high and lower
more homogeneous segments with more uniform noise levels.
Furthermore, it adds support for pursuit event classification.
Just as the original algorithm, its frame of reference is world centered,
i.e. the gaze coordinates have a reference to a stimulation set-up with a fixed
position in the world such as x and y coordinates in pixel of a computer screen,
and it is meant to be used with eye tracking data from participants viewing
static (e.g. images) or dynamic (e.g. videos) stimuli, recorded with remote or
tower-mounted eye trackers.
Importantly, it is built and distributed as free, open source software,
and can be easily obtained and executed with free tools.
We evaluated \remodnav\ on three different
datasets from conventional paradigms, and dynamic, feature-rich stimulation (high and lower
quality), and relate its performance to the algorithm comparison by
\cite{Andersson2017}.
@ -285,57 +318,95 @@ quality), and relate its performance to the algorithm comparison by
%technical details required for implementation of the tool by other
%developers.}}
Like NH, \remodnav\ is a \textit{velocity-based} event detection algorithm.
Compared to \textit{dispersion-based} algorithms, these types of algorithms are
less susceptible to noise and spatio-temporal precision, and are thus applicable to
a wide range of sampling frequencies. Furthermore, any influence of
biologically implausible velocities and accelerations can be prevented with the
use of appropriate filters and thresholds \citep{holmqvist2011eye}.
The algorithm comprises two major steps: preprocessing and event detection. A
general overview and pseudo-code are shown in \fig{alg}. The following
Event classification algorithms can be broadly grouped into \textit{velocity-} and
\textit{dispersion-}based algorithms. The former rely on velocity thresholds to
differentiate between different eye movement events, while the latter classify
eye movements based on the size of the region the recorded data falls
into for a given amount of time \citep{holmqvist2011eye}. Both types of algorithms
are common (see e.g., \citet{hessels2017noise} for a recent dispersion-based,
and e.g., \citet{van2018gazepath} for a recent velocity-based solution,
and see \citet{dalveren2019evaluation} for an evaluation of common algorithms
of both types).
Like NH, \remodnav\ is a \textit{velocity-based} event classification algorithm.
The algorithm comprises two major steps: preprocessing and event classification. The following
sections detail individual analysis steps. For each step relevant algorithm
parameters are given in parenthesis. \tab{parameters} summarizes all
parameters, and lists their default values.
\begin{figure*}
\includegraphics[width=1\textwidth]{img/remodnav_algorithm.pdf}
\caption{\remodnav workflow. Optional steps and configurable parameters are in bold.}
\label{fig:alg}
\end{figure*}
parameters are given in parenthesis.
\fig{alg} provides an overview of the algorithm's main components.
\tab{parameters} summarizes all parameters, and lists their default values.
The computational definitions of the different eye movements
\citep{hessels2018eye} are given within the event classification description.
Note, however, that some of the computational definitions of eye movements can be
adjusted to comply to alternative definitions by changing the algorithms'
parameters.
\subsection*{Preprocessing}
The goal of data preprocessing is to compute a time series of eye movement
velocities on which the event detection algorithm can be executed, while jointly
reducing non-movement-related noise in the data as much as possible.
velocities on which the event classification algorithm can be executed, while jointly
reducing non-eyemovement-related noise in the data as much as possible.
First, implausible spikes in the coordinate time series are removed with a
heuristic spike filter \citep{stampe1993} (\fig{alg}A, 1). This filter is
heuristic spike filter \citep{stampe1993} (\fig{alg}, P1). This filter is
standard in many eye tracking toolboxes and often used for preprocessing
\citep[\eg][]{Nystrom2010AnData}.
\citep[\eg][]{Friedman2018}.
%
Data samples around signal loss (\eg eye blinks) can be nulled in order to
remove spurious movement signals (\param{dilate\_nan},
\param{min\_blink\_duration}; \fig{alg}A, 2).
Data samples around signal loss (\eg eye blinks) can be set to non-numeric values (NaN)
in order to eliminate spurious movement signals without shortening the time series
(\param{dilate\_nan}, \param{min\_blink\_duration}; \fig{alg}, P2). This is
motivated by the fact that blinks can produce artifacts in the eye-tracking signal when the
eyelid closes and re-opens \citep{choe2016pupil}.
%
Coordinate time series are temporally filtered in two different ways
(\fig{alg}A, 3). A relatively large median filter (\param{median\_
filter\_length}) is used to emphasize long-distance saccades. This type of
\fig{alg}, P3). A relatively large median filter
(\param{median\_filter\_length}) is used to emphasize large amplitude saccades. This type of
filtered data is later used for a coarse segmentation of a time series into
shorter intervals between major saccades.
%
Separately, data are also smoothed with a Savitzky-Golay filter
(\param{savgol\_ \{length,polyord\}}). All event detection beyond the
(\param{savgol\_ \{length,polyord\}}). All event classification beyond the
localization of major saccades for time series chunking is performed on this
type of filtered data.
After spike-removal and temporal filtering, movement velocities are computed
(\fig{alg}A, 4-5). To disregard biologically implausible measurements, a
After spike-removal and temporal filtering, movement velocities are computed.
To disregard biologically implausible measurements, a
configurable maximum velocity (\param{max\_vel}) is enforced---any samples
exceeding this threshold are replaced by this set value.
%The result of a default preprocessing procedure is displayed in \fig{preproc}.
%
%\begin{figure}
% \includegraphics[width=0.5\textwidth]{img/preproc.pdf}
% \caption{Examplary preprocessing.}
% \label{fig:preproc}
%\end{figure}
\begin{figure*}
\includegraphics[width=1\textwidth]{img/flowchart_2.pdf}
\caption{Schematic algorithm overview of \remodnav.
Panel A: Preprocessing. The two plots show raw (blue) and preprocessed (black)
time series after preprocessing with the default parameter values
(see Table \ref{tab:parameters} for details).
Panel B: Adaptive saccade velocity computation and time series chunking.
Starting from an initial velocity threshold (\param{velthresh\_startvelocity}),
a global velocity threshold is iteratively computed. The time series is chunked
into intervals between the fastest saccades across the complete recording.
Panel C: Saccade and PSO classification.
Saccade on- and offsets, and PSO on- and offsets are classified based on adaptive
velocity thresholds computed within the respective event contexts.
The default context is \unit[1]{s} centered on the peak velocity for saccadic
events used for time series chunking, and the entire time series chunk for
intersaccadic intervals. PSOs are classified into low- or high-velocity PSOs
depending on whether they exceed the saccade onset- or peak-velocity threshold.
Panel D: Fixation and pursuit classification.
Remaining unlabeled segments are filtered with a Butterworth filter. Samples
exceeding a configurable pursuit velocity threshold (\param{pursuit\_velthresh})
are classified as pursuits, and segments that do not qualify as pursuits are
classified as fixations.
}
\label{fig:alg}
\end{figure*}
\begin{table*}[tbp]
\caption{Exhaustive list of algorithm parameters, their default values, and units.}
\label{tab:parameters}
@ -355,16 +426,16 @@ exceeding this threshold are replaced by this set value.
\unit[0.02]{s}\\
\texttt{dilate\_nan} &
duration for which to replace data by missing data markers on either side of a
signal-loss window &
signal-loss window (\fig{alg}, P2)&
\unit[0.01]{s}\\
\texttt{median\_filter\_length} &
smoothing median-filter size (for initial data chunking only) &
smoothing median-filter size (for initial data chunking only) (\fig{alg}, P3)&
\unit[0.05]{s}\\
\texttt{savgol\_length} &
size of Savitzky-Golay filter for noise reduction&
size of Savitzky-Golay filter for noise reduction (\fig{alg}, P3)&
\unit[0.019]{s}\\
\texttt{savgol\_polyord} &
polynomial order of Savitzky-Golay filter for noise reduction&
polynomial order of Savitzky-Golay filter for noise reduction (\fig{alg}, P3)&
2\\
\texttt{max\_vel} &
maximum velocity threshold, will replace value with maximum, and issue
@ -373,56 +444,56 @@ exceeding this threshold are replaced by this set value.
\citep[default value based on ][]{holmqvist2011eye}&
\unit[1000]{deg/s}\\
\\\multicolumn{3}{l}{\textit{Event detection}} \\
\\\multicolumn{3}{l}{\textit{Event classification}} \\
\texttt{min\_saccade\_duration} &
minimum duration of a saccade event candidate &
minimum duration of a saccade event candidate (\fig{alg}, E3) &
\unit[0.01]{s}\\
\texttt{max\_pso\_duration} &
maximum duration of a post-saccadic oscillation (glissade) candidate &
maximum duration of a post-saccadic oscillation (glissade) (\fig{alg}, E3) &
\unit[0.04]{s}\\
\texttt{min\_fixation\_duration} &
minimum duration of a fixation event candidate &
minimum duration of a fixation event candidate (\fig{alg}, E4)&
\unit[0.04]{s}\\
\texttt{min\_pursuit\_duration} &
minimum duration of a pursuit event candidate &
minimum duration of a pursuit event candidate (\fig{alg}, E4)&
\unit[0.04]{s}\\
\texttt{min\_intersaccade\_duration} &
no saccade detection is performed in windows shorter than twice this value, plus minimum saccade and PSO duration&
no saccade classification is performed in windows shorter than twice this value, plus minimum saccade and PSO duration (\fig{alg}, E2)&
\unit[0.04]{s}\\
\texttt{noise\_factor} &
adaptive saccade onset threshold velocity is the median absolute deviation of velocities in the window of interest, times this factor (peak velocity threshold is twice the onset velocity); increase for noisy data to reduce false positives \citep[equivalent: 3.0]{Nystrom2010AnData}&
adaptive saccade onset threshold velocity is the median absolute deviation of velocities in the window of interest, times this factor (peak velocity threshold is twice the onset velocity); increase for noisy data to reduce false positives \citep[equivalent: 3.0]{Nystrom2010AnData}(\fig{alg}E1)&
5\\
\texttt{velthresh\_startvelocity} &
start value for adaptive velocity threshold algorithm \citep{Nystrom2010AnData}, should
be larger than any conceivable minimum saccade velocity &
be larger than any conceivable minimum saccade velocity (\fig{alg}, E1)&
\unit[300]{deg/s}\\
\texttt{max\_initial\_saccade\_freq} &
maximum saccade frequency for initial detection of major saccades, initial data
maximum saccade frequency for initial classification of major saccades, initial data
chunking is stopped if this frequency is reached (should be smaller than an expected
(natural) saccade frequency in a particular context), default based on literature reports of a natural, free-viewing saccade frequency of \unit[$\sim$1.7 $\pm$0.3]{Hz} during a movie stimulus \citep{amit2017temporal} &
(natural) saccade frequency in a particular context), default based on literature reports of a natural, free-viewing saccade frequency of \unit[$\sim$1.7 $\pm$0.3]{Hz} during a movie stimulus \citep{amit2017temporal} (\fig{alg}E1)&
\unit[2]{Hz}\\
\texttt{saccade\_context\_window\_length} &
size of a window centered on any velocity peak for adaptive determination of
saccade velocity thresholds (for initial data chunking only) &
saccade velocity thresholds (for initial data chunking only) (\fig{alg}, E2)&
\unit[1]{s}\\
\texttt{lowpass\_cutoff\_freq} &
cut-off frequency of a Butterworth low-pass filter applied to determine drift
velocities in a pursuit event candidate &
velocities in a pursuit event candidate (\fig{alg}, E4)&
\unit[4]{Hz}\\
\texttt{pursuit\_velthresh} &
fixed drift velocity threshold to distinguish periods of pursuit from periods of fixation; higher than natural ocular drift velocities during fixations \citep[\eg ][]{GOLTZ1997789,cherici2012} &
fixed drift velocity threshold to distinguish periods of pursuit from periods of fixation; higher than natural ocular drift velocities during fixations \citep[\eg ][]{GOLTZ1997789,cherici2012} (\fig{alg}, E4)&
\unit[2]{deg/s}\\
\end{tabular}
\end{table*}
\subsection*{Event detection}
\subsection*{Event classification}
\subsubsection*{Saccade velocity threshold}
Except for a few modifications, \remodnav\ employs the adaptive saccade
detection algorithm proposed by \cite{Nystrom2010AnData}, where saccades are
classification algorithm proposed by \cite{Nystrom2010AnData}, where saccades are
initially located by thresholding the velocity time series by a critical value.
Starting from an initial velocity threshold (\param{velthresh\_startvelocity},
termed $PT_1$ in NH), the critical value is determined adaptively by computing
@ -448,17 +519,29 @@ PT_n = median({V}_{n-1}) + F \times MAD({V}_{n-1})
%
where $MAD$ is the median absolute deviation, and $F$ is a
scalar parameter of the algorithm.
This iterative process is illustrated in \fig{alg}, E1 (upper panel).
% Adina: removed in favor of new algorithm overview
%\begin{figure}
% \includegraphics[width=0.5\textwidth]{img/vel_est_1.pdf}
% \caption{Iterative, global estimation of velocity thresholds
% for saccades (SACC), and high/low velocity post saccadic oscillations (HPSO/LPSO).
% The method is adapted from \cite{Nystrom2010AnData}, but is modified to use robust statistics
% with median absolute deviation (MAD) as a measure of variability, more suitable
% for data with a non-normal distribution.}
% \label{fig:velest1}
%\end{figure}
\subsection*{Time series chunking}
As the algorithm aims to be applicable to prolonged recordings without an
inherent trial structure and inhomogeneous noise levels, the time series needs
As the algorithm aims to be applicable to prolonged recordings with
potentially inhomogeneous noise levels, the time series needs
to be split into shorter chunks to prevent the negative impact of sporadic
noise flares on the aforementioned adaptive velocity thresholding procedure.
\remodnav\ implements this chunking by determining a critical velocity on a
median-filtered (\param{median\_ filter\_length}) time series comprising the
full duration of a recording (\fig{alg}D). Due to potentially elevated noise
\remodnav\ implements this time-series chunking by determining a critical velocity on a
median-filtered (\param{median\_filter\_length}) time series comprising the
full duration of a recording (\fig{alg}, E2). Due to potentially elevated noise
levels, the resulting threshold tends to overestimate an optimal threshold.
Consequently, only periods of fastest eye movements will exceed this threshold.
All such periods of consecutive above-threshold velocities are weighted by the
@ -467,17 +550,17 @@ selecting such events sequentially (starting with the largest sums), until a
maximum average frequency across the whole time series is reached
(\param{max\_initial\_saccade\_ freq}). The resulting chunks represent data
intervals between saccades of maximum magnitude in the respective data.
\fig{alg}, E3 (right) exemplifies event classification within such an intersaccadic interval.
\subsection*{Classification of saccades and post-saccadic oscillations}
\subsection*{Detection of saccades and post-saccadic oscillations}
Detection of these event types is identical to the NH algorithm, only the data
Classification of these event types is identical to the NH algorithm, only the data
context and metrics for determining the velocity thresholds differ. For
saccades that also represent time series chunk boundaries (event label
\texttt{SACC}), a context of \unit[1]{s}
(\param{saccade\_context\_window\_ length}) centered on the peak velocity is
used by default, for any other saccade (event label \texttt{ISAC}) the entire
time series chunk represents that context (\fig{alg}E).
time series chunk represents that context (\fig{alg}, E3).
Peak velocity threshold and on/offset velocity threshold are then determined by
equation \ref{eq:threshold} with $F$ set to $2\times\mathtt{noise\_factor}$ and
@ -496,18 +579,28 @@ and high-velocity oscillations (event label \texttt{HPSO} or \texttt{IHPS}),
where the velocity exceeds the saccade onset or peak velocity threshold,
respectively.
\subsection*{Pursuit and fixation detection}
% Adina: Removed in favor of new algorithm overview
%\begin{figure}
% \includegraphics[width=0.5\textwidth]{img/vel_est_2.pdf}
% \caption{Iterative event classification between major saccades (SACC).
% The algorithm reports saccades within major saccade windows (ISAC),
% high/low velocity post saccadic oscillations after ISAC events (IHPS/ILPS),
% fixations (FIXA), and smooth pursuits (PURS).}
% \label{fig:velest2}
%\end{figure}
\subsection*{Pursuit and fixation classification}
For all remaining, unlabeled time series segments that are longer than a
minimum duration (\param{min\_fixation\_ duration}), velocities are low-pass
filtered (Butterworth, \param{lowpass\_cutoff\_freq}). Any segments
exceeding a minimum velocity threshold (\param{pursuit\_velthresh}) are
classified as pursuit (event label \texttt{PURS}). Pursuit on/offset detection
classified as pursuit (event label \texttt{PURS}). Pursuit on/offset classification
uses the same approach as that for saccades: search for local minima preceding
and following the above threshold velocities.
%
Any remaining segment that does not qualify as a pursuit event is classified
as a fixation (event label \texttt{FIXA}).
as a fixation (event label \texttt{FIXA}) (\fig{alg}, E4).
\subsection*{Operation}\label{op}
@ -538,21 +631,27 @@ remodnav <inputfile> <outputfile> \
where \texttt{<inputfile>} is the name of a tab-separated-value (TSV) text file
with one gaze coordinate sample per line. An input file can have any number of
columns, only the first two columns are read and interpreted as $X$ and $Y$
coordinates. The second argument \texttt{<outputfile>} is the file name of a
coordinates. Note that this constrains input data to a dense data representation,
i.e. either data from eye trackers with fixed sampling frequency throughout the
recording, or sparse data that has been transformed into a dense representation
beforehand.
The second argument \texttt{<outputfile>} is the file name of a
BIDS-compliant \citep{gorgolewski2016brain} TSV text file that will contain a
report on one detected eye movement event per line, with onset and offset time,
report on one classified eye movement event per line, with onset and offset time,
onset and offset coordinates, amplitude, peak velocity, median velocity and
average velocity. The remaining arguments are the only two mandatory
parameters: the conversion factor from pixels to visual degrees, \ie the visual
angle of a single (square) pixel (\texttt{<px2deg>} in \unit{deg}), and the
temporal sampling rate (\texttt{<sampling\_rate>} in \unit{Hz}).
Any other supported parameter can be added to the program invocation to override
the default values.
All additionally supported parameters (sorted by algorithm step) with their
A complete list of supported parameters (sorted by algorithm step) with their
description and default value, are listed in \tab{parameters}.
While the required user input is kept minimal, the number of configurable
parameters is purposefully large to facilitate optimal parameterization for
data with specific properties. Besides the list of detected events, a
visualization of the detection results, together with a time course of
data with specific properties. Besides the list of classified events, a
visualization of the classification results, together with a time course of
horizontal and vertical gaze position, and velocities is provided for
illustration and initial quality assessment of algorithm performance on each
input data file.
@ -576,7 +675,7 @@ input data file.
The selection of datasets and analyses for validating algorithm performance was
guided by three objectives: 1) compare to other existing
solutions; 2) demonstrate plausible results on data from prolonged gaze
coordinate recordings during natural viewing; and 3) illustrate result
coordinate recordings during viewing of dynamic, feature-rich stimuli; and 3) illustrate result
robustness on lower-quality data. The following three sections each introduce a
dataset and present the validation results for these objectives. All analysis
presented here are performed using default parameters (\tab{parameters}), with
@ -586,7 +685,7 @@ no dataset-specific tuning other than the built-in adaptive behavior.
\subsection*{Algorithm comparison}\label{ana_1}
Presently, \cite{Andersson2017} represents the most comprehensive comparative
study on eye movement detection algorithms. Moreover, the dataset employed
study on eye movement classification algorithms. Moreover, the dataset employed
in that study was made publicly available. Consequently, evaluating \remodnav\
performance on these data and using their metrics offers a straightforward
approach to relate this new development to alternative solutions.
@ -612,10 +711,15 @@ did not assign the same label. We limited this analysis to all time points that
were labeled as fixation, saccade, PSO, or pursuit by any method, hence
ignoring the rarely used NaN/blinks or ``undefined" category. For a direct
comparison with the results in \cite{Andersson2017}, the analysis was repeated
while also excluding samples labeled as pursuit. \tab{mclf} shows the
while also excluding samples labeled as pursuit.
In the labeled data, there was no distinction made between high- and low-velocity
PSOs, potentially because the literature following \citet{Nystrom2010AnData}
did not adopt their differentiation of PSOs into velocity categories.
All high- and low-velocity PSOs classified by \remodnav\ were therefore
collapsed into a single PSO category. \tab{mclf} shows the
misclassification rates for all pairwise comparisons, in all stimulus types.
In comparison to the NH algorithm, after which the proposed work was modelled,
\remodnav performed consistently better (32/93/70\% average misclassification for NH,
\remodnav\ performed consistently better (32/93/70\% average misclassification for NH,
vs. \imgMNALMclfWOP/\dotsRAALMclfWOP/ \videoRAALMclfWOP\% worst
misclassification for \remodnav\ in categories images, dots, and videos). Compared to all ten
algorithms evaluated in \citet{Andersson2017}, \remodnav\ exhibits the lowest
@ -683,14 +787,21 @@ for \remodnav]{Startsev2018}.
\end{table}
\fig{conf} shows confusion patterns for a comparison of algorithm
classifications with human labeling. While \remodnav\ does not achieve a
classifications with human labeling and displays the similarity between
classification decisions with Jaccard indices \citep[JI; ][]{jaccard1901etude}.
The JI is bound in range [0, 1] with higher values indicating higher similarity.
A value of 0.93 in the upper left cell of the very first matrix in \fig{conf}
for example indicates that 93\% of frames that are labeled as a fixation by
human coders RA and MN are the same. This index allows to quantify the
similarity of classifications independent of values in other cells.
While \remodnav\ does not achieve a
labeling similarity that reaches the human inter-rater agreement, it still
performs well. In particular, the relative magnitude of agreement with each
individual human coder for fixations, saccades, and PSOs, resembles the
agreement between the human coders. Classification of smooth
pursuits is consistent with human labels for the categories moving dots, and
videos. However, there is a substantial confusion of fixation and pursuit for
the static images. In a real-world application of \remodnav, pursuit detection
the static images. In a real-world application of \remodnav, pursuit classification
could be disabled (by setting a high pursuit velocity threshold) for data from
static images, if the occurrence of pursuit events can be ruled out a priori.
For this evaluation, however, no such intervention was made.
@ -718,6 +829,61 @@ For this evaluation, however, no such intervention was made.
\end{figure*}
\begin{table}[tbp]
% table caption is above the table
\caption{Cohen's Kappa reliability between human coders (MN, RA), and \remodnav\ (AL)
with each of the human coders.
}
\label{tab:kappa} % Give a unique label
% For LaTeX tables use
\begin{tabular*}{0.5\textwidth}{c @{\extracolsep{\fill}}llll}
\textbf {Fixations} & & & \\
\hline\noalign{\smallskip}
Comparison & Images & Dots & Videos \\
\noalign{\smallskip}\hline\noalign{\smallskip}
MN versus RA & \kappaRAMNimgFix & \kappaRAMNdotsFix & \kappaRAMNvideoFix \\
AL versus RA & \kappaALRAimgFix & \kappaALRAdotsFix & \kappaALRAvideoFix \\
AL versus MN & \kappaALMNimgFix & \kappaALMNdotsFix & \kappaALMNvideoFix \\
\noalign{\smallskip}
\textbf{Saccades} & & & \\
\hline\noalign{\smallskip}
Comparison & Images & Dots & Videos \\
\noalign{\smallskip}\hline\noalign{\smallskip}
MN versus RA & \kappaRAMNimgSac & \kappaRAMNdotsSac & \kappaRAMNvideoSac \\
AL versus RA & \kappaALRAimgSac & \kappaALRAdotsSac & \kappaALRAvideoSac \\
AL versus MN & \kappaALMNimgSac & \kappaALMNdotsSac & \kappaALMNvideoSac \\
\noalign{\smallskip}
\textbf{PSOs} & & & \\
\hline\noalign{\smallskip}
Comparison & Images & Dots & Videos \\
\noalign{\smallskip}\hline\noalign{\smallskip}
MN versus RA & \kappaRAMNimgPSO & \kappaRAMNdotsPSO & \kappaRAMNvideoPSO \\
AL versus RA & \kappaALRAimgPSO & \kappaALRAdotsPSO & \kappaALRAvideoPSO \\
AL versus MN & \kappaALMNimgPSO & \kappaALMNdotsPSO & \kappaALMNvideoPSO \\
\noalign{\smallskip}\hline
\end{tabular*}
\end{table}
In addition to the confusion analysis and again following \citet{Andersson2017},
we computed Cohen's Kappa \citep{cohen1960coefficient} for an additional measure
of similarity between human and algorithm performance. It quantifies the
sample-by-sample agreement between two ratings following equation \ref{eq:kappa}:
%
\begin{equation}\label{eq:kappa}
K = \frac{P_o - P_c}{1- P_c}
\end{equation}
%
where $P_o$ is the observed proportion of agreement between the ratings, and
$P_c$ is the proportion of chance agreement. A value of $K=1$ indicates perfect
agreement, and $K=0$ indicates chance level agreement.
Table \ref{tab:kappa} displays the resulting values
between the two human experts, and \remodnav\ with each of the experts, for
each stimulus category and the three event types used in \citet{Andersson2017},
namely fixations, saccades, and PSOs (compare to \citet{Andersson2017}, table 7).
For all event types and stimulus categories, \remodnav\ performs on par or better
than the original NH algorithm, and in many cases on par or better than the best
of all algorithms evaluated in \citet{Andersson2017} within an event or stimulus type.
In order to further rank the performance of the proposed algorithm with respect
to the ten algorithms studied in \citet{Andersson2017}, we followed their
approach to compute root mean square deviations (RMSD) from human labels for
@ -726,28 +892,27 @@ durations, plus number of events) for each stimulus category (images, dots,
videos) and event type (fixations, saccades, PSOs, pursuits). This measure
represents a scalar distribution dissimilarity score that can be used as an
additional comparison metric of algorithm performance that focuses on overall
number and durations of detected events, instead of sample-by-sample
number and durations of classified events, instead of sample-by-sample
misclassification. The RMSD measure has a lower bound of $0.0$ (identical to
the average of both human raters), with higher values indicating larger
differences \citep[for detail information on the calculation of this metric
see][]{Andersson2017}.
\tab{rmsd} reproduces \citet[Tables
3-6]{Andersson2017}, and the RMSD calculation for the added rows on \remodnav\
is based on the scores for the human raters given in these original tables. As
\tab{rmsd} is modelled after \citet[Tables
3-6]{Andersson2017}, appended with \remodnav, showing RMSD based on the scores of human raters given in the original tables. As
acknowledged by the authors, the absolute value of the RMSD scores is not
informative due to scaling with respect to the respective maximum value of each
characteristic. Therefore, we converted RMSDs for each algorithm and event
type into zero-based ranks (lower is more human-like).
The LNS algorithm \citep{Larsson2013} was found to have the most human-like
performance for saccade and PSO detection in \cite{Andersson2017}. \remodnav\
performance for saccade and PSO classification in \cite{Andersson2017}. \remodnav\
performs comparable to LNS for both event types (saccades: $2.0$ vs. $3.3$;
PSOs: $2.3$ vs. $2.0$, mean rank across stimulus categories for LNS and \remodnav,
respectively).
Depending on the stimulus type, different algorithms performed best for
fixation detection. NH performed best for images and videos, but worst for
fixation classification. NH performed best for images and videos, but worst for
moving dots. \remodnav\ outperforms all other algorithms in the dots category,
and achieves rank 5 and 6 (middle range) for videos and images, respectively.
Across all stimulus and event categories, \remodnav\ achieves a mean ranking
@ -757,11 +922,15 @@ of $2.9$, and a mean ranking of $3.2$ when not taking smooth pursuit into accoun
% table caption is above the table
\caption{Comparison of event duration statistics (mean, standard deviation, and number
of events) for image, dot, and video
stimuli. This table reproduces \citet[Tables 3-6]{Andersson2017}, and
stimuli. This table is modeled after \citet[Tables 3-6]{Andersson2017}, and
root-mean-square-deviations (RMSD) from human raters are shown
for fixations, saccades, PSOs, and pursuit as zero-based ranks (rank zero
is closest to the average of the two human raters). Rows for \remodnav\
have been added.}
is closest to the average of the two human raters). Summary statistics for
all algorithms used in \citet{Andersson2017} were taken from their publicly
available GitHub repository
(github.com/richardandersson/EyeMovementDetectorEvaluation). Cohens Kappa
was computed for the complete set of algorithms in \citet{Andersson2017} and
\remodnav .}
\label{tab:rmsd} % Give a unique label
% For LaTeX tables use
\begin{small}
@ -831,25 +1000,24 @@ of $2.9$, and a mean ranking of $3.2$ when not taking smooth pursuit into accoun
\end{small}
\end{table*}
Taken together, \remodnav\ yields classification results that are, on average,
more human-like than any other algorithm tested on the dataset and metrics put
forth by \citet{Andersson2017}. In particular, its performance largely equals
or exceeds that of the original NH algorithm. NH outperforms it only for
fixation detection in the image and video category, but in these categories
fixation classification in the image and video category, but in these categories
\remodnav\ also classifies comparatively well. These results are an indication
that the changes to the NH algorithm proposed here to improve upon its
robustness are not detrimental to its performance on data from conventional
paradigms and stimuli.
\subsection*{Prolonged natural viewing}\label{ana_2}
\subsection*{Prolonged viewing of dynamic stimuli}\label{ana_2}
Given that \remodnav\ yielded plausible results for the "video" stimulus
category data in the \citet{Andersson2017} dataset (\fig{conf}, and
\tab{rmsd}, right columns), we determined
whether it is capable of analyzing data from dynamic stimulation without a
trial structure.
whether it is capable of analyzing data from dynamic stimulation in prolonged
(\unit[$~15$]{min}) recordings.
As a test dataset we used publicly available eye tracking data from the
\textit{studyforrest.org} project, where 15~participants were recorded watching
@ -860,7 +1028,7 @@ Ontario, Canada) and a sampling rate of \unit[1000]{Hz}. The movie stimulus was
presented on a \unit[$522\times294$]{mm} LCD monitor at a resolution of
\unit[$1920\times1280$]{px} and a viewing distance of \unit[85]{cm}. Participants
watched the movie in eight approximately \unit[15]{min} long segments,
with measurement recalibration before every segment.
with recalibration of the eye tracker before every segment.
\begin{figure*}[tbp]
\includegraphics[trim=0 8mm 3mm 0,clip,width=.5\textwidth]{img/mainseq_lab}
@ -890,10 +1058,10 @@ match previous reports in the literature, such as a strong bias towards short
\unit[10-20]{ms} \citep[Fig.~11]{Nystrom2010AnData}, and a non-Gaussian saccade
duration distribution located below \unit[100]{ms} \citep[Fig.~8, albeit for
static scene perception]{Nystrom2010AnData}.
%
Overall, the presented summary statistics suggest that \remodnav\ is capable
of detecting eye movements with plausible characteristics, in prolonged
gaze recordings without a trial structure. A visualization of such a detection
of classifying eye movements with plausible characteristics, in prolonged
gaze recordings. A visualization of such a classification
result is depicted in \fig{remodnav} (top row).
\begin{figure*}
@ -922,18 +1090,23 @@ result is depicted in \fig{remodnav} (top row).
\begin{figure*}[tbp]
\includegraphics[trim=0 8mm 0 0,clip,width=1\textwidth]{img/remodnav_lab.pdf} \\
\includegraphics[trim=0 0 0 2mm,clip,width=1\textwidth-1mm]{img/remodnav_mri.pdf}\\
\caption{Exemplary eye movement detection results for the same \unit[10]{s} excerpt
\includegraphics[trim=0 0 0 0,clip,width=1\textwidth]{img/remodnav_mri.pdf}\\
\includegraphics[trim=0 0 0 0,clip,width=0.49\textwidth]{img/rawtrace_lab.pdf}
\includegraphics[trim=0 0 0 0,clip,width=0.49\textwidth]{img/rawtrace_mri.pdf}\\
\caption{Exemplary eye movement classification results for the same \unit[10]{s} excerpt
of a movie stimulus for a single participant in the high quality lab sample (top),
and in the lower quality MRI sample (bottom). The plots show filtered gaze coordinates
(black), computed velocity time series (gray) overlayed on the eye movement event segmentation
with periods of fixation (green), pursuit (beige), saccades (blue), and
high/low-velocity post-saccadic oscillations (dark/light purple). The variable noise
level, and prolonged signal loss (white) visible in the MRI sample represent
a challenge for algorithms. \remodnav\ uses an adaptive approach that determines major
and in the lower quality MRI sample (middle). The plots show filtered gaze
coordinates (black), computed velocity time series (gray) overlayed on the eye
movement event segmentation with periods of fixation (green), pursuit (beige),
saccades (blue), and high/low-velocity post-saccadic oscillations (dark/light
purple). The bottom panel shows the first \unit[2]{s} of unfiltered gaze coordinates
(black) and unfiltered velocity time series (gray) for lab (left) and mri (right)
sample in greater detail. The variable noise level, and prolonged signal loss (white
in top panel) visible in the MRI sample represent a challenge for algorithms.
\remodnav\ uses an adaptive approach that determines major
saccade events first, and subsequently tunes the velocity threshold to short time
windows between these events. Figures like this accompany the program output to
facilitate quality control and discovery of inappropriate preprocessing and detection
facilitate quality control and discovery of inappropriate preprocessing and classification
parameterization.}
\label{fig:remodnav}
@ -944,9 +1117,9 @@ result is depicted in \fig{remodnav} (top row).
An explicit goal for \remodnav\ development was robust performance on
lower-quality data. While lack of quality cannot be arbitrarily compensated and
will inevitably lead to misses in eye movement detection, it is beneficial for
any further analysis if operation on noisy data does not introduce unexpected
event property biases.
can inevitably lead to misses in eye movement classification if too high,
it is beneficial for any further analysis if operation on noisy data does not
introduce unexpected event property biases.
In order to investigate noise-robustness we ran \remodnav\ on another
publicly available dataset from the \textit{studyforrest.org} project, where 15
@ -962,15 +1135,45 @@ previous stimulation setup. The eye tracking camera was mounted outside the
scanner bore and recorded the participants' left eye at a distance of about
\unit[100]{cm}. Compared to the lab-setup, physical limitations of the scanner
environment, and sub-optimal infrared illumination led to substantially
noisier data, as evident from a generally higher amount of data loss and a
larger spatial uncertainty \citep[Technical Validation]{Hanke2016}. An example
of the amplified and variable noise pattern is shown in \fig{remodnav} (bottom
row, black lines). Except for the differences in stimulation setup, all other
noisier data, as evident from a larger spatial uncertainty
\citep[Technical Validation]{Hanke2016}, a generally higher amount of data loss,
and more events with a velocity above \unit[800]{deg/s}.
Following common data quality criteria used to warrant exclusion by
\citet{Holmqvist2012}, a higher amount of zero values, a greater number of
events with a velocity above \unit[800]{deg/s}, and lower spatial accuracy
can be indicative of lower quality data.
The average amount of data loss in the MRI sample was three times higher than
in the laboratory setting (15.1\% versus 4.1\% in the lab), with six out
of 15 subjects having one or more movie segments with data loss greater than 30\%.
In the laboratory setting, in comparison, zero out of 15 subjects had one or
more movie segments with data loss greater than 30\% \citep[Table 1]{Hanke2016}.
Figure \ref{vels} highlights the higher amount of extreme velocities in the MRI sample,
even though the stimulus size was smaller than in the laboratory setting.
Finally, the average spatial accuracy at the start of a recording, assessed with a 13-point
calibration procedure, was \unit[0.58]{degrees of visual angle} for the MRI sample and
\unit[0.45]{degrees} for the lab sample \citep[Technical Validation]{Hanke2016}.
An example of the amplified and variable noise pattern is shown in \fig{remodnav} (bottom
row, gray lines). Except for the differences in stimulation setup, all other
aspects of data acquisition, eye tracker calibration, and data processing
were identical to the previous dataset.
\begin{figure}
\includegraphics[width=0.5\textwidth]{img/velhist.pdf}
\caption{
Comparison of sample velocity distributions for MRI and laboratory setting
across all measurements and participants (excluding samples during periods
of signal-loss). The MRI sample exhibits a larger fraction of higher
velocities, despite a 30\% smaller stimulus size.
}
\label{vels}
\end{figure}
We performed the identical analysis as before, in order to compare performance
between a high and lower-quality data acquisition.
between a high and lower-quality data acquisition. This approach differs from
the common approach of adding increasing levels of artificial noise to data
(as done for example in \citet{hessels2017noise}),
but bears the important advantage of incorporating real lower-quality data
characteristics instead of potentially inappropriate or unnatural noise.
Figures~\ref{fig:overallComp}-\ref{fig:remodnav} depict the results for the
lab-quality dataset, and the MRI-scanner dataset in the top and bottom rows,
respectively.
@ -979,14 +1182,14 @@ respectively.
% data loss, perhaps in addition to a subjective rating from the person
% responsible for the recording"
Overall, the detection results exhibit strong similarity, despite the potential
Overall, the classification results exhibit strong similarity, despite the potential
behavioral impact of watching a movie while lying on their back and looking
upwards on the participants, or the well known effect of increasing fatigue
upwards on the participants, or the well known effect of increasing fatigue \citep{wakefulness}
during a two-hour session in an MRI-scanner. In particular, saccade amplitude
and peak velocity exhibit a clear main-sequence relationship that resembles
that found for the lab acquisition (\fig{overallComp}). Duration distributions
for fixations, PSOs, and pursuits are strikingly similar between the two
datasets (\fig{dist}), except for a generally lower number of detected events
datasets (\fig{dist}), except for a generally lower number of classified events
for the MRI experiment, which could be explained by the higher noise level and
fraction of signal loss. There is a notable difference regarding the saccade
duration distributions, with a bias towards shorter saccades in the MRI
@ -997,56 +1200,79 @@ dataset. This effect may be attributable to the differences in stimulus size
\section*{Conclusion}\label{con}
Based on the adaptive, velocity-based algorithm for fixation, saccade, and PSO
detection by \cite{Nystrom2010AnData}, we have developed an improved algorithm
that, in contrast to the original, performs robustly on prolonged recordings
with dynamic stimulation, without a trial structure and variable noise levels,
and also supports the detection of smooth pursuit events. Through a series of
classification by \cite{Nystrom2010AnData}, we have developed an improved algorithm
that performs robustly on prolonged or short recordings
with dynamic stimulation, with potentially variable noise levels,
and also supports the classification of smooth pursuit events. Through a series of
validation analyses we have shown that its performance is comparable to or
better than ten other contemporary detection algorithms, and that plausible
detection results are achieved on high and lower quality data.
better than ten other contemporary algorithms, and that plausible
classification results are achieved on high and lower quality data.
These aspects of algorithm capabilities and performance suggest that \remodnav\
is a state-of-the-art tool for eye movement detection with particular relevance
for emerging complex, naturalistic data collections paradigms, such as
mobile or outdoor aquisition, or the combination of eye tracking and functional
is a state-of-the-art tool for eye movement classification with particular relevance
for emerging complex data collections paradigms with dynamic stimulation, such as
the combination of eye tracking and functional
MRI in simultaneous measurements.
The proposed algorithm is rule-based, hence can be applied to data without
prior training, apart from the adaptive estimation of velocity thresholds.
This aspect distinguishes it from other recent developments based on deep
neural networks \citep{Startsev2018}, and machine-learning in general
\citep{Zemblys2018}. Such algorithms tend to require substantial amount of
(labeled) training data, which can be a critical limitation in the context of a
research study. However, in its present form \remodnav\ cannot be used for
\citep{Zemblys2018}.
Some statistical learning algorithms require (labeled) training data,
which can be a limitation in the context of a research study.
However, in its present form \remodnav\ cannot be used for
real-time data analysis, as its approach for time series chunking is based
on an initial sorting of major saccade events across the entire time series.
The proposed algorithm presently does not support the detection of eye blinks
The proposed algorithm presently does not support the classification of eye blinks
as a category distinct from periods of general signal loss. While such a feature
could potentially be added, the current default preprocessing aims at removing
blink-related signal. Lastly, the evaluation results presented here are based
blink-related signal. The algorithm maintains a distinction between high- and
low-velocity PSOs first introduced by \citet{Nystrom2010AnData}, although, to our knowledge,
the present literature does not make use of such a distinction. Algorithm users
are encouraged to decide on a case-by-case basis whether to lump these event categories
together into a general PSO category, as done in our own validation analyses.
As a general remark it is also noteworthy that eye tracking systems using
pupil corneal reflection (pupil-CR) eye tracking may bias data towards premature PSO onset times and
inflated PSO peak velocities (see \citet{HOOGE20166}).
In deciding whether and how to interpret PSO events, it needs to be considered
whether the eye tracking device may have introduced biases in the data.
Lastly, the evaluation results presented here are based
on data with a relatively high temporal resolution (\unit[0.5 and 1]{kHz}). While
the algorithm does not impose any hard constraints on data acquisition parameters,
its performance on data from low-end, consumer grade hardware (\eg \unit[50]{Hz}
sampling rate) has not been tested.
Just as \cite{Andersson2017}, we considered human raters as a gold standard
reference for event detection when evaluating algorithms. The implications of
reference for event classification when evaluating algorithms. The implications of
the results presented herein are hence only valid if this assumption is
warranted. Some authors voice concerns \cite[\eg][]{5523936}, regarding
potential biases that may limit generalizability. Nevertheless, human-made
event labels are a critical component of algorithm validation, as pointed out
by \cite{Hooge2018}.
The validation analyses presented here are based on three different datasets:
a manually annoted dataset \citep{Andersson2017}, and two datasets with
prolonged recordings using movie stimuli \citep{Hanke2016}.
Beyond our own validation, a recent evaluation of nine different smooth
pursuit algorithms by Startsev, Agtzidis and Dorr as part of their recent
paper \citep{Startsev2018} also provides metrics for \remodnav.
In their analysis, algorithm performance was evaluated against a partially
hand-labelled eye movement annotation of the Hollywood2 dataset \citep{Mathe2012}.
We refrain from restating their methodology or interpreting their results here,
but encourage readers to consult this independent
report\footnote{https://www.michaeldorr.de/smoothpursuit/}.
\remodnav\ aims to be a readily usable tool, available as cross platform
compatible, free and open source software, with a simple command line interface
and carefully chosen default settings. However, as evident from numerous
algorithm evaluations
\citep[\eg][]{Andersson2017,Larsson2013,Zemblys2018,5523936} different
\citep[\eg][]{Andersson2017,Larsson2013,Zemblys2018,5523936}, different
underlying stimulation, and data characteristics can make certain algorithms or
parameterizations more suitable than others for particular applications. The
provided implementation of the \remodnav\ algorithm \citep{michael_hanke_2019_2651042}
acknowledges this fact by
exposing a range of parameters through its user interface that can be altered
in order to tune the detection for a particular use case.
in order to tune the classification for a particular use case.
The latest version of \remodnav\ can be installed from
PyPi\footnote{https://pypi.org/project/remodnav} via \texttt{pip install
@ -1060,7 +1286,10 @@ repository\footnote{https://github.com/psychoinformatics-de/paper-remodnav/}.
All required input data, from \cite{Andersson2017} and the
\textit{studyforrest.org} project, are referenced in this repository at precise
versions as DataLad\footnote{\url{http://datalad.org}} subdatasets, and can be
obtained on demand.
obtained on demand. The repository constitutes an automatically reproducible
research object, and readers interested in verifying the results and claims of
our paper can recompute and plot all results with a single command after cloning
the repository.
\subsection*{Author contributions}
@ -1116,7 +1345,7 @@ labeled eye tracking dataset used for validation under an open-source license.
\end{acknowledgements}
\bibliographystyle{spbasic} % basic style, author-year citations
\bibliography{EyeGaze,tools,references}
\bibliography{references}
% References can be listed in any standard referencing style that uses a numbering system
% (i.e. not Harvard referencing style), and should be consistent between references within

767
references.bib
View file

@ -1,67 +1,72 @@
@Article{ALH+2017,
author="Andersson, Richard
and Larsson, Linnea
and Holmqvist, Kenneth
and Stridh, Martin
and Nystr{\"o}m, Marcus",
title="One algorithm to rule them all? An evaluation and discussion of ten eye movement event-detection algorithms",
journal="Behavior Research Methods",
year="2017",
month="Apr",
day="01",
volume="49",
number="2",
pages="616--637",
abstract="Almost all eye-movement researchers use algorithms to parse raw data and detect distinct types of eye movement events, such as fixations, saccades, and pursuit, and then base their results on these. Surprisingly, these algorithms are rarely evaluated. We evaluated the classifications of ten eye-movement event detection algorithms, on data from an SMI HiSpeed 1250 system, and compared them to manual ratings of two human experts. The evaluation focused on fixations, saccades, and post-saccadic oscillations. The evaluation used both event duration parameters, and sample-by-sample comparisons to rank the algorithms. The resulting event durations varied substantially as a function of what algorithm was used. This evaluation differed from previous evaluations by considering a relatively large set of algorithms, multiple events, and data from both static and dynamic stimuli. The main conclusion is that current detectors of only fixations and saccades work reasonably well for static stimuli, but barely better than chance for dynamic stimuli. Differing results across evaluation methods make it difficult to select one winner for fixation detection. For saccade detection, however, the algorithm by Larsson, Nystr{\"o}m and Stridh (IEEE Transaction on Biomedical Engineering, 60(9):2484--2493,2013) outperforms all algorithms in data from both static and dynamic stimuli. The data also show how improperly selected algorithms applied to dynamic data misestimate fixation and saccade properties.",
issn="1554-3528",
doi="10.3758/s13428-016-0738-9"
author="Andersson, Richard
and Larsson, Linnea
and Holmqvist, Kenneth
and Stridh, Martin
and Nystr{\"o}m, Marcus",
title="One algorithm to rule them all? An evaluation and discussion of ten eye movement event-detection algorithms",
journal="Behavior Research Methods",
year="2017",
month="Apr",
day="01",
volume="49",
number="2",
pages="616--637",
abstract="Almost all eye-movement researchers use algorithms to parse raw data and detect distinct types of eye movement events, such as fixations, saccades, and pursuit, and then base their results on these. Surprisingly, these algorithms are rarely evaluated. We evaluated the classifications of ten eye-movement event detection algorithms, on data from an SMI HiSpeed 1250 system, and compared them to manual ratings of two human experts. The evaluation focused on fixations, saccades, and post-saccadic oscillations. The evaluation used both event duration parameters, and sample-by-sample comparisons to rank the algorithms. The resulting event durations varied substantially as a function of what algorithm was used. This evaluation differed from previous evaluations by considering a relatively large set of algorithms, multiple events, and data from both static and dynamic stimuli. The main conclusion is that current detectors of only fixations and saccades work reasonably well for static stimuli, but barely better than chance for dynamic stimuli. Differing results across evaluation methods make it difficult to select one winner for fixation detection. For saccade detection, however, the algorithm by Larsson, Nystr{\"o}m and Stridh (IEEE Transaction on Biomedical Engineering, 60(9):2484--2493,2013) outperforms all algorithms in data from both static and dynamic stimuli. The data also show how improperly selected algorithms applied to dynamic data misestimate fixation and saccade properties.",
issn="1554-3528",
doi="10.3758/s13428-016-0738-9"
}
@article{jaccard1901etude,
title={{\'E}tude comparative de la distribution florale dans une portion des Alpes et des Jura},
author={Jaccard, Paul},
journal={Bull Soc Vaudoise Sci Nat},
volume={37},
pages={547--579},
year={1901}
title={{\'E}tude comparative de la distribution florale dans une portion des Alpes et des Jura},
author={Jaccard, Paul},
journal={Bull Soc Vaudoise Sci Nat},
volume={37},
pages={547--579},
year={1901}
}
@article{amit2017temporal,
title={Temporal dynamics of saccades explained by a self-paced process},
author={Amit, Roy and Abeles, Dekel and Bar-Gad, Izhar and Yuval-Greenberg, Shlomit},
journal={Scientific reports},
volume={7},
number={1},
pages={886},
year={2017},
doi = {10.1038/s41598-017-00881-7},
publisher={Nature Publishing Group}
title={Temporal dynamics of saccades explained by a self-paced process},
author={Amit, Roy and Abeles, Dekel and Bar-Gad, Izhar and Yuval-Greenberg, Shlomit},
journal={Scientific reports},
volume={7},
number={1},
pages={886},
year={2017},
doi = {10.1038/s41598-017-00881-7},
publisher={Nature Publishing Group}
}
@article{bahill1975main,
title={The main sequence, a tool for studying human eye movements},
author={Bahill, A Terry and Clark, Michael R and Stark, Lawrence},
journal={Mathematical Biosciences},
volume={24},
number={3-4},
pages={191--204},
year={1975},
publisher={Elsevier},
url={https://doi.org/10.1016/0025-5564(75)90075-9}
title={The main sequence, a tool for studying human eye movements},
author={Bahill, A Terry and Clark, Michael R and Stark, Lawrence},
journal={Mathematical Biosciences},
volume={24},
number={3-4},
pages={191--204},
year={1975},
publisher={Elsevier},
url={https://doi.org/10.1016/0025-5564(75)90075-9}
}
@article{GOLTZ1997789,
title = "Vertical Eye Position Control in Darkness: Orbital Position and Body Orientation Interact to Modulate Drift Velocity",
journal = "Vision Research",
volume = "37",
number = "6",
pages = "789 - 798",
year = "1997",
issn = "0042-6989",
doi = "https://doi.org/10.1016/S0042-6989(96)00217-9",
author = "H.C Goltz and E.L Irving and M.J Steinbach and M EizenamnIZENMAN"
title = "Vertical Eye Position Control in Darkness: Orbital Position and Body Orientation Interact to Modulate Drift Velocity",
journal = "Vision Research",
volume = "37",
number = "6",
pages = "789 - 798",
year = "1997",
issn = "0042-6989",
doi = "https://doi.org/10.1016/S0042-6989(96)00217-9",
author = "H.C Goltz and E.L Irving and M.J Steinbach and M EizenamnIZENMAN"
}
@article{cherici2012,
author = {Cherici, Claudia and Kuang, Xutao and Poletti, Martina and Rucci, Michele},
title = "{Precision of sustained fixation in trained and untrained observers}",
@ -75,19 +80,665 @@ author = "H.C Goltz and E.L Irving and M.J Steinbach and M EizenamnIZENMAN"
doi = {10.1167/12.6.31},
}
@thesis{Sch2017,
author = {Ulrike Schnaithman},
title = {Combining and testing filter and detection algorithms for post-experimental analysis of eye tracking data on dynamic stimuli},
note = {B.Sc. thesis submitted to the faculty of natural sciences at the Otto von Guericke University, Magdeburg, Germany},
year = 2017
author = {Ulrike Schnaithman},
title = {Combining and testing filter and detection algorithms for post-experimental analysis of eye tracking data on dynamic stimuli},
note = {B.Sc. thesis submitted to the faculty of natural sciences at the Otto von Guericke University, Magdeburg, Germany},
year = 2017
}
@misc{michael_hanke_2019_2651042,
author = {Michael Hanke and
author = {Michael Hanke and
Asim H Dar and
Adina Wagner},
title = {psychoinformatics-de/remodnav: Submission time},
month = apr,
year = 2019,
doi = {10.5281/zenodo.2651042},
title = {psychoinformatics-de/remodnav: Submission time},
month = apr,
year = 2019,
doi = {10.5281/zenodo.2651042},
}
@article{wakefulness,
title = "Decoding Wakefulness Levels from Typical fMRI Resting-State Data Reveals Reliable Drifts between Wakefulness and Sleep",
journal = "Neuron",
volume = "82",
number = "3",
pages = "695 - 708",
year = "2014",
issn = "0896-6273",
doi = "https://doi.org/10.1016/j.neuron.2014.03.020",
url = "http://www.sciencedirect.com/science/article/pii/S0896627314002505",
author = "Enzo Tagliazucchi and Helmut Laufs"
}
@article{Hannula2010,
abstract = {Results of several investigations indicate that eye movements can reveal memory for elements of previous experience. These effects of memory on eye movement behavior can emerge very rapidly, changing the efficiency and even the nature of visual processing without appealing to verbal reports and without requiring conscious recollection. This aspect of eye-movement based memory investigations is particularly useful when eye movement methods are used with special populations (e.g., young children, elderly individuals, and patients with severe amnesia), and also permits use of comparable paradigms in animals and humans, helping to bridge different memory literatures and permitting cross-species generalizations. Unique characteristics of eye movement methods have produced findings that challenge long-held views about the nature of memory, its organization in the brain, and its failures in special populations. Recently, eye movement methods have been successfully combined with neuroimaging techniques such as fMRI, single-unit recording, and MEG, permitting more sophisticated investigations of memory. Ultimately, combined use of eye-tracking with neuropsychological and neuroimaging methods promises to provide a more comprehensive account of brain-behavior relationships and adheres to the {\&}{\#}8220;converging evidence{\&}{\#}8221; approach to cognitive neuroscience.},
author = {Hannula, Deborah E. and Althoff, Robert R and Warren, David E and Riggs, Lily and Cohen, Neal J and Ryan, Jennifer D},
doi = {10.3389/fnhum.2010.00166},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Hannula et al. - 2010 - Worth a glance using eye movements to investigate the cognitive neuroscience of memory.pdf:pdf},
issn = {16625161},
journal = {Frontiers in Human Neuroscience},
keywords = {Amnesia,Eye Movements,Hippocampus,MEG,Memory,fMRI},
month = {oct},
pages = {166},
publisher = {Frontiers},
title = {{Worth a glance: using eye movements to investigate the cognitive neuroscience of memory}},
volume = {4},
year = {2010}
}
@article{Andersson2017,
abstract = {Almost all eye-movement researchers use algorithms to parse raw data and detect distinct types of eye movement events, such as fixations, saccades, and pursuit, and then base their results on these. Surprisingly, these algorithms are rarely evaluated. We evaluated the classifications of ten eye-movement event detection algorithms, on data from an SMI HiSpeed 1250 system, and compared them to manual ratings of two human experts. The evaluation focused on fixations, saccades, and post-saccadic oscillations. The evaluation used both event duration parameters, and sample-by-sample comparisons to rank the algorithms. The resulting event durations varied substantially as a function of what algorithm was used. This evaluation differed from previous evaluations by considering a relatively large set of algorithms, multiple events, and data from both static and dynamic stimuli. The main conclusion is that current detectors of only fixations and saccades work reasonably well for static stimuli, but barely better than chance for dynamic stimuli. Differing results across evaluation methods make it difficult to select one winner for fixation detection. For saccade detection, however, the algorithm by Larsson, Nystr{\"{o}}m and Stridh (IEEE Transaction on Biomedical Engineering, 60(9):2484-2493,2013) outperforms all algorithms in data from both static and dynamic stimuli. The data also show how improperly selected algorithms applied to dynamic data misestimate fixation and saccade properties.},
author = {Andersson, Richard and Larsson, Linnea and Holmqvist, Kenneth and Stridh, Martin and Nystr{\"{o}}m, Marcus},
doi = {10.3758/s13428-016-0738-9},
issn = {1554-3528},
journal = {Behavior Research Methods},
keywords = {Eye-tracking,Inter-rater reliability,Parsing},
month = {apr},
number = {2},
pages = {616--637},
pmid = {27193160},
title = {{One algorithm to rule them all? An evaluation and discussion of ten eye movement event-detection algorithms}},
volume = {49},
year = {2017}
}
@inproceedings{Holmqvist2012,
address = {New York, New York, USA},
author = {Holmqvist, Kenneth and Nystr{\"{o}}m, Marcus and Mulvey, Fiona},
booktitle = {Proceedings of the Symposium on Eye Tracking Research and Applications - ETRA '12},
doi = {10.1145/2168556.2168563},
isbn = {9781450312219},
keywords = {accuracy,data quality,eye movements,eye tracker,latency,precision},
pages = {45},
publisher = {ACM Press},
title = {{Eye tracker data quality}},
year = {2012}
}
@article{Toiviainen2014,
abstract = {We investigated neural correlates of musical feature processing with a decoding approach. To this end, we used a method that combines computational extraction of musical features with regularized multiple regression (LASSO). Optimal model parameters were determined by maximizing the decoding accuracy using a leave-one-out cross-validation scheme. The method was applied to functional magnetic resonance imaging (fMRI) data that were collected using a naturalistic paradigm, in which participants' brain responses were recorded while they were continuously listening to pieces of real music. The dependent variables comprised musical feature time series that were computationally extracted from the stimulus. We expected timbral features to obtain a higher prediction accuracy than rhythmic and tonal ones. Moreover, we expected the areas significantly contributing to the decoding models to be consistent with areas of significant activation observed in previous research using a naturalistic paradigm with fMRI. Of the six musical features considered, five could be significantly predicted for the majority of participants. The areas significantly contributing to the optimal decoding models agreed to a great extent with results obtained in previous studies. In particular, areas in the superior temporal gyrus, Heschl's gyrus, Rolandic operculum, and cerebellum contributed to the decoding of timbral features. For the decoding of the rhythmic feature, we found the bilateral superior temporal gyrus, right Heschl's gyrus, and hippocampus to contribute most. The tonal feature, however, could not be significantly predicted, suggesting a higher inter-participant variability in its neural processing. A subsequent classification experiment revealed that segments of the stimulus could be classified from the fMRI data with significant accuracy. The present findings provide compelling evidence for the involvement of the auditory cortex, the cerebellum and the hippocampus in the processing of musical features during continuous listening to music.},
author = {Toiviainen, Petri and Alluri, Vinoo and Brattico, Elvira and Wallentin, Mikkel and Vuust, Peter},
doi = {10.1016/J.NEUROIMAGE.2013.11.017},
issn = {1053-8119},
journal = {NeuroImage},
month = {mar},
pages = {170--180},
publisher = {Academic Press},
title = {{Capturing the musical brain with Lasso: Dynamic decoding of musical features from fMRI data}},
volume = {88},
year = {2014}
}
@article{Holsanova2006,
abstract = {The aim of this article is to compare general assumptions about newspaper reading with eye-tracking data from readers' actual interaction with a newspaper. First, we extract assumptions about the way people read newspapers from socio-semiotic research. Second, we apply these assumptions by analysing a newspaper spread; this is done without any previous knowledge of actual reading behaviour. Finally, we use eye-tracking to empirically examine so-called entry points and reading paths. Eye movement data on reading newspaper spreads are analysed in three different ways: the time sequence in which different areas attract attention is calculated in order to determine reading priorities; the amount of time spent on different areas is calculated in order to determine which areas have been read most; the depth of attention is calculated in order to determine how carefully those areas have been read. General assumptions extracted from the socio-semiotic framework are compared to the results of the actual behaviour of subjects reading the newspaper spread. The results show that the empirical data confirm some of the extracted assumptions. The reading paths of the five subjects participating in the eye-tracking tests suggest that there are three main categories of readers: editorial readers, overview readers and focused readers.},
author = {Holsanova, Jana and Rahm, Henrik and Holmqvist, Kenneth},
doi = {10.1177/1470357206061005},
issn = {1470-3572},
journal = {Visual Communication},
month = {feb},
number = {1},
pages = {65--93},
publisher = {Sage PublicationsSage CA: Thousand Oaks, CA},
title = {{Entry points and reading paths on newspaper spreads: comparing a semiotic analysis with eye-tracking measurements}},
volume = {5},
year = {2006}
}
@article{Gordon2006,
author = {Gordon, Peter C. and Hendrick, Randall and Johnson, Marcus and Lee, Yoonhyoung},
doi = {10.1037/0278-7393.32.6.1304},
issn = {1939-1285},
journal = {Journal of Experimental Psychology: Learning, Memory, and Cognition},
number = {6},
pages = {1304--1321},
title = {{Similarity-based interference during language comprehension: Evidence from eye tracking during reading.}},
volume = {32},
year = {2006}
}
@article{Tikka2012,
abstract = {We outline general theoretical and practical implications of what we promote as enactive cinema for the neuroscientific study of online socio-emotional interaction. In a real-time functional magnetic resonance imaging (rt-fMRI) setting, participants are immersed in cinematic experiences that simulate social situations. While viewing, their physiological reactions - including brain responses - are tracked, representing implicit and unconscious experiences of the on-going social situations. These reactions, in turn, are analysed in real-time and fed back to modify the cinematic sequences they are viewing while being scanned. Due to the engaging cinematic content, the proposed setting focuses on living-by in terms of shared psycho-physiological epiphenomena of experience rather than active coping in terms of goal-oriented motor actions. It constitutes a means to parametrically modify stimuli that depict social situations and their broader environmental contexts. As an alternative to studying the variation of brain responses as a function of a priori fixed stimuli, this method can be applied to survey the range of stimuli that evoke similar responses across participants at particular brain regions of interest.},
author = {Tikka, Pia and V{\"{a}}ljam{\"{a}}e, Aleksander and de Borst, Aline W. and Pugliese, Roberto and Ravaja, Niklas and Kaipainen, Mauri and Takala, Tapio},
doi = {10.3389/fnhum.2012.00298},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Tikka et al. - 2012 - Enactive cinema paves way for understanding complex real-time social interaction in neuroimaging experiments.pdf:pdf},
issn = {1662-5161},
journal = {Frontiers in Human Neuroscience},
keywords = {Brain-Computer-Interfaces,enactive cinema,generative storytelling,implicit interaction,real-time fMRI,social neuroscience,two-way feedback},
month = {nov},
pages = {298},
publisher = {Frontiers},
title = {{Enactive cinema paves way for understanding complex real-time social interaction in neuroimaging experiments}},
volume = {6},
year = {2012}
}
@techreport{Larsson2016,
author = {Larsson and Linn{\'{e}}a},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Larsson, Linn{\'{e}}a - 2016 - P O B o x 1 1 7 2 2 1 0 0 L u n d 4 6 4 6-2 2 2 0 0 0 0 Event Detection in Eye-Tracking Data for Use in Applic.pdf:pdf},
title = {{P O B o x 1 1 7 2 2 1 0 0 L u n d + 4 6 4 6-2 2 2 0 0 0 0 Event Detection in Eye-Tracking Data for Use in Applications with Dynamic Stimuli}},
url = {http://portal.research.lu.se/portal/files/6192499/8600514.pdf},
year = {2016}
}
@article{Nystrom2010AnData,
title = {{An adaptive algorithm for fixation, saccade, and glissade detection in eyetracking data}},
year = {2010},
journal = {Behavior Research Methods},
author = {Nystr{\"{o}}m, Marcus and Holmqvist, Kenneth},
number = {1},
month = {2},
pages = {188--204},
volume = {42},
publisher = {Springer-Verlag},
doi = {10.3758/BRM.42.1.188},
issn = {1554-351X}
}
@Article{Stampe1993,
author="Stampe, Dave M.",
title="Heuristic filtering and reliable calibration methods for video-based pupil-tracking systems",
journal="Behavior Research Methods, Instruments, {\&} Computers",
year="1993",
month="Jun",
day="01",
volume="25",
number="2",
pages="137--142",
abstract="Methods for enhancing the accuracy of fixation and saccade detection and the reliability of calibration in video gaze-tracking systems are discussed. The unique aspects of the present approach include effective low-delay noise reduction prior to the detection of fixation changes, monitoring of gaze position in real time by the operator, identification of saccades as small as 0.5{\textdegree} while eliminating false fixations, and a quick, high-precision, semiautomated calibration procedure.",
issn="1532-5970",
doi="10.3758/BF03204486"
}
@article{dorr2010variability,
title={Variability of eye movements when viewing dynamic natural scenes},
author={Dorr, Michael and Martinetz, Thomas and Gegenfurtner, Karl R and Barth, Erhardt},
journal={Journal of vision},
volume={10},
number={10},
pages={28--28},
year={2010},
publisher={The Association for Research in Vision and Ophthalmology},
doi={10.1167/10.10.28}
}
@Article{Duchowski2002,
author="Duchowski, Andrew T.",
title="A breadth-first survey of eye-tracking applications",
journal="Behavior Research Methods, Instruments, {\&} Computers",
year="2002",
month="Nov",
day="01",
volume="34",
number="4",
pages="455--470",
abstract="Eye-tracking applications are surveyed in a breadth-first manner, reporting on work from the following domains: neuroscience, psychology, industrial engineering and human factors, marketing/advertising, and computer science. Following a review of traditionally diagnostic uses, emphasis is placed on interactive applications, differentiating between selective and gaze-contingent approaches.",
issn="1532-5970",
doi="10.3758/BF03195475"
}
@book{holmqvist2011eye,
title={Eye tracking: A comprehensive guide to methods and measures},
author={Holmqvist, Kenneth and Nystr{\"o}m, Marcus and Andersson, Richard and Dewhurst, Richard and Jarodzka, Halszka and Van de Weijer, Joost},
year={2011},
publisher={OUP Oxford}
}
@inproceedings{munn2008fixation,
title={Fixation-identification in dynamic scenes: Comparing an automated algorithm to manual coding},
author={Munn, Susan M and Stefano, Leanne and Pelz, Jeff B},
booktitle={Proceedings of the 5th symposium on Applied perception in graphics and visualization},
pages={33--42},
year={2008},
doi={10.1145/1394281.1394287},
organization={ACM}
}
@article{LARSSON2015145,
title = "Detection of fixations and smooth pursuit movements in high-speed eye-tracking data",
journal = "Biomedical Signal Processing and Control",
volume = "18",
pages = "145 - 152",
year = "2015",
issn = "1746-8094",
doi = "https://doi.org/10.1016/j.bspc.2014.12.008",
author = "Linnéa Larsson and Marcus Nystr{\"{o}}m and Richard Andersson and Martin Stridh",
keywords = "Signal processing, Eye-tracking, Smooth pursuit",
abstract = "A novel algorithm for the detection of fixations and smooth pursuit movements in high-speed eye-tracking data is proposed, which uses a three-stage procedure to divide the intersaccadic intervals into a sequence of fixation and smooth pursuit events. The first stage performs a preliminary segmentation while the latter two stages evaluate the characteristics of each such segment and reorganize the preliminary segments into fixations and smooth pursuit events. Five different performance measures are calculated to investigate different aspects of the algorithm's behavior. The algorithm is compared to the current state-of-the-art (I-VDT and the algorithm in [11]), as well as to annotations by two experts. The proposed algorithm performs considerably better (average Cohen's kappa 0.42) than the I-VDT algorithm (average Cohen's kappa 0.20) and the algorithm in [11] (average Cohen's kappa 0.16), when compared to the experts’ annotations."
}
@Article{Komogortsev2013,
author="Komogortsev, Oleg V.
and Karpov, Alex",
title="Automated classification and scoring of smooth pursuit eye movements in the presence of fixations and saccades",
journal="Behavior Research Methods",
year="2013",
month="Mar",
day="01",
volume="45",
number="1",
pages="203--215",
abstract="Ternary eye movement classification, which separates fixations, saccades, and smooth pursuit from the raw eye positional data, is extremely challenging. This article develops new and modifies existing eye-tracking algorithms for the purpose of conducting meaningful ternary classification. To this end, a set of qualitative and quantitative behavior scores is introduced to facilitate the assessment of classification performance and to provide means for automated threshold selection. Experimental evaluation of the proposed methods is conducted using eye movement records obtained from 11 subjects at 1000 Hz in response to a step-ramp stimulus eliciting fixations, saccades, and smooth pursuits. Results indicate that a simple hybrid method that incorporates velocity and dispersion thresholding allows producing robust classification performance. It is concluded that behavior scores are able to aid automated threshold selection for the algorithms capable of successful classification.",
issn="1554-3528",
doi="10.3758/s13428-012-0234-9"
}
@Article{Zemblys2018,
author="Zemblys, Raimondas
and Niehorster, Diederick C.
and Holmqvist, Kenneth",
title="gazeNet: End-to-end eye-movement event detection with deep neural networks",
journal="Behavior Research Methods",
year="2018",
month="Oct",
day="17",
abstract="Existing event detection algorithms for eye-movement data almost exclusively rely on thresholding one or more hand-crafted signal features, each computed from the stream of raw gaze data. Moreover, this thresholding is largely left for the end user. Here we present and develop gazeNet, a new framework for creating event detectors that do not require hand-crafted signal features or signal thresholding. It employs an end-to-end deep learning approach, which takes raw eye-tracking data as input and classifies it into fixations, saccades and post-saccadic oscillations. Our method thereby challenges an established tacit assumption that hand-crafted features are necessary in the design of event detection algorithms. The downside of the deep learning approach is that a large amount of training data is required. We therefore first develop a method to augment hand-coded data, so that we can strongly enlarge the data set used for training, minimizing the time spent on manual coding. Using this extended hand-coded data, we train a neural network that produces eye-movement event classification from raw eye-movement data without requiring any predefined feature extraction or post-processing steps. The resulting classification performance is at the level of expert human coders. Moreover, an evaluation of gazeNet on two other datasets showed that gazeNet generalized to data from different eye trackers and consistently outperformed several other event detection algorithms that we tested.",
issn="1554-3528",
doi="10.3758/s13428-018-1133-5"
}
@article{real_world,
author = {Matusz, Pawel J. and Dikker, Suzanne and Huth, Alexander G. and Perrodin, Catherine},
title = {Are We Ready for Real-world Neuroscience?},
journal = {Journal of Cognitive Neuroscience},
volume = {31},
number = {3},
pages = {327-338},
year = {2019},
doi = {10.1162/jocn\_e\_01276},
note ={PMID: 29916793},
abstract = { Real-world environments are typically dynamic, complex, and multisensory in nature and require the support of top–down attention and memory mechanisms for us to be able to drive a car, make a shopping list, or pour a cup of coffee. Fundamental principles of perception and functional brain organization have been established by research utilizing well-controlled but simplified paradigms with basic stimuli. The last 30 years ushered a revolution in computational power, brain mapping, and signal processing techniques. Drawing on those theoretical and methodological advances, over the years, research has departed more and more from traditional, rigorous, and well-understood paradigms to directly investigate cognitive functions and their underlying brain mechanisms in real-world environments. These investigations typically address the role of one or, more recently, multiple attributes of real-world environments. Fundamental assumptions about perception, attention, or brain functional organization have been challenged—by studies adapting the traditional paradigms to emulate, for example, the multisensory nature or varying relevance of stimulation or dynamically changing task demands. Here, we present the state of the field within the emerging heterogeneous domain of real-world neuroscience. To be precise, the aim of this Special Focus is to bring together a variety of the emerging “real-world neuroscientific” approaches. These approaches differ in their principal aims, assumptions, or even definitions of “real-world neuroscience” research. Here, we showcase the commonalities and distinctive features of the different “real-world neuroscience” approaches. To do so, four early-career researchers and the speakers of the Cognitive Neuroscience Society 2017 Meeting symposium under the same title answer questions pertaining to the added value of such approaches in bringing us closer to accurate models of functional brain organization and cognitive functions.}
}
@Article{Hooge2018,
author="Hooge, Ignace T. C.
and Niehorster, Diederick C.
and Nystr{\"o}m, Marcus
and Andersson, Richard
and Hessels, Roy S.",
title="Is human classification by experienced untrained observers a gold standard in fixation detection?",
journal="Behavior Research Methods",
year="2018",
month="Oct",
day="01",
volume="50",
number="5",
pages="1864--1881",
abstract="Manual classification is still a common method to evaluate event detection algorithms. The procedure is often as follows: Two or three human coders and the algorithm classify a significant quantity of data. In the gold standard approach, deviations from the human classifications are considered to be due to mistakes of the algorithm. However, little is known about human classification in eye tracking. To what extent do the classifications from a larger group of human coders agree? Twelve experienced but untrained human coders classified fixations in 6 min of adult and infant eye-tracking data. When using the sample-based Cohen's kappa, the classifications of the humans agreed near perfectly. However, we found substantial differences between the classifications when we examined fixation duration and number of fixations. We hypothesized that the human coders applied different (implicit) thresholds and selection rules. Indeed, when spatially close fixations were merged, most of the classification differences disappeared. On the basis of the nature of these intercoder differences, we concluded that fixation classification by experienced untrained human coders is not a gold standard. To bridge the gap between agreement measures (e.g., Cohen's kappa) and eye movement parameters (fixation duration, number of fixations), we suggest the use of the event-based F1 score and two new measures: the relative timing offset (RTO) and the relative timing deviation (RTD).",
issn="1554-3528",
doi="10.3758/s13428-017-0955-x"
}
@ARTICLE{5523936,
author={O. V. {Komogortsev} and D. V. {Gobert} and S. {Jayarathna} and D. H. {Koh} and S. M. {Gowda}},
journal={IEEE Transactions on Biomedical Engineering},
title={Standardization of Automated Analyses of Oculomotor Fixation and Saccadic Behaviors},
year={2010},
volume={57},
number={11},
pages={2635-2645},
keywords={biomechanics;biomedical optical imaging;eye;image classification;medical image processing;oculomotor fixation;saccadic behaviors;standardization;eye movement classification algorithms;stimulus-evoked task;threshold-value selection;Standardization;Classification algorithms;Computer science;Logic;Humans;Visual system;Psychology;Permission;Brain injuries;Alzheimer's disease;Analysis;baseline;eye-movement classification;oculomotor behavior;Adolescent;Adult;Algorithms;Female;Fixation, Ocular;Humans;Male;Saccades;Young Adult},
doi={10.1109/TBME.2010.2057429},
ISSN={0018-9294},
month={Nov}
}
@article{gorgolewski2016brain,
title={The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments},
author={Gorgolewski, Krzysztof J and Auer, Tibor and Calhoun, Vince D and Craddock, R Cameron and Das, Samir and Duff, Eugene P and Flandin, Guillaume and Ghosh, Satrajit S and Glatard, Tristan and Halchenko, Yaroslav O and others},
journal={Scientific Data},
volume={3},
pages={160044},
year={2016},
doi={10.1038/sdata.2016.44},
publisher={Nature Publishing Group}
}
@article{carl1987pursuits,
author = {Carl, J. R. and Gellman, R. S.},
title = {Human smooth pursuit: stimulus-dependent responses},
journal = {Journal of Neurophysiology},
volume = {57},
number = {5},
pages = {1446-1463},
year = {1987},
doi = {10.1152/jn.1987.57.5.1446},
note ={PMID: 3585475},
abstract = { We studied pursuit eye movements in seven normal human subjects with the scleral search-coil technique. The initial eye movements in response to unpredictable changes in target motion were analyzed to determine the effect of target velocity and position on the latency and acceleration of the response. By restricting our analysis to the presaccadic portion of the response we were able to eliminate any saccadic interactions, and the randomized stimulus presentation minimized anticipatory responses. This approach has allowed us to characterize a part of the smooth-pursuit system that is dependent primarily on retinal image properties. The latency of the smooth-pursuit response was very consistent, with a mean of 100 +/- 5 ms to targets moving 5 degrees/s or faster. The responses were the same whether the velocity step was presented when the target was initially stationary or after tracking was established. The latency did increase for lower velocity targets; this increase was well described by a latency model requiring a minimum target movement of 0.028 degrees, in addition to a fixed processing time of 98 ms. The presaccadic accelerations were fairly low, and increased with target velocity until an acceleration of about 50 degrees/s2 was reached for target velocities of 10 degrees/s. Higher velocities produced only a slight increase in eye acceleration. When the target motion was adjusted so that the retinal image slip occurred at increasing distances from the fovea, the accelerations declined until no presaccadic response was measurable when the image slip started 15 degrees from the fovea. The smooth-pursuit response to a step of target position was a brief acceleration; this response occurred even when an oppositely directed velocity stimulus was present. The latency of the pursuit response to such a step was also approximately 100 ms. This result seems consistent with the idea that sensory pathways act as a low-pass spatiotemporal filter of the retinal input, effectively converting position steps into briefly moving stimuli. There was a large asymmetry in the responses to position steps: the accelerations were much greater when the position step of the target was away from the direction of tracking, compared with steps in the direction of tracking. The asymmetry may be due to the addition of a fixed slowing of the eyes whenever the target image disappears from the foveal region. When saccades were delayed by step-ramp stimuli, eye accelerations increased markedly approximately 200 ms after stimulus onset.(ABSTRACT TRUNCATED AT 400 WORDS)}
}
@Article{Startsev2018,
author="Startsev, Mikhail
and Agtzidis, Ioannis
and Dorr, Michael",
title="1D CNN with BLSTM for automated classification of fixations, saccades, and smooth pursuits",
journal="Behavior Research Methods",
year="2018",
month="Nov",
day="08",
abstract="Deep learning approaches have achieved breakthrough performance in various domains. However, the segmentation of raw eye-movement data into discrete events is still done predominantly either by hand or by algorithms that use hand-picked parameters and thresholds. We propose and make publicly available a small 1D-CNN in conjunction with a bidirectional long short-term memory network that classifies gaze samples as fixations, saccades, smooth pursuit, or noise, simultaneously assigning labels in windows of up to 1 s. In addition to unprocessed gaze coordinates, our approach uses different combinations of the speed of gaze, its direction, and acceleration, all computed at different temporal scales, as input features. Its performance was evaluated on a large-scale hand-labeled ground truth data set (GazeCom) and against 12 reference algorithms. Furthermore, we introduced a novel pipeline and metric for event detection in eye-tracking recordings, which enforce stricter criteria on the algorithmically produced events in order to consider them as potentially correct detections. Results show that our deep approach outperforms all others, including the state-of-the-art multi-observer smooth pursuit detector. We additionally test our best model on an independent set of recordings, where our approach stays highly competitive compared to literature methods.",
issn="1554-3528",
doi="10.3758/s13428-018-1144-2"
}
@article{Schutz2011,
author = {Schutz, A. C. and Braun, D. I. and Gegenfurtner, K. R.},
doi = {10.1167/11.5.9},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Schutz, Braun, Gegenfurtner - 2011 - Eye movements and perception A selective review.pdf:pdf},
issn = {1534-7362},
journal = {Journal of Vision},
keywords = {eye,eye movement,motion,motion perception,noise,object recognition,perception,pursuit, smooth,saccades},
mendeley-groups = {EyeGaze},
month = {sep},
number = {5},
pages = {9--9},
publisher = {The Association for Research in Vision and Ophthalmology},
title = {{Eye movements and perception: A selective review}},
volume = {11},
year = {2011}
}
@article{cohen1960coefficient,
title={A coefficient of agreement for nominal scales},
author={Cohen, Jacob},
journal={Educational and psychological measurement},
volume={20},
number={1},
pages={37--46},
year={1960},
publisher={Sage Publications Sage CA: Thousand Oaks, CA}
}
@article{hessels2018eye,
title={Is the eye-movement field confused about fixations and saccades? A survey among 124 researchers},
author={Hessels, Roy S and Niehorster, Diederick C and Nystr{\"o}m, Marcus and Andersson, Richard and Hooge, Ignace TC},
journal={Royal Society open science},
volume={5},
number={8},
pages={180502},
year={2018},
publisher={The Royal Society}
}
@inproceedings{holmqvist2012eye,
title={Eye tracker data quality: what it is and how to measure it},
author={Holmqvist, Kenneth and Nystr{\"o}m, Marcus and Mulvey, Fiona},
booktitle={Proceedings of the symposium on eye tracking research and applications},
pages={45--52},
year={2012},
organization={ACM}
}
@article{maguire2012studying,
title={Studying the freely-behaving brain with fMRI},
author={Maguire, Eleanor A},
journal={Neuroimage},
volume={62},
number={2},
pages={1170--1176},
year={2012},
publisher={Elsevier}
}
@article{choe2016pupil,
title={Pupil size dynamics during fixation impact the accuracy and precision of video-based gaze estimation},
author={Choe, Kyoung Whan and Blake, Randolph and Lee, Sang-Hun},
journal={Vision research},
volume={118},
pages={48--59},
year={2016},
publisher={Elsevier}
}
@inproceedings{Mathe2012,
author = {Mathe, Stefan and Sminchisescu, Cristian},
title = {Dynamic Eye Movement Datasets and Learnt Saliency Models for Visual Action Recognition},
year = {2012},
isbn = {9783642337086},
publisher = {Springer-Verlag},
address = {Berlin, Heidelberg},
booktitle = {Proceedings, Part II, of the 12th European Conference on Computer Vision --- ECCV 2012 - Volume 7573},
pages = {842–856},
numpages = {15}
}
@article{Friedman2018,
author = {Friedman, Lee and Rigas, Ioannis and Abdulin, Evgeny and Komogortsev, Oleg V.},
doi = {10.3758/s13428-018-1050-7},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Friedman et al. - 2018 - A novel evaluation of two related and two independent algorithms for eye movement classification during read(2).pdf:pdf},
issn = {1554-3528},
journal = {Behavior Research Methods},
month = {aug},
number = {4},
pages = {1374--1397},
publisher = {Springer US},
title = {{A novel evaluation of two related and two independent algorithms for eye movement classification during reading}},
volume = {50},
year = {2018}
}
@article{Hanke2016,
abstract = {A {\textless}i{\textgreater}studyforrest{\textless}/i{\textgreater} extension, simultaneous fMRI and eye gaze recordings during prolonged natural stimulation},
author = {Hanke, Michael and Adelh{\"{o}}fer, Nico and Kottke, Daniel and Iacovella, Vittorio and Sengupta, Ayan and Kaule, Falko R. and Nigbur, Roland and Waite, Alexander Q. and Baumgartner, Florian and Stadler, J{\"{o}}rg},
doi = {10.1038/sdata.2016.92},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Hanke et al. - 2016 - A studyforrest extension, simultaneous fMRI and eye gaze recordings during prolonged natural stimulation.pdf:pdf},
issn = {2052-4463},
journal = {Scientific Data},
keywords = {Attention,Cortex,Language,Neural encoding,Visual system},
month = {oct},
pages = {160092},
publisher = {Nature Publishing Group},
title = {{A studyforrest extension, simultaneous fMRI and eye gaze recordings during prolonged natural stimulation}},
volume = {3},
year = {2016}
}
@article{HantaoLiu2011,
abstract = {Since the human visual system (HVS) is the ultimate assessor of image quality, current research on the design of objective image quality metrics tends to include an important feature of the HVS, namely, visual attention. Different metrics for image quality prediction have been extended with a computational model of visual attention, but the resulting gain in reliability of the metrics so far was variable. To better understand the basic added value of including visual attention in the design of objective metrics, we used measured data of visual attention. To this end, we performed two eye-tracking experiments: one with a free-looking task and one with a quality assessment task. In the first experiment, 20 observers looked freely to 29 unimpaired original images, yielding us so-called natural scene saliency (NSS). In the second experiment, 20 different observers assessed the quality of distorted versions of the original images. The resulting saliency maps showed some differences with the NSS, and therefore, we applied both types of saliency to four different objective metrics predicting the quality of JPEG compressed images. For both types of saliency the performance gain of the metrics improved, but to a larger extent when adding the NSS. As a consequence, we further integrated NSS in several state-of-the-art quality metrics, including three full-reference metrics and two no-reference metrics, and evaluated their prediction performance for a larger set of distortions. By doing so, we evaluated whether and to what extent the addition of NSS is beneficial to objective quality prediction in general terms. In addition, we address some practical issues in the design of an attention-based metric. The eye-tracking data are made available to the research community {\textless}citerefgrp{\textgreater}{\textless}citeref refid="ref1"/{\textgreater}{\textless}/citerefgrp{\textgreater}.},
author = {Liu, Hantao and Heynderickx, Ingrid},
doi = {10.1109/TCSVT.2011.2133770},
isbn = {1051-8215},
issn = {10518215},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
keywords = {Eye tracking,image quality assessment,objective metric,saliency map,visual attention},
month = {jul},
number = {7},
pages = {971--982},
title = {{Visual attention in objective image quality assessment: Based on eye-tracking data}},
volume = {21},
year = {2011}
}
@article{Larsson2013,
abstract = {A novel algorithm for detection of saccades and postsaccadic oscillations in the presence of smooth pursuit movements is proposed. The method combines saccade detection in the acceleration domain with specialized on- and offset criteria for saccades and postsaccadic oscillations. The performance of the algorithm is evaluated by comparing the detection results to those of an existing velocity-based adaptive algorithm and a manually annotated database. The results show that there is a good agreement between the events detected by the proposed algorithm and those in the annotated database with Cohen's kappa around 0.8 for both a development and a test database. In conclusion, the proposed algorithm accurately detects saccades and postsaccadic oscillations as well as intervals of disturbances.},
author = {Larsson, Linnea and Nystr{\"{o}}m, Marcus and Stridh, Martin},
doi = {10.1109/TBME.2013.2258918},
isbn = {1558-2531 (Electronic) 0018-9294 (Linking)},
issn = {15582531},
journal = {IEEE Transactions on Biomedical Engineering},
keywords = {Eye-tracking,signal processing,smooth pursuit},
month = {sep},
number = {9},
pages = {2484--2493},
pmid = {23625350},
title = {{Detection of saccades and postsaccadic oscillations in the presence of smooth pursuit}},
volume = {60},
year = {2013}
}
@article{Hanke2014,
abstract = {A high-resolution 7-Tesla fMRI dataset from complex natural stimulation with an audio movie},
author = {Hanke, Michael and Baumgartner, Florian J. and Ibe, Pierre and Kaule, Falko R. and Pollmann, Stefan and Speck, Oliver and Zinke, Wolf and Stadler, J{\"{o}}rg},
doi = {10.1038/sdata.2014.3},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Hanke et al. - 2014 - A high-resolution 7-Tesla fMRI dataset from complex natural stimulation with an audio movie.pdf:pdf},
issn = {2052-4463},
journal = {Scientific Data},
keywords = {Auditory system,Functional magnetic resonance imaging,Language,Perception},
month = {may},
pages = {140003},
publisher = {Nature Publishing Group},
title = {{A high-resolution 7-Tesla fMRI dataset from complex natural stimulation with an audio movie}},
volume = {1},
year = {2014}
}
@article{Harris2014,
abstract = {Face-selective regions in the amygdala and posterior superior temporal sulcus (pSTS) are strongly implicated in the processing of transient facial signals, such as expression. Here, we measured neural responses in participants while they viewed dynamic changes in facial expression. Our aim was to explore how facial expression is represented in different face-selective regions. Short movies were generated by morphing between faces posing a neutral expression and a prototypical expression of a basic emotion (either anger, disgust, fear, happiness or sadness). These dynamic stimuli were presented in block design in the following four stimulus conditions: (1) same-expression change, same-identity, (2) same-expression change, different-identity, (3) different-expression change, same-identity, and (4) different-expression change, different-identity. So, within a same-expression change condition the movies would show the same change in expression whereas in the different-expression change conditions each movie would have a different change in expression. Facial identity remained constant during each movie but in the different identity conditions the facial identity varied between each movie in a block. The amygdala, but not the posterior STS, demonstrated a greater response to blocks in which each movie morphed from neutral to a different emotion category compared to blocks in which each movie morphed to the same emotion category. Neural adaptation in the amygdala was not affected by changes in facial identity. These results are consistent with a role of the amygdala in category-based representation of facial expressions of emotion.},
author = {Harris, Richard J and Young, Andrew W and Andrews, Timothy J},
doi = {10.1016/j.neuropsychologia.2014.01.005},
file = {:C$\backslash$:/Users/Asim H. Dar/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Harris, Young, Andrews - 2014 - Dynamic stimuli demonstrate a categorical representation of facial expression in the amygdala.pdf:pdf},
issn = {1873-3514},
journal = {Neuropsychologia},
keywords = {Emotion,Expression,Face,fMRI},
month = {apr},
number = {100},
pages = {47--52},
pmid = {24447769},
publisher = {Elsevier},
title = {{Dynamic stimuli demonstrate a categorical representation of facial expression in the amygdala.}},
volume = {56},
year = {2014}
}
@Misc{JOP+2001,
author = {Eric Jones and Travis Oliphant and Pearu Peterson and others},
title = {{SciPy}: Open source scientific tools for {Python}},
year = {2001--},
url = "http://www.scipy.org"
}
@book{oliphant2006guide,
title={A guide to NumPy},
author={Oliphant, Travis E},
volume={1},
year={2006},
publisher={Trelgol Publishing USA}
}
@article{hunter2007matplotlib,
title={Matplotlib: A 2D graphics environment},
author={Hunter, John D},
journal={Computing in science \& engineering},
volume={9},
number={3},
pages={90--95},
year={2007},
publisher={IEEE},
doi={10.1109/MCSE.2007.55}
}
@inproceedings{mckinney2010data,
title={Data structures for statistical computing in python},
author={McKinney, Wes and others},
booktitle={Proceedings of the 9th Python in Science Conference},
volume={445},
pages={51--56},
year={2010},
organization={Austin, TX}
}
@inproceedings{seabold2010statsmodels,
title={Statsmodels: Econometric and statistical modeling with python},
author={Seabold, Skipper and Perktold, Josef},
booktitle={9th Python in Science Conference},
year={2010},
}
@Misc{HH+2013,
author = {Yaroslav O. Halchenko and Michael Hanke and others},
title = {{DataLad}: perpetual decentralized management of digital objects},
year = {2013--},
url = "http://datalad.org",
doi={10.5281/zenodo.1470735}
}
@article{hessels2017noise,
title={Noise-robust fixation detection in eye movement data: Identification by two-means clustering (I2MC)},
author={Hessels, Roy S and Niehorster, Diederick C and Kemner, Chantal and Hooge, Ignace TC},
journal={Behavior research methods},
volume={49},
number={5},
pages={1802--1823},
year={2017},
publisher={Springer}
}
@article{HOOGE20166,
title = "The pupil is faster than the corneal reflection (CR): Are video based pupil-CR eye trackers suitable for studying detailed dynamics of eye movements?",
journal = "Vision Research",
volume = "128",
pages = "6 - 18",
year = "2016",
issn = "0042-6989",
doi = "https://doi.org/10.1016/j.visres.2016.09.002",
url = "http://www.sciencedirect.com/science/article/pii/S0042698916301031",
author = "Ignace Hooge and Kenneth Holmqvist and Marcus Nyström",
keywords = "Saccades, Pupil, Corneal reflection",
abstract = "Most modern video eye trackers use the p-CR (pupil minus CR) technique to deal with small relative movements between the eye tracker camera and the eye. We question whether the p-CR technique is appropriate to investigate saccade dynamics. In two experiments we investigated the dynamics of pupil, CR and gaze signals obtained from a standard SMI Hi-Speed eye tracker. We found many differences between the pupil and the CR signals. Differences concern timing of the saccade onset, saccade peak velocity and post-saccadic oscillation (PSO). We also obtained that pupil peak velocities were higher than CR peak velocities. Saccades in the eye trackers’ gaze signal (that is constructed from p-CR) appear to be excessive versions of saccades in the pupil signal. We conclude that the pupil-CR technique is not suitable for studying detailed dynamics of eye movements."
}
@article{dalveren2019evaluation,
title={Evaluation of Ten Open-Source Eye-Movement Classification Algorithms in Simulated Surgical Scenarios},
author={Dalveren, Gonca Gokce Menekse and Cagiltay, Nergiz Ercil},
journal={IEEE Access},
volume={7},
pages={161794--161804},
year={2019},
publisher={IEEE}
}
@article{van2018gazepath,
title={Gazepath: An eye-tracking analysis tool that accounts for individual differences and data quality},
author={van Renswoude, Daan R and Raijmakers, Maartje EJ and Koornneef, Arnout and Johnson, Scott P and Hunnius, Sabine and Visser, Ingmar},
journal={Behavior research methods},
volume={50},
number={2},
pages={834--852},
year={2018},
publisher={Springer}
}

102
results_def.tex
View file

@ -519,3 +519,105 @@
\newcommand{\rankPURvideoMN}{1}
\newcommand{\rankPURvideoRA}{0}
\newcommand{\rankPURvideoRE}{2}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: TH34_img_vy_labelled_{}.mat
% #
% Truncate labels to shorter sample: 4988
\newcommand{\kappaRAMNimgFix}{0.84}
\newcommand{\kappaALRAimgFix}{0.55}
\newcommand{\kappaALMNimgFix}{0.52}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: TH34_img_vy_labelled_{}.mat
% #
% Truncate labels to shorter sample: 4988
\newcommand{\kappaRAMNimgSac}{0.91}
\newcommand{\kappaALRAimgSac}{0.78}
\newcommand{\kappaALMNimgSac}{0.78}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: TH34_img_vy_labelled_{}.mat
% #
% Truncate labels to shorter sample: 4988
\newcommand{\kappaRAMNimgPSO}{0.76}
\newcommand{\kappaALRAimgPSO}{0.59}
\newcommand{\kappaALMNimgPSO}{0.58}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL27_trial17_labelled_{}.mat
% #
% Truncate labels to shorter sample: 454
\newcommand{\kappaRAMNdotsFix}{0.65}
\newcommand{\kappaALRAdotsFix}{0.37}
\newcommand{\kappaALMNdotsFix}{0.45}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL27_trial17_labelled_{}.mat
% #
% Truncate labels to shorter sample: 454
\newcommand{\kappaRAMNdotsSac}{0.81}
\newcommand{\kappaALRAdotsSac}{0.72}
\newcommand{\kappaALMNdotsSac}{0.78}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL27_trial17_labelled_{}.mat
% #
% Truncate labels to shorter sample: 454
\newcommand{\kappaRAMNdotsPSO}{0.62}
\newcommand{\kappaALRAdotsPSO}{0.38}
\newcommand{\kappaALMNdotsPSO}{0.41}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: TH38_video_dolphin_fov_labelled_{}.mat
% #
% Truncate labels to shorter sample: 4044
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL23_video_triple_jump_labelled_{}.mat
% #
% Truncate labels to shorter sample: 2821
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL27_video_triple_jump_labelled_{}.mat
% #
% Truncate labels to shorter sample: 2822
\newcommand{\kappaRAMNvideoFix}{0.65}
\newcommand{\kappaALRAvideoFix}{0.44}
\newcommand{\kappaALMNvideoFix}{0.39}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: TH38_video_dolphin_fov_labelled_{}.mat
% #
% Truncate labels to shorter sample: 4044
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL23_video_triple_jump_labelled_{}.mat
% #
% Truncate labels to shorter sample: 2821
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL27_video_triple_jump_labelled_{}.mat
% #
% Truncate labels to shorter sample: 2822
\newcommand{\kappaRAMNvideoSac}{0.87}
\newcommand{\kappaALRAvideoSac}{0.76}
\newcommand{\kappaALMNvideoSac}{0.79}
% #
% # %INCONSISTENCY Found label length mismatch between coders for: TH38_video_dolphin_fov_labelled_{}.mat
% #
% Truncate labels to shorter sample: 4044
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL23_video_triple_jump_labelled_{}.mat
% #
% Truncate labels to shorter sample: 2821
% #
% # %INCONSISTENCY Found label length mismatch between coders for: UL27_video_triple_jump_labelled_{}.mat
% #
% Truncate labels to shorter sample: 2822
\newcommand{\kappaRAMNvideoPSO}{0.65}
\newcommand{\kappaALRAvideoPSO}{0.45}
\newcommand{\kappaALMNvideoPSO}{0.51}

51
tools.bib
View file

@ -1,51 +0,0 @@
@Misc{JOP+2001,
author = {Eric Jones and Travis Oliphant and Pearu Peterson and others},
title = {{SciPy}: Open source scientific tools for {Python}},
year = {2001--},
url = "http://www.scipy.org"
}
@book{oliphant2006guide,
title={A guide to NumPy},
author={Oliphant, Travis E},
volume={1},
year={2006},
publisher={Trelgol Publishing USA}
}
@article{hunter2007matplotlib,
title={Matplotlib: A 2D graphics environment},
author={Hunter, John D},
journal={Computing in science \& engineering},
volume={9},
number={3},
pages={90--95},
year={2007},
publisher={IEEE},
doi={10.1109/MCSE.2007.55}
}
@inproceedings{mckinney2010data,
title={Data structures for statistical computing in python},
author={McKinney, Wes and others},
booktitle={Proceedings of the 9th Python in Science Conference},
volume={445},
pages={51--56},
year={2010},
organization={Austin, TX}
}
@inproceedings{seabold2010statsmodels,
title={Statsmodels: Econometric and statistical modeling with python},
author={Seabold, Skipper and Perktold, Josef},
booktitle={9th Python in Science Conference},
year={2010},
}
@Misc{HH+2013,
author = {Yaroslav O. Halchenko and Michael Hanke and others},
title = {{DataLad}: perpetual decentralized management of digital objects},
year = {2013--},
url = "http://datalad.org",
doi={10.5281/zenodo.1470735}
}