Decoding eye-movement confounds from EPI time-series

Hey all,

If you are worried that your decoding accuracy results between 2 conditions might have been confounded by the eye-movement but you have no in-scanner eye-tracking data.

How much can you use the EPI signal change from the eyes themselves to rule out this possibility?

Only found this rough univariate approach but I wonder if anything multivariate has been tried.


1 Like

Here is a review of literature on the subject with some methods to track eye position directly from epi series:


LaConte, S. M.; Peltier, S. J.; Heberlein, K. A.; Hu, X. P. Predictive Eye Estimation Regression (PEER) for Simultaneous Eye Tracking and FMRI. 1.


Laconte, S.; Glielmi, C.; Heberlein, K.; Hu, X. Verifying Visual Fixation to Improve FMRI with Predictive Eye Estimation Regression (PEER). 1.




Son, J.; Ai, L.; Xu, T.; Colcombe, S.; LaConte, S.; Lisinski, J.; Klein, A.; Craddock, C.; Milham, M. Evaluating FMRI-Based Estimation of Eye Movements during Naturalistic Viewing. bioRxiv 2018 .


McNabb, C. B.; Lindner, M.; Shen, S.; Murayama, K.; Burgess, L. G.; Johnstone, T. Inter-Slice Leakage and Intra-Slice Aliasing in Simultaneous Multi-Slice Echo-Planar Images. bioRxiv 2018 .


Brodoehl, S.; Witte, O. W.; Klingner, C. M. Measuring Eye States in Functional MRI. BMC Neuroscience 2016 , 17 (1).


Fanea, L.; Fagan, A. J. Review: Magnetic Resonance Imaging Techniques in Ophthalmology. Molecular Vision 2012 , 23.


Zhang, X.; Ross, T. J.; Jo Salmeron, B.; Yang, S.; Yang, Y.; Stein, E. A. Single Subject Task-Related BOLD Signal Artifact in a Real-Time FMRI Feedback Paradigm. Human Brain Mapping 2011 , 32 (4), 592–600.


Sathian, K.; Lacey, S.; Stilla, R.; Gibson, G. O.; Deshpande, G.; Hu, X.; LaConte, S.; Glielmi, C. Dual Pathways for Haptic and Visual Perception of Spatial and Texture Information. NeuroImage 2011 , 57 (2), 462–475.


Keck, I. R.; Fischer, V.; Puntonet, C. G.; Lang, E. W. Eye Movement Quantification in Functional MRI Data by Spatial Independent Component Analysis. In Independent Component Analysis and Signal Separation ; Adali, T., Jutten, C., Romano, J. M. T., Barros, A. K., Eds.; Springer Berlin Heidelberg: Berlin, Heidelberg, 2009; Vol. 5441, pp 435–442.


Beauchamp, M. S. Detection of Eye Movements from FMRI Data. Magnetic Resonance in Medicine 2003 , 49 (2), 376–380.


Tregellas, J. R.; Tanabe, J. L.; Miller, D. E.; Freedman, R. Monitoring Eye Movements during FMRI Tasks with Echo Planar Images. Human Brain Mapping 2002 , 17 (4), 237–243.


Porter, D. A.; Calamante, F.; Gadian, D. G.; Connelly, A. The Effect of Residual Nyquist Ghost in Quantitative Echo-Planar Diffusion Imaging. Magnetic Resonance in Medicine 1999 , 42 (2), 385–392.<385::AID-MRM21>3.0.CO;2-J.


Chen, W.; Zhu, X.-H. Suppression of Physiological Eye Movement Artifacts in Functional MRI Using Slab Presaturation. Magnetic Resonance in Medicine 1997 , 38 (4), 546–550.


Awesome! I admit I had no idea where to even start looking. Very helpful! Thanks!

@Mohamed_Rezk you are seeing this?

@Remi-Gau : thanks :slight_smile: I am now.

Hi all,

Had no idea that so much work has been done on this!

However, I remain somewhat unsure as to whether you will be able to convince reviewers that using theses approaches really eliminates the possibility that what you are seeing is not due to eye movements (after all we have in-scanner eye trackers for a reason). Whether you can make it convincing depends on a lot of things, like the task (are eye movements critical, e.g. like in a visual search), the anatomical area (is it an area known to encode eye movements ), the magnitude of the effect, etc.

One possibility may be to run a control with eye tracking outside of the scanner. That may be more time and cost effective.

Just my 2 cents.

1 Like

Yeah it is by no means a silver bullet I agree and if you can have in-scanner eye tracking data it would be best (thought they are a pain to set up!).

As usual if you can decode between conditions from the eye-data on your EPIs, it might likely mean that you have a problem ; but if you can’t decode it does not necessarily mean that everything is peachy.

The only “advantage” it might have is that your is more “matched” in terms of sampling rate, SNR, movement confounds… that your eye tracking data. But I might be missing something.