No evidence for confounding orientation-dependent fixational eye movements under baseline conditions

Marcel van Gerven,Rob van Lier,Jordy Thielen

doi:10.1038/s41598-018-30221-2

Abstract

Decoding has become a standard analysis technique for contemporary cognitive neuroscience. Already more than a decade ago, it was shown that orientation information could be decoded from functional magnetic resonance imaging voxel time series. However, the underlying neural mechanism driving the decodable information is still under debate. Here, we investigated whether eye movements and pupil dilation during attempted fixation and passive viewing of visually presented square-wave grating stimuli could explain orientation decoding. We hypothesized that there are confounding orientation-dependent fixational eye movements (e.g., microsaccades), which systematically alter brain activity, and hence can be the source of decodable information. We repeated one of the original orientation decoding studies, but recorded eye movements instead of brain activity. We found no evidence that stimulus orientation can be decoded from eye movements under baseline conditions, but cannot rule out the potential confounding effect of eye movements under different conditions. With this study, we emphasize the importance, and show the implications of such potential confounding eye movements for decoding studies and cognitive neuroscience in general.

Highlights

Nowadays, multivariate pattern analysis (MVPA) is one of the standard methods to study brain function[1]
One of the first applications of MVPA involved the decoding of perceived stimulus orientation from patterns of blood oxygenation level dependent (BOLD) activation recorded with functional magnetic resonance imaging
When the direction of eye movements is tightly coupled to which stimulus is presented, brain activity is evoked that is specific to the presented stimulus, which in turn can be used by a decoder to classify the stimulus

Summary

Introduction

Multivariate pattern analysis (MVPA) is one of the standard methods to study brain function[1]. The population of neurons contained within one voxel could represent the full range of possible orientations uniformly This renders it unlikely that orientation tuning is a source of decodable information in these orientation decoding studies. It is postulated that weak biases emerge because of the imperfect alignment of voxels with the columnar organization of visual cortex, which could give rise to an orientation preference at the voxel level and support orientation decoding These small biases might be too weak to be picked up by univariate analyses, but MVPA can combine many voxels and thereby create a reliable classifier. It is important to note that the source of decodable information in this case would have nothing to do with a sensitivity to the presented stimulus (e.g., orientation selectivity), but instead can be completely explained by systematic eye movements

Methods

Results

Conclusion