Eye Movement-Related Confounds in Neural Decoding of Visual Working Memory Representations

Area of Science:

• Cognitive Neuroscience
• Neuroimaging Analysis

Background:

• Neural decoding, including multivariate pattern analysis (MVPA), is increasingly used in cognitive neuroimaging.
• These methods aim to decode neural signals to understand representations in brain activity.
• Visual working memory (VWM) has been a key area benefiting from these techniques.

Purpose of the Study:

• To investigate neural representations in visual working memory (VWM) using multivariate decoding.
• To assess the temporal dynamics of neural representations during a VWM/imagery task.
• To identify and address potential confounds in neural decoding analyses.

Main Methods:

• Magnetoencephalography (MEG) was used to record neural signals from human volunteers.
• Multivariate decoding analyses were applied to the MEG data.
• Gaze position analysis was performed to detect eye movements.
• A functional localizer targeting bottom-up sensory signals was developed as a potential remedy.

Main Results:

• Initial decoding analyses suggested insights into VWM representations.
• Gaze analysis revealed that eye movements systematically contaminated the MEG decoding results.
• Original analyses, without accounting for eye movements, could lead to invalid conclusions.
• A novel decoding approach using a functional localizer successfully avoided eye movement confounds.

Conclusions:

• Eye movements represent a pervasive confound in neural decoding studies, particularly in VWM research.
• Researchers must be aware of and actively mitigate eye movement artifacts for valid neuroimaging results.
• Employing specific functional localizers can provide a more reliable method for decoding neural representations.