Decoupling of BOLD amplitude and pattern classification of orientation-selective activity in human visual cortex

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Neuroimaging

Background:

• Multivariate pattern analysis (MVPA) decodes neural representations from distributed fMRI activity patterns.
• Previous studies show reliable MVPA decoding in early visual cortex for stimuli in working memory or internally generated, despite low BOLD signals.

Purpose of the Study:

• Investigate how reliable stimulus information is decoded from early visual cortex when overall BOLD signal is low.
• Examine the relationship between univariate BOLD response differences and multivariate decoding accuracy.

Main Methods:

• Reanalyzed fMRI data from 24 participants imagining or maintaining a visual grating in working memory.
• Divided V1, V2, and V3 voxels by orientation preference.
• Compared time courses of mean BOLD responses to preferred/non-preferred orientations with multivariate decoding performance.

Main Results:

• Decoding accuracy correlated with small, reliable univariate differences in BOLD responses to preferred vs. non-preferred stimuli.
• The time course of BOLD differences mirrored multivariate classification accuracy.
• Classification reliability strongly correlated with the magnitude of these BOLD signal differences.

Conclusions:

• Small stimulus-specific BOLD responses, alongside large non-stimulus-specific task-related activity, enable decoding.
• Task-related BOLD signals in early visual cortex are an amalgam of specific and non-specific responses.
• Non-specific responses, not evident during internal generation, explain decoding from low overall BOLD signals.