Using fMRI to examine nonlinear mixed selectivity tuning to task and category in the human brain

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Neuroimaging

Background:

• Nonlinear mixed selectivity in neurons allows primate brains to adapt representations to changing task contexts.
• Previous studies relied on invasive neural recordings, limiting human brain research.
• Understanding this neural computation is key to deciphering cognitive flexibility.

Purpose of the Study:

• To demonstrate noninvasive measurement of nonlinear mixed selectivity tuning in the human brain using fMRI pattern decoding.
• To investigate the joint representation of object category and task information in human visual streams.
• To determine if task modulations occur when visual input is constant, varying only cognitive operations.

Main Methods:

• fMRI pattern decoding was employed to measure neural representations.
• Participants performed oddball detection and one-back repetition tasks on identical visual stimuli.
• Early, ventral, and dorsal visual stream areas were analyzed for joint category-task tuning.

Main Results:

• Moderate but significant evidence for nonlinear mixed selectivity to object category and task was found in fMRI patterns.
• This tuning was detected in both human ventral and dorsal visual areas.
• Ventral stream coding involved representational geometry shifts, preserving content, while dorsal stream coding reshaped geometry, altering content.

Conclusions:

• Nonlinear mixed selectivity neurons exist and cluster at an fMRI-visible scale in human ventral and dorsal streams.
• Task-dependent representational changes in the ventral stream preserve category information, while dorsal stream changes alter it.
• This study validates fMRI pattern decoding for studying complex neural computations noninvasively.