Shape Selectivity of Middle Superior Temporal Sulcus Body Patch Neurons

Area of Science:

• Neuroscience
• Computational Neuroscience
• Primate Vision

Background:

• Functional MRI studies reveal specific brain regions in primates activated by visual body images.
• The middle superior temporal sulcus body (MSB) patch in macaques contains neurons selective for body stimuli.
• Previous research indicates shape is a significant factor in MSB neuron responses.

Purpose of the Study:

• To investigate and model the precise shape selectivity of single MSB neurons.
• To determine the extent to which shape, rather than semantic meaning, drives MSB neuron responses.

Main Methods:

• Recorded responses of single MSB neurons to diverse shapes.
• Employed an adaptive stimulus sampling procedure to refine shape stimuli.
• Fitted neuron responses using models based on shape parameters, pixel data, and convolutional neural networks (CNNs).

Main Results:

• Deep convolutional layers of CNNs provided the best fit for MSB neuron shape selectivity, explaining a median of 77% of the variance.
• Goodness-of-fit improved with deeper CNN layers but decreased in fully connected layers.
• Human ratings indicated that while many top shapes were animal-like, they didn't always resemble a body.

Conclusions:

• Deep CNNs successfully model single-unit shape selectivity in MSB neurons.
• Shape, particularly contour features, is a primary driver of MSB neuron responses.
• Semantic or category knowledge plays a minor role in the shape selectivity of individual MSB neurons.