Decoding Pixel-Level Image Features From Two-Photon Calcium Signals of Macaque Visual Cortex.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Computer Vision

Background:

• Visual scenes contain complex features crucial for brain cognition.
• Predicting neural responses from stimulus images is a key research area.
• Extracting features from neuronal responses remains a challenge.

Purpose of the Study:

• To develop a method for extracting features from neuronal responses.
• To investigate how the brain processes artificial patterns versus natural images.
• To understand neural encoding principles through decoding strategies.

Main Methods:

• Recorded two-photon calcium neural data from the visual cortex of awake macaque monkeys.
• Utilized stimuli including artificial patterns and natural images.
• Employed a deep neural network decoder inspired by image segmentation.

Main Results:

• Decoding natural images relied on a few strongly responding neurons, consistent with sparse coding.
• Decoding artificial patterns required a larger number of neurons.
• Pretraining the model on artificial patterns enabled extraction of salient natural scene features and category information from natural images.

Conclusions:

• The study provides a novel approach to understanding neural encoding by reverse-engineering decoding strategies.
• Distinct neural processing mechanisms exist for artificial patterns and natural images.
• Deep learning models can effectively decode neural activity and extract meaningful visual features.