Decoding Visual fMRI Stimuli from Human Brain Based on Graph Convolutional Neural Network.

Area of Science:

• Neuroscience
• Computer Science
• Machine Learning

Background:

• Brain decoding aims to predict external stimuli from brain activity, with visual information being a key focus.
• Understanding visual cortex mechanisms relies on decoding functional magnetic resonance imaging (fMRI) data.
• Existing brain decoding algorithms struggle with accurately extracting stimulus features from fMRI.

Purpose of the Study:

• To propose an advanced brain decoding algorithm utilizing a graph convolution network (GCN) for improved fMRI analysis.
• To enhance the extraction of functional correlation features within the brain's visual regions.
• To overcome limitations of traditional methods in decoding visual stimuli from brain responses.

Main Methods:

• Selection of 11 key regions of interest (ROIs) within human visual cortex to minimize noise.
• Utilizing a deep 3D convolutional neural network for feature extraction from selected ROIs.
• Employing GCN with residual connections to capture inter-regional functional correlations and prevent gradient disappearance.

Main Results:

• The proposed GCN-based algorithm achieved a high recognition accuracy of 98.67% on a public dataset.
• Demonstrated superior performance compared to existing state-of-the-art brain decoding algorithms.
• Residual connections effectively integrated multi-level features, improving overall accuracy.

Conclusions:

• The novel GCN algorithm significantly improves the accuracy of brain decoding for visual stimuli from fMRI data.
• This approach offers a more effective method for understanding brain visual function mechanisms.
• The algorithm shows promise for future applications in neuroscience research and brain-computer interfaces.