s-TBN: A New Neural Decoding Model to Identify Stimulus Categories From Brain Activity Patterns.

Area of Science:

• Neurocomputing Science
• Neuroimaging Analysis
• Machine Learning

Background:

• Brain network patterns contain spatiotemporal information crucial for understanding brain activation.
• Traditional machine learning methods struggle to extract multidimensional structural information from brain networks.
• Tensor decomposition offers a powerful approach to mine unique spatiotemporal characteristics from complex data.

Purpose of the Study:

• To propose a novel stimulus-constrained Tensor Brain Network (s-TBN) model for neural decoding.
• To enhance the extraction of multidimensional structural information from brain networks.
• To improve the accuracy of decoding external stimuli from neuroimaging data.

Main Methods:

• Developed a stimulus-constrained Tensor Brain Network (s-TBN) model incorporating tensor decomposition.
• Integrated stimulus category-constraint information into the model.
• Validated the s-TBN model on magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) datasets.

Main Results:

• The s-TBN model achieved accuracy improvements of over 11.06% and 18.46% compared to existing methods on two different neuroimaging datasets.
• Demonstrated superior performance in extracting discriminative features from brain network data.
• Showcased particular effectiveness in decoding object stimuli with semantic information.

Conclusions:

• The proposed s-TBN model significantly outperforms traditional methods in neural decoding.
• Tensor decomposition combined with stimulus constraints is highly effective for extracting complex brain network features.
• This approach offers a promising advancement for neurocomputing and understanding brain responses to stimuli.