Brain Network State Transformer: Leveraging State Functional Connectivity for Enhanced Brain Network Analysis.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Functional Magnetic Resonance Imaging (fMRI) research increasingly focuses on dynamic functional connectivity (DFC) over static approaches.
• Existing DFC methods face challenges in balancing temporal resolution with the interpretability of brain network patterns.

Purpose of the Study:

• To introduce the Brain Network State Transformer (BNST), a novel framework for enhanced brain network analysis using fMRI.
• To improve the interpretability and predictive power of DFC analysis by identifying and modeling distinct brain states.

Main Methods:

• The BNST framework employs Deep Clustering to identify recurring brain states from DFC matrices.
• It utilizes State-Based Rechunking to reorganize BOLD time series based on identified states.
• A Transformer-Based Feature Extraction mechanism models intra-state and inter-state relationships for prediction tasks.

Main Results:

• BNST demonstrated effectiveness on ABCD and HCP fMRI datasets for both classification and regression tasks.
• The framework successfully captured structured temporal dynamics in brain activity.
• BNST improved prediction performance compared to existing DFC approaches.

Conclusions:

• BNST offers a structured representation of brain activity by identifying distinct brain states and their functional significance.
• This approach enhances the interpretability of DFC, aligning network dynamics with cognitive and neural processes.
• BNST represents a significant advancement in analyzing dynamic brain connectivity for neuroimaging research.