Sub-Connection Learning for fMRI-Based Brain Functional Network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Brain functional network analysis uses functional magnetic resonance imaging (fMRI) to model brain connectivity.
• Current AI methods often analyze global connectivity, potentially missing localized disorder-specific patterns.
• Neurological disorders often impact specific, localized functional connections within the brain.

Purpose of the Study:

• To develop a novel sub-connection learning method for more accurate neurological disorder diagnosis.
• To identify diagnostically specific brain connections while suppressing irrelevant ones.
• To improve upon existing AI-based brain functional network analysis techniques.

Main Methods:

• A dynamic functional connectivity construction strategy was used to create functional connectivity matrices.
• A sub-connection Mask Learning strategy with a multi-head self-attention mechanism was designed to learn adaptive connection masks.
• Multi-mask Fusion and Mask Iterative Optimization strategies were employed to refine mask quality.

Main Results:

• The proposed method achieved high diagnostic accuracies of 72.30% on the ABIDE I dataset and 80.99% on the ADNI dataset.
• The model successfully identified diagnostically specific connections and suppressed noise connections.
• Experimental results demonstrated superior performance compared to state-of-the-art methods.

Conclusions:

• The novel sub-connection learning method effectively enhances diagnostic accuracy for neurological disorders.
• Focusing on specific, localized brain connections improves upon global connectivity analysis.
• The AI-driven approach shows significant potential for clinical applications in diagnosing brain disorders.