Atlas-free Brain Network Transformer

Researchers have developed an atlas-free Brain Network Transformer (BNT) that uses individualized brain parcellations from subject-specific fMRI data instead of standardized brain atlases. The approach outperformed existing methods in sex classification and brain age prediction tasks, offering improved precision and robustness for neuroimaging biomarkers and clinical diagnostics.

• →Atlas-free BNT eliminates limitations of fixed brain atlases including spatial misalignment and functional heterogeneity.
• →The method uses ROI-to-voxel connectivity features in standardized voxel-based feature space for individualized analysis.
• →Experimental results show consistent outperformance over state-of-the-art methods including elastic net, BrainGNN, and Graphormer.
• →The approach demonstrates significant improvements in precision, robustness, and generalizability of brain network analyses.
• →Applications include enhanced neuroimaging biomarkers and clinical diagnostic tools for personalized precision medicine.