SUFT: Sparse and Uncertain Fusion Transformers for Multi-Atlas Brain Network Analysis

Abstract

The existing multi-atlas brain network analysis methods rely on some simple fusion methods (i.e., add and concatenation) and do not consider the information redundancy caused by increased brain regions. To improve upon these, we propose the Sparse and Uncertain Fusion Transformers (SUFT) for multi-atlas brain network analysis. First, multi-atlas brain network data is derived from original fMRI data using various atlases. An attention enhancement module then extracts and preserves essential features for minimal information loss during selection. These processed features are fed into a selection module that identifies disease-related brain regions based on their importance scores. The model utilizes only these sparse features for further processing. Finally, an uncertain fusion module assesses the confidence level of each atlas and implements a strategy to integrate results at the evidence level. Experimental results on 1138 subjects from the SRPBS dataset demonstrate that our SUFT outperforms several state-of-the-art methods in identifying brain disorders.