Research on the Auxiliary Diagnosis Method of Parkinson's Disease Based on FMRI

With the accelerating global aging population, the incidence of Parkinson's disease (PD) has been increasing annually, verwhelming families and society. However, traditional clinical diagnostic methods suffer from high subjectivity and elevated misdiagnosis rates. To address these challenges in PD diagnosis, this study proposes an enhanced Spatio-Temporal Linear Attention Transformer (STLAT) deep learning model based on the classical Transformer encoder. By extracting temporal and spatial features from resting-state functional magnetic resonance imaging (rs-fMRI), we introduce a multi-scale attention mechanism and an explainability quantification method to achieve accurate classification of PD patients and visualized diagnostic decision-making. Our approach constructs a standardized preprocessing pipeline to extract spatiotemporal features of brain regions and innovatively designs a spatiotemporal cross-attention module to capture dynamic functional connectivity patterns. Furthermore, we incorporate the integrated gradients saliency quantification (IG-SQ) method to interpret the decision rationale of the neural network. Experimentally, the method is proved to be correct up to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{9 8. 6 0 \%}$</tex> in PD binary classification. Through explainability analysis, we successfully localize key pathological regions, including the basal ganglia and the default mode network (DMN). This study provides an explainable neuroimaging biomarker analysis framework for objective PD diagnosis and pathological mechanism research.