Multiple Sclerosis Severity Classification in MRI using SCAN-ExOrU-Net and Siren-CNN

Abstract

Multiple Sclerosis (MS) is a progressive neurological disorder that leads to significant structural and functional changes in the brain, impacting both white and gray matter. Magnetic Resonance Imaging (MRI) plays a critical role in MS diagnosis and monitoring, yet existing lesion-based classification methods often fail to capture broader pathological changes beyond visible lesions. This research proposes an advanced severity classification framework leveraging multimodal MRI data (T1, T2, and FLAIR) and integrating both lesion-specific and non-lesion-specific attributes. The proposed approach employs a novel Spatial-Channel Attention Networks Exponential Orthogonal U-Net (SCAN-ExOrU-Net) for precise lesion segmentation and Siren-CNN for severity classification. Additionally, a Multiscale K-Co-occurrence Clustering (MK-CoC) method is introduced for tissue grouping, while Dynamic Causal Modeling (DCM) generates connectivity matrices to analyze brain network alterations. Feature extraction combines morphological, textural, and connectivity-based attributes, and severity classification is enhanced using Dynamic Fuzzy Log Rule Prioritized Logic (DFLRPL). Experimental validation using publicly available MRI datasets demonstrates the proposed framework's superior accuracy in classifying MS severity compared to existing methodologies. This research provides a comprehensive and robust MS severity assessment model, potentially improving clinical decision-making and patient management.