Precision Lesion Profiling in Multiple Sclerosis: A Novel Pipeline for EDSS Prediction

Multiple Sclerosis (MS) diagnosis and progression prediction remain challeng- ing due to the scarcity of labeled medical data. This paper presents a robust pipeline that effectively extracts meaningful lesion features from MRI scans and employs machine learning (ML) techniques to predict the Expanded Disability Status Scale (EDSS) scores, even with limited data availability. By leverag- ing advanced lesion detection, volume estimation, and distribution analysis, the proposed pipeline achieves reliable results despite the inherent challenges of small, imbalanced datasets. Various ML models, including Support Vector Machines (SVM), Decision Trees, Multilayer Perceptron (MLP), and Boosting algorithms (Random Forest, XGBoost, AdaBoost), were assessed. To mitigate label imbalance, SMOTE and semi-supervised learning techniques were incor- porated, further improving model stability. Notably, a simple MLP achieved an accuracy of 87% on a dataset with only 60 patients, demonstrating the effec- tiveness of our pipeline in low-data scenarios and its ability to generalize well with limited labeled medical data showing its potential for real-world clinical applications.