Predictors of short-term, relapse-independent progression in multiple sclerosis: A machine learning approach based on clinical data and conventional MRI-derived features.
Authors
Affiliations (10)
Affiliations (10)
- Department of Human Neurosciences, Sapienza University of Rome, Italy; Multiple Sclerosis Center, San Pietro Fatebenetratelli, Rome, Italy.
- Department of Human Neurosciences, Sapienza University of Rome, Italy.
- Department of Neurosciences and Reproductive and Odontostomatological Sciences, University "Federico II", 80131 Naples, Italy.
- Department of Advanced Biomedical Sciences, University "Federico II", Naples, Italy; Department of Radiology and Nuclear Medicine, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands.
- Department of Medicine, Surgery and Dentistry, University of Salerno, Baronissi, Italy.
- Department of Neurosciences, San Camillo-Forlanini Hospital, Rome, Italy.
- Department of Advanced Biomedical Sciences, University "Federico II", Naples, Italy; Department of Radiology and Nuclear Medicine, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands; Queen Square Multiple Sclerosis Centre, Department of Neuroinflammation, UCL Queen Square Institute of Neurology, University College London, London, UK.
- Department of Advanced Biomedical Sciences, University "Federico II", Naples, Italy.
- Department of Human Neurosciences, Sapienza University of Rome, Italy; IRCCS Neuromed, Pozzilli, IS, Italy.
- Department of Medicine, Surgery and Dentistry, University of Salerno, Baronissi, Italy. Electronic address: [email protected].
Abstract
Progression independent of relapse activity (PIRA) contributes to long-term disability in multiple sclerosis (MS), even in early stages. However, predicting short-term PIRA in routine clinical settings remains a challenge. To develop and evaluate machine learning (ML) models to predict PIRA in relapsing MS using routinely available clinical and conventional MRI-derived features. We developed two ML models to predict PIRA at 24 and 36 months in relapsing MS using baseline and longitudinal clinical and conventional MRI-derived data including brain and spine lesion burden, atrophy, and change in structural connectivity (ChaCo) scores. A Naïve Bayes classifier was trained after feature selection and class balancing with Synthetic Minority Over-sampling Technique (SMOTE). Among 186 patients, 12.4% experienced PIRA at 24 months. In a longitudinal subset (n = 81), 19.7% developed PIRA at 36 months. The 24-month model, achieved moderate discriminative performance (AUC = 0.73), mainly driven by baseline features. The 36-month model, including baseline disability, brain volume and volume change over time, new cervical cord lesions and baseline ChaCo features, showed improved accuracy (AUC = 0.83). ML models using clinical and conventional MRI features can predict short-term PIRA with moderate-to-high accuracy. Incorporating imaging changes over time enhances prediction and may support earlier individualized treatment strategies.