Skip to content
Review Open access

Application of machine learning to MRI biomarkers in predicting progression and PIRA in multiple sclerosis: a systematic review

Oct 2026 · ANPAbstract_CBN26 · 0 citations

Abstract

Background: Multiple Sclerosis (MS) is the leading cause of non-traumatic neurological disability in young adults. Structural magnetic resonance imaging (MRI), by focusing on lesion load, fails to reflect the patient's actual disability because it does not capture diffuse tissue damage and early atrophy. To monitor Progression Independent of Relapse Activity (PIRA), the application of Machine Learning (ML) and Artificial Intelligence (AI) algorithms to advanced MRI biomarkers emerges as a strategic advance. Objective: To evaluate the efficacy of ML/AI applied to MRI biomarkers in predicting MS progression, comparing their performance with conventional radiological analysis and traditional statistical models. Methods: Systematic review of the PubMed database over the past 10 years, encompassing MS patients with different clinical manifestations. The intervention analyzed was the application of AI/ML algorithms to MRI data. After independent duplicate screening using Rayyan software (176 articles), 55 were included in the final analysis. Disease progression was assessed using the Expanded Disability Status Scale (EDSS), occurrence of PIRA, and long-term disability accumulation. Results: AI predictive models (such as Neural Networks and XGBoost) significantly outperformed conventional methods in predicting clinical trajectory. AI confirmed that microstructural volume markers, especially thalamic volume, cortical atrophy, and spinal cord atrophy, are far more robust predictors of physical and cognitive decline than the traditional lesion volume metric. Unsupervised learning identified new MRI-based progression subtypes, such as deep early atrophy and cortical atrophy subtypes, drastically improving risk stratification. The use of the "Brain Age" metric (Brain-PAD) by AI revealed accelerated structural aging in MS, predicting future disability accumulation more rapidly than chronological age. Multimodal integration, combining MRI with serum biomarkers (sNfL, GFAP) or genetic data, substantially increased predictive accuracy. Furthermore, AI demonstrated superior sensitivity compared to standard radiological reporting in detecting new or expanding lesions, even modifying treatment management in approximately 10% of evaluated cases. Conclusion: The transition to precision medicine in MS requires overcoming quantitative limitations. ML combined with advanced MRI biomarkers offers a tool superior to the current standard for anticipating decline trajectories and PIRA risk, enabling early interventions and optimized monitoring of diffuse neurodegeneration.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.