Machine Learning-Based Combination of the Central Vein Sign, Cortical Lesions and Paramagnetic Rim Lesions: A Web-Based Tool for the Diagnosis of Multiple Sclerosis

Wynen, Maxence;Vanden Bulcke, Colin;et.al.
(2025) Brain Communications — (2025)

Files

ssrn-5192817_title.pdf
  • Open Access
  • Adobe PDF
  • 116.08 KB
Download

Details

Authors
Abstract
Background: Multiple sclerosis (MS) diagnostic criteria lack optimal specificity, leading to potential misdiagnosis. Advanced MRI biomarkers like the central vein sign (CVS), cortical lesions (CL), and paramagnetic rim lesions (PRL) are highly MS specific and could potentially improve diagnostic accuracy. Methods: We applied machine learning (ML) techniques to a retrospective, multicentric dataset of 364 MS/MS-mimic (204/118) and 42 prodromal MS/non-MS (26/16) adult patients, incorporating CVS, CL, and PRL biomarkers. We compared (5x2CV combined F-test) the diagnostic performance of 71 ML models using full-count and simplified biomarker assessments against the baseline dissemination in space (DIS) McDonald criteria. The aim was to evaluate the MS diagnostic power of combining these biomarkers in an MRI-only diagnostic framework. Findings: 51 models outperformed the DIS criteria, achieving balanced accuracy improvements of up to 13·0%. Twelve of these models only used simplified biomarker assessments. There was no significant difference between the best full-count and simplified models (p=0·29). Results were confirmed in two out-of-domain test sets, with simplified assessments performing better than their full-count counterparts. Interpretation: Within a non-invasive MRI-only diagnostic framework, we show that the incorporation of advanced imaging biomarkers into the MS-MRI diagnostic criteria significantly enhances the diagnostic accuracy - especially when using simplified CVS, CL and PRL assessments. The study also provides a publicly available online diagnostic tool, facilitating further interaction, validation and clinical support (https://www.msdiagnostictool.org).
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Wynen, M., Vanden Bulcke, C., & et al. (2025). Machine Learning-Based Combination of the Central Vein Sign, Cortical Lesions and Paramagnetic Rim Lesions: A Web-Based Tool for the Diagnosis of Multiple Sclerosis. Brain Communications. Accepted/in-press. https://doi.org/10.2139/ssrn.5192817 (Original work published 2025)