AUTHORS: Wynen M., Vanden Bulcke C., Borrelli S., Gordaliza P.M., Stölting A., Macq B., Du Pasquier R., Filippi M., Absinta M., Reich D.S., Bach Cuadra M., Maggi P.

European Committee for Treatment and Research in Multiple Sclerosis, : , Copenhagen, September 2024

ABSTRACT

Current differential diagnosis dissemination in space (DIS) McDonald criteria (1) tend to favor sensitivity over specificity (2). This leads to early detection of the disease (3) but also increases the risk of incorrectly diagnosing patients with MS-mimicking diseases (4). New advanced MRI biomarkers – the central vein sign (CVS), cortical lesions (CLs), and paramagnetic rim lesions (PRLs) show promising high specificity for MS. However, the full-count assessment of these advanced biomarkers is time-consuming and often incompatible with clinical practice. Objectives Using machine learning (ML), we investigate: • the incorporation of CVS, CLs, and PRLs to improve MS differential diagnosis sensitivity-specificity trade-off • the diagnostic performance of more practical, simplified advanced biomarker assessments Finally, the publication of an online tool allows the interaction with trained ML models. Conclusions • CVS emerges as the most diagnostically powerful biomarker • ML models combining CVS, CL, and PRL show the highest MS diagnostic performance and clearly outperform the current MS DIS diagnostic criteria. • Simplified assessments are competitive against full-count assessments, considerably reducing the time burden associated with image analysis.

BibTex

http://hdl.handle.net/2078.1/293910

Module: