For the roughly 2.9 million people living with multiple sclerosis worldwide, routine MRI surveillance is a clinical cornerstone — but long scan times strain both patients and imaging suites. A finding that cuts acquisition time nearly in half while preserving diagnostic accuracy could meaningfully reshape how disease activity is monitored, particularly for patients who struggle with prolonged scanner exposure.
This prospective study enrolled 94 participants — 77 with confirmed MS and 17 with suspected chronic inflammatory CNS disease — who each underwent both a conventional 3D T2 SPACE dark-fluid sequence (5 minutes, 1 second) and a deep-learning-reconstructed version (2 minutes, 48 seconds) at 1.5 Tesla. The primary endpoint was interchangeability for detecting new white matter lesions under the 2024 revised McDonald criteria, evaluated across periventricular, cortical/juxtacortical, and infratentorial regions by three independent readers plus one experienced neuroradiologist using certified lesion-detection software. The predefined equivalence margin was 5%, and inter-reader reliability was quantified using Gwet's AC1 and AC2 coefficients. Secondary endpoints included total lesion count comparability and subjective image quality ratings.
This work sits at a productive intersection of deep-learning image reconstruction and neuroradiology — a field where prior studies have demonstrated speed gains but rarely rigorously tested clinical interchangeability using preregistered equivalence margins. The 2024 McDonald criteria revision makes lesion-level precision increasingly consequential, raising the bar for any accelerated technique aiming to replace established sequences. A 44% reduction in acquisition time is clinically significant: it could allow more patients to be scanned per session, reduce motion artifact in less tolerant patients, and lower the cumulative burden on healthcare systems without downgrading diagnostic confidence. Key limitations include the single-field-strength design (1.5T only), a relatively modest cohort, and the inherently cross-sectional interchangeability design rather than longitudinal tracking of lesion evolution. If replicated at 3T — where most MS centers now operate — this approach could represent a genuine workflow shift rather than an incremental refinement.