Artificial Intelligence for Gadolinium-Sparing Detection of Active Multiple Sclerosis Lesions: A Structured Narrative...

Gadolinium-enhanced T1-weighted magnetic resonance imaging (MRI; T1c+) remains the reference standard for identifying active inflammatory lesions in multiple sclerosis (MS). However, cumulative exposure raises concerns regarding tissue deposition, cost, scan duration, and patient burden. Artificial intelligence (AI) has emerged as a potential strategy to reduce reliance on contrast administration.

We conducted a structured narrative review of studies evaluating AI-based approaches for detecting active MS lesions without routine gadolinium use. Eligible studies included those that generated synthetic contrast-equivalent images or directly classified lesion activity using non-contrast MRI. Outcomes included diagnostic test accuracy, reconstruction fidelity, model characteristics, and methodological quality, which were assessed using an adapted QUADAS-2 framework.

Among studies reporting diagnostic accuracy against a gadolinium reference standard, sensitivity ranged from 72% to 100% and specificity from 66% to 96%. Only one study reported a complete contingency table; a second reported confusion matrices that are not reproduced here, and only one classifier included external multicenter validation. Image-synthesis studies demonstrated visually plausible outputs but were frequently validated against non-gadolinium targets, limiting clinical applicability. Common limitations included lack of prospective validation, limited reporting transparency, and insufficient evaluation of generalizability.

AI-based approaches demonstrate feasibility for gadolinium-sparing MS imaging but are not yet sufficient to replace contrast-enhanced MRI in routine clinical practice. A plausible near-term role is as a triage tool to identify low-risk follow-up examinations in which gadolinium administration may be deferred. Prospective multicenter studies with standardized reporting and true post-contrast reference standards are needed before clinical adoption.