Deep Learning for Predicting Enhancing Lesions in Multiple Sclerosis from Noncontrast MRI.

Area of Science:

• Radiology
• Artificial Intelligence
• Neurology

Background:

• Enhancing lesions on MRI are key indicators of active multiple sclerosis (MS).
• Current methods require contrast agents, which carry risks and costs.
• Predicting these lesions without contrast is a significant clinical need.

Purpose of the Study:

• To assess the efficacy of deep learning algorithms in identifying MS-related enhancing lesions on unenhanced MRI scans.
• To determine if AI can predict disease activity without the need for gadolinium-based contrast agents.

Main Methods:

• A convolutional neural network (CNN) was developed to classify enhancing lesions on unenhanced multiparametric MRI.
• The network analyzed 1970 scans from 1008 patients, using postcontrast T1-weighted images as ground truth.
• Performance was evaluated using fivefold cross-validation, calculating sensitivity, specificity, and AUC.

Main Results:

• The deep learning model achieved moderate to high accuracy in predicting enhancing lesions.
• Slice-wise prediction sensitivity was 78% (±4.3) and specificity was 73% (±2.7).
• Participant-wise prediction yielded AUCs of 0.82 (±0.02) for slice-wise and 0.75 (±0.03) for participant-wise analysis.

Conclusions:

• Deep learning models can effectively identify multiple sclerosis enhancing lesions using conventional, unenhanced MRI sequences.
• This AI-driven approach demonstrates potential for accurate, non-contrast-enhanced monitoring of MS disease activity.
• The findings suggest a future where contrast-agent-free MRI analysis improves MS patient care.