Applying Deep Learning to Accelerated Clinical Brain Magnetic Resonance Imaging for Multiple Sclerosis

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Magnetic resonance (MR) scans are crucial for monitoring multiple sclerosis (MS).
• Accelerating MR scan times is needed to reduce patient discomfort, scheduling delays, and costs.
• Deep learning (DL) offers potential for improving accelerated MR scan quality.

Purpose of the Study:

• To evaluate the clinical utility of a DL model in enhancing image quality for accelerated brain MR scans in people with MS (PwMS).
• To assess if DL-improved accelerated scans retain diagnostic accuracy for key volumetric and lesion metrics.

Main Methods:

• Fast 3D T1w BRAVO and T2w FLAIR MRI sequences were acquired alongside conventional scans.
• A DL model was trained to reconstruct high-quality images from accelerated scans.
• Clinical volumetrics (brain, thalamic, gray matter, white matter) and T2 lesion volume were compared across conventional, fast, and DL-enhanced scans.

Main Results:

• Statistically significant but minor differences were observed in T1w volumetrics between conventional and DL-enhanced fast scans.
• No significant differences were found in the correlation between T1w volumetrics and patient-reported outcomes for MS symptom burden and disability.

Conclusions:

• A DL model can effectively enhance the image quality of accelerated brain MR scans for PwMS.
• DL-enhanced accelerated MR scans show potential for maintaining clinical relevance in MS monitoring and outcome assessment.