Deep learning-accelerated 3D flair for white matter lesion detection in multiple sclerosis: a feasibility study.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Deep learning (DL)-based image reconstruction (DLBIR) offers potential for accelerated MRI acquisition and improved image quality.
• Conventional 3D FLAIR (3D-FLAIR_SOC) is a standard sequence for evaluating multiple sclerosis (MS).

Purpose of the Study:

• To compare the image quality and lesion detection capabilities of a DLBIR-based 3D FLAIR (3D-FLAIR_DL) against conventional 3D FLAIR (3D-FLAIR_SOC) in MS patients.
• To assess the impact of DLBIR on diagnostic confidence and quantitative imaging metrics.

Main Methods:

• A prospective, reader-blinded study involving 26 MS patients who underwent both 3D-FLAIR_DL and 3D-FLAIR_SOC sequences on a 3T scanner.
• Two neuroradiologists evaluated image quality using Likert-like scales, with specific grading for lesion conspicuity (<3 mm and ≥3 mm).
• Quantitative analysis included lesion counts, apparent signal-to-noise ratio (aSNR), and apparent contrast-to-noise ratio (aCNR).

Main Results:

• 3D-FLAIR_DL demonstrated significantly superior qualitative image quality scores (p < 0.001) with high inter-reader agreement.
• DLBIR improved lesion conspicuity and diagnostic confidence, particularly for smaller lesions (<3 mm), detecting 29 additional lesions.
• Quantitative metrics showed significantly higher aSNR and aCNR for DLBIR images (p < 0.001), with a 32% reduction in acquisition time.

Conclusions:

• DLBIR 3D FLAIR significantly enhances lesion detection and overall image quality in MS patients.
• The findings support the potential integration of DLBIR 3D FLAIR into standard MS imaging protocols.
• Further validation in larger, diverse cohorts is recommended as DLBIR algorithms continue to evolve.