Effects of Deep Learning-Based Reconstruction on the Quality of Accelerated Contrast-Enhanced Neck MRI

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Conventional MRI techniques can be limited by scan time and image quality.
• Deep learning (DL) offers potential for image reconstruction and acceleration.

Purpose of the Study:

• To compare the image quality of deep learning-reconstructed turbo spin-echo (DL-TSE) versus conventionally interpolated turbo spin-echo (Conv-TSE) in contrast-enhanced neck MRI.
• To assess the impact of DL-TSE on scan time and diagnostic performance.

Main Methods:

• Contrast-enhanced T1-weighted DL-TSE and Conv-TSE images were acquired from 106 patients using 3T scanners.
• DL-TSE utilized a specific algorithm for interpolation and denoising, reducing scan times.
• Image quality metrics including non-uniformity, contrast-to-noise ratio, and full width at half maximum (FWHM) were evaluated.
• Neuroradiologists assessed overall quality, sharpness, mucosal conspicuity, artifacts, and lesion detection.

Main Results:

• DL-TSE demonstrated sharper stationary phantom scans and higher contrast-to-noise ratio in patient scans.
• Significantly improved regional mucosal FWHM was observed in DL-TSE, particularly in the oropharynx and hypopharynx.
• DL-TSE yielded higher overall image quality and sharpness, with improved mucosal conspicuity and lesion detection confidence.
• While DL-TSE reduced extracorporeal ghost artifacts, artifacts overlapping anatomical structures were slightly more pronounced.

Conclusions:

• DL-based reconstruction in accelerated neck MRI significantly enhances image quality, sharpness, and mucosal conspicuity in motion-prone areas.
• DL-TSE improves lesion detection confidence and enables substantial scan time reduction.
• Despite some artifact trade-offs, DL-TSE demonstrates enhanced diagnostic performance in contrast-enhanced neck MRI.