Prospective Evaluation of Accelerated Brain MRI Using Deep Learning-Based Reconstruction: Simultaneous Application to...

Area of Science:

• Radiology and Imaging Science
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Conventional brain MRI scans can be time-consuming, potentially limiting patient throughput and comfort.
• Deep learning-based reconstruction offers a potential solution for accelerating MRI acquisition and improving image quality.

Purpose of the Study:

• To prospectively evaluate the impact of accelerated deep learning-based reconstruction (Accel-DL) on brain MRI quality.
• To assess the reduction in scan time achieved by Accel-DL compared to conventional MRI techniques.
• To determine if Accel-DL affects volumetric quantification accuracy in brain structures and lesions.

Main Methods:

• 150 participants underwent brain MRI using conventional and Accel-DL methods on three different 3T scanner vendors.
• Accel-DL utilized optimized parameters and U-Net-based software for reconstruction from undersampled data.
• Image quality, SNR, CNR, scan times, and volumetric measurements (regional structures, WMHs) were assessed and compared.

Main Results:

• Accel-DL reduced scan time by an average of 39.4% while significantly improving overall image quality, structure delineation, and reducing artifacts.
• Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were significantly increased with Accel-DL.
• Volumetric measurements showed no significant differences for most regions and WMH segmentation, except for deep gray matter and leukocortical lesions.

Conclusions:

• Accel-DL substantially reduces brain MRI scan time and enhances image quality across various sequences.
• The technique demonstrates robustness in maintaining accurate volumetric quantification, including lesion assessment.
• Accel-DL represents a promising advancement for efficient and high-quality neuroimaging.