Vessel wall imaging-dedicated deep learning (VWI-DL): Toward a 5-min clinically robust MR protocol

Motivation: Long scan times (8-12 minutes) in whole-brain MR vessel wall imaging (VWI) cause patient discomfort and motion artifacts, limiting clinical utilization. Faster VWI with preserved image quality is highly desirable. Goal(s): To develop a VWI-dedicated deep learning model to substantially accelerate data acquisition without compromising image quality. Approach: We developed a multi-view deep learning model using SwinIR as the backbone, trained it on 12-min VWI raw data, and used a voting mechanism across views to enhance consistency. Both retrospective and prospective testing were performed. Results: The model reduced scan time to <6 minutes with PSNR and SSIM improvements of 9.69 and 0.28. Impact: The VWI-dedicated deep learning model enables faster data acquisition with preserved image quality compared to standard clinical protocols, which may enhance VWI's robustness and clinical throughput and thus promote its widespread clinical adoption.