Image quality of combined PROPELLER and deep learning reconstruction using an accelerated T2-weighted imaging technique for prostate MRI

Abstract

Motion artifacts can degrade prostate MRI quality. PROPELLER read-out (PR), a radial acquisition technique, is an established approach that can reduce motion artifacts but requires longer acquisition times. We explored whether combining PR with accelerated acquisition and reconstruction methods yields image quality gains while reducing scan times. To compare the image quality of standard Cartesian (CR) and PROPELLER read-outs (PR) using an accelerated fast spin echo-based technique (VORTEX), combined with deep learning reconstruction (DLR). In 80 consecutive patients, VORTEX with CR was performed for the axial, sagittal, and coronal planes with scan time of 84, 50, and 54 s, respectively, while VORTEX with PR was performed in the coronal (71 s) and sagittal planes (92 s), for a total of five imaging datasets. All datasets were reconstructed using a commercially available DLR (AIR TM Recon DL). Five radiologists performed image quality assessments. Inter-reader agreement was estimated using Gwet’s AC2 statistics. All VORTEX derived images were assigned a T 2 w image PI-QUAL score of ≥3, and all radiologists perceived fewer artifacts in the sagittal and coronal PROPELLER images ( p < .001). Four out of five radiologists also observed both superior overall image quality and reduced noise for all PROPELLER derived images ( p < .001). Inter-rater agreement was good to very good for all assessments (AC2 0.77–0.88). VORTEX with PR and DLR can achieve fast high-quality T 2 w sagittal and coronal imaging of prostate, with all acquisitions completed within 95 s per imaging plane and a total triplanar acquisition time of 4 min 7 s.