Evaluation of deep learning reconstruction in accelerated knee MRI: comparison of visual and diagnostic performance metrics.
Authors
Affiliations (2)
Affiliations (2)
- Department of Radiology, The First Affiliated Hospital, Guangxi Medical University, No 6 Shuangyong Road, 530021, Nanning, Guangxi Zhuang Autonomous Region, China.
- Department of Radiology, The First Affiliated Hospital, Guangxi Medical University, No 6 Shuangyong Road, 530021, Nanning, Guangxi Zhuang Autonomous Region, China. [email protected].
Abstract
To investigate the clinical value of deep learning reconstruction (DLR) in accelerated magnetic resonance imaging (MRI) of the knee and compare its visual quality and diagnostic performance metrics with conventional fast spin-echo T2-weighted imaging with fat suppression (FSE-T2WI-FS). This prospective study included 116 patients with knee injuries. All patients underwent both conventional FSE-T2WI-FS and DLR-accelerated FSE-T2WI-FS scans on a 1.5‑T MRI scanner. Two radiologists independently evaluated overall image quality, artifacts, and image sharpness using a 5-point Likert scale. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of lesion regions were measured. Subjective scores were compared using the Wilcoxon signed-rank test, SNR/CNR differences were analyzed via paired t tests, and inter-reader agreement was assessed using Cohen's kappa. The accelerated sequences with DLR achieved a 36 % reduction in total scan time compared to conventional sequences (p < 0.05), shortening acquisition from 9 min 50 s to 6 min 15 s. Moreover, DLR demonstrated superior artifact suppression and enhanced quantitative image quality, with significantly higher SNR and CNR (p < 0.001). Despite these improvements, diagnostic equivalence was maintained: No significant differences were observed in overall image quality, sharpness (p > 0.05), or lesion detection rates. Inter-reader agreement was good (κ> 0.75), further validating the clinical reliability of the DLR technique. Using DLR-accelerated FSE-T2WI-FS reduces scan time, suppresses artifacts, and improves quantitative image quality while maintaining diagnostic accuracy comparable to conventional sequences. This technology holds promise for optimizing clinical workflows in MRI of the knee.