AI-assisted acceleration combined with deep learning of the knee: feasibility of anomaly detection

Abstract

To evaluate the performance of deep learning reconstruction (DLR) for knee magnetic resonance imaging (MRI), and to assess the diagnostic value of combining DLR with artificial intelligence-assisted compressed sensing (ACS) technology in detecting knee abnormalities. This prospective study included 10 healthy volunteers and 40 patients with knee abnormalities. Patients were divided into Conventional (Con), ACS, and ACS + DLR groups. Acceleration factors were optimized in a pilot study on healthy volunteers (acceleration factor: 1–3). Images of different groups were compared in acquisition time, signal to noise ratio (SNR), contrast to noise ratio (CNR), subjective image quality, and diagnostic confidence. Two musculoskeletal radiologists with 10 and 15 years of experience independently reviewed all images in a blinded manner. The diagnostic criteria were based on standard MRI features, including signal intensity alterations, morphological deformities, and structural discontinuities. Inter-observer agreement was assessed using intraclass correlation coefficients. Differences in SNR, CNR, subjective image quality, and diagnostic confidence among Con, ACS, and ACS + DLR sequences were compared using one-way ANOVA or Friedman tests, as appropriate. Compared with Con group, ACS and ACS + DLR groups reduced acquisition time by 55%, and the SNR and CNR were significantly improved. Subjective image quality and diagnostic confidence scores in both the ACS and ACS + DLR groups were significantly higher than those in the Con group. Notably, compared with ACS group, the ACS + DLR group demonstrated superior performance in terms of image quality and diagnostic confidence, with significantly higher scores. Moreover, this combined technique enabled clearer visualization of meniscal tear grading, reduced implant-related artifacts, and enhanced the delineation of bone marrow edema and surrounding soft tissues. In knee MRI examinations, ACS combined with DLR reduces scan time while maintaining or improving image quality, thereby enhancing the detection of subtle structural lesions.