Prospective study assessing the validity of accelerated 2D Fast Spin Echo (2D FSE) based high-resolution knee MRI and T2 mapping using deep learning reconstruction

TL;DR

This study shows that deep learning reconstruction can speed up knee MRI scans while maintaining image quality and improving detection of cartilage damage.

Contribution

The study validates the clinical utility of deep learning reconstruction in accelerating high-resolution knee MRI and T2 mapping.

Findings

DLR improved image sharpness and objective metrics (SNR/CNR) compared to conventional FSE.

DLR-based imaging showed higher diagnostic performance for cartilage lesions than standard FSE.

Combining DLR with T2 mapping improved diagnostic accuracy for cartilage injury detection.

Abstract

A routine fast spin-echo (FSE) MRI protocol is widely used to evaluate structural injuries of the knee. While adding a T2 mapping sequence to this protocol increases sensitivity for detecting early cartilage lesions, it is time-consuming. Deep learning reconstruction (DLR) can provide accelerated, high-quality imaging but requires further clinical validation. This study aimed to evaluate the image quality and diagnostic performance of an accelerated, high-resolution 2D FSE protocol using DLR. Additionally, it sought to determine whether a routine MR imaging protocol combining an accelerated T2 mapping sequence with DLR could improve diagnostic performance for the detection of cartilage lesions, using arthroscopy as the reference standard.

A total of 92 patients underwent 2D FSE based routine knee imaging on 3.0T MRI, 39 of whom also underwent sagittal T2 mapping with different k-space-based acceleration factor of 2 or 3 and then reconstructed using conventional and deep learning reconstruction algorithms as FSEO, FSEDLR, T2O − ARC=2,3, and T2DLR − ARC=2,3 and arthroscopy of the knee joint. Two radiologists subjectively and objectively evaluated both FSEO and FSEDLR. The inter-reader agreement for each pathology and image quality score was assessed using Cohen’s κ. The objective metrics (SNR/CNR) between sequences was analyzed using paired t-test or Wilcoxon signed-rank test according to data normality. A two-sided p-value of less than 0.05 was considered statistically significant. Additionally, diagnostic performance of routine knee MR images and DLR or non-DLR T2 measurements respectively for grading knee cartilage were also compared. Articular cartilage was categorized according to International Cartilage Repair Society (ICRS). Each articular surface was then evaluated at arthroscopy. Receiver operating characteristic curve (ROC) was used to analyze diagnostic performance using arthroscopic results as reference.

Inter-reader agreement of subjective assessment ranged from 0.70 (95% CI: 0.46–0.94,) to 0.89 (95% CI: 0.79–0.99,) and higher score on FSEDLR than FSEO. Sharpness for FSEDLR was rated to be excellent (median Likert score, 5; range, 5–5), higher compared to FSEO (median Likert score, 5; range,4–5),(P < 0.001)). Both SNR and CNR of FSEDLR were higher than those of FSEO (P < 0.001). Inter-reader agreement was almost perfect, withκvalues between 0.94 (95% CI: 0.85-1.0) to 1.00 (95% CI: 1.0–1.0) for the detection of internal derangement and substantial to almost perfect between 0.70 (95% CI:0.52–0.88) and 0.93 (95% CI:0.85-1.0) for the assessment of cartilage defects. FSEDLR (Reader 1 AUC, 0.77; 95% CI: 0.69–0.84 and Reader 2 AUC, 0.86; 95% CI: 0.78–0.91) had higher diagnostic performance than FSEO (Reader 1 AUC, 0.74; 95% CI: 0.66–0.81 and Reader 2 AUC, 0.80; 95% CI: 0.72–0.86; P = 0.005) for articular cartilage lesions. Moreover, Reader 1 achieved the higher diagnostic efficacy (AUC, 0.84; 95% CI: 0.76–0.90) in differentiating normal-appearing from injury-visible cartilage when using both routine FSEDLR images and T2DLR − ARC=2,while Reader 2 achieved an AUC of 0.86 (95% CI: 0.78–0.91) with routine FSEDLR images.

Our preliminary results indicate that the accelerated DLR FSE protocol provided diagnostic performance equivalent to the standard protocol for internal derangement, with potential improvement for the detection of cartilage lesions, while delivering higher image quality and quantitative T2 data within a clinically feasible scan time. These findings suggest its potential value for a comprehensive and efficient assessment of knee injury.

The online version contains supplementary material available at 10.1186/s12891-025-09482-2.

Figures