Diagnostic efficacy of deep learning-based denoising of low-field 0.55 T compared to conventional 3 T knee MRI.

Objectives

To evaluate the diagnostic performance of low-field 0.55 T knee magnetic resonance imaging (MRI) with and without deep learning (DL)-based image denoising techniques compared to conventional 3 T MRI.

Methods

This retrospective study included 33 knee MRIs from 26 patients (10 women, mean age 52 years; ±16.9 years) with knee pain (mean duration 2.7 years; ±4.8 years) who underwent both 0.55 T and 3 T MRI. DL-based denoising (ImT-MRD) was applied to all 0.55 T datasets. Four radiologists assessed image quality, anatomical landmark visibility (4-point Likert scale), and diagnostic confidence for pathology detection (10-point scale). Diagnostic accuracy, including sensitivity and specificity, was calculated using consensus reading as standard of reference. Agreement was performed using Lin's concordance correlation coefficient (CCC).

Results

0.55 T without DL showed a moderate sensitivity of 0.83, a specificity of 0.82, and an overall diagnostic accuracy of 0.83. For 3 T, sensitivity, specificity, and diagnostic accuracy were 0.95, 0.98, and 0.97, respectively. 0.55 T with DL improved sensitivity to 0.97 and diagnostic accuracy to 0.98. Agreement with 3 T was excellent for 0.55 T with DL (CCC: 0.94) and substantial for 0.55 T without DL (CCC: 0.78). Image quality was significantly higher for 0.55 T with DL compared to 0.55 T without DL (2.92vs.2.55;p = 0.005), yet 3 T outperformed both (3.44;p < 0.001). Diagnostic confidence was significantly higher for 0.55 T with DL compared to 0.55 T without DL (8.21vs.7.66;p < 0.001), with 3 T demonstrating the highest confidence levels overall (8.60; all p < 0.001).

Conclusion

DL-based denoising of 0.55 T knee MRI significantly enhances the diagnostic performance, demonstrating results comparable to conventional 3 T, which were not achieved with 0.55 T MRI alone.