Application of Deep Learning-Based Reconstruction to Magnetic Resonance Imaging of Canine Stifle Joint in Healthy...

Stifle joint magnetic resonance imaging (MRI) is a valuable modality for diagnosing stifle joint disorders. In veterinary practice, reducing slice thickness is crucial for the detailed assessment of intra-articular structures. Recent advancements in deep learning-based reconstruction (DLR) have overcome the conventional trade-off between spatial resolution and noise, mitigating image quality degradation in thin-slice MRI. We hypothesized that DLR-applied stifle joint MRI could surpass conventional MRI in terms of diagnostic performance, especially with reduced slice thicknesses. This prospective, comparative pilot study compared conventional versus DLR-applied stifle joint MRI data of eight healthy beagle dogs by using sagittal T2-weighted fat saturation (T2W_FS) and proton density-weighted fat saturation (PDW_FS) sequences. The following groups were formed on the basis of slice thickness: conventional 2 mm group (₂CON) and DLR-applied 2, 1.5, and 1 mm groups (₂DLR, _1.5DLR, and ₁DLR, respectively). Quantitative analysis assessed the signal-to-noise ratio (SNR) and contrast-to-noise ratio, whereas qualitative analysis evaluated the structural visibility of the cranial cruciate ligament (CCL), meniscus, and bone; perceived SNR; and overall image quality using a four-point Likert scale. In both T2W_FS and PDW_FS sequences, ₂DLR, _1.5DLR, and ₁DLR exhibited a significantly higher SNR than ₂CON. Moreover, in both T2W_FS and PDW_FS sequences, the DLR groups scored significantly higher than ₂CON for all qualitative indices. Notably, ₁DLR exhibited the highest CCL visibility, followed by _1.5DLR and ₂DLR. In conclusion, DLR-applied stifle joint MRI can reduce slice thickness and enhance image quality and anatomical delineation, potentially increasing the diagnostic accuracy of stifle joint disorders.