Deep Learning-Based Reconstruction Improves Image Quality in Canine Cranial Abdominal MRI: A Prospective Pilot Study

The application of deep learning-based reconstruction (DLR) to magnetic resonance imaging (MRI) has been recently introduced in human and veterinary medicine to improve image quality without prolonging acquisition time. We hypothesized that cranial abdominal MRI with DLR would have superior image quality than conventional MRI. This prospective comparative pilot study aimed to compare cranial abdominal MR image quality with versus without DLR application in dogs. Transverse T2-weighted and T1-weighted images with and without contrast enhancement of the cranial abdomen region were performed in 10 clinically healthy dogs. The original and DLR images were generated from a single set of raw data using the DLR network embedded reconstruction pathway. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated for quantitative analysis. Qualitative analysis included evaluation of the edge sharpness of the liver and pancreas, the conspicuity of the adrenal glands, respiratory motion artifacts, coarseness, and overall image quality using a 4-point scale. Quantitative and qualitative values were compared between images with DLR and their counterparts. The mean SNR and CNR of all images with DLR were significantly higher than those of their counterparts (p < 0.05). Qualitative analysis parameters, except for respiratory motion artifacts, were significantly superior in images with DLR compared with counterparts for all sequences (p < 0.05). The present study demonstrated that cranial abdominal MRI with DLR is feasible to improve image quality without prolonging acquisition time.