Abstract: Objective The purpose of this study was to investigate deep learning-based high-resolution magnetic resonance imaging (MRI) in the differential diagnosis of focal liver lesions(FLL). Methods 98 patients with FLL were admitted to Second People's Hospital, Lianyungang City between January 2023 and December 2023, and all underwent hrMRI scan. The imaging was repeatedly read by Residual Network under PyTorch, persons encountered in the first half of the year were selected as training set, and those in the second half of the year were acted as validation set. Tissues obtained by fine needle aspiration biopsies or by surgery were routinely pathologically examined. Area under receiver operating characteristic curve (AUC) was applied to evaluate diagnostic performance. Results The eligible image quality rate for diagnosis by deep learning-based hrMRI was 93.9%, much higher than 84.7% by hrMRI (P＜0.05); of 98 patients with FLL, histo-pathological examination showed malignant lesions in 55 cases (cholangiocarcinoma in 5 and hepatocellular carcinoma in 50), and benign lesions in 43 cases (focal nodular hyperplasia in 25 and cirrhotic nodules in 18); sensitivity, specificity, accuracy, positive predictive value and negative predictive value by deep learning-based hrMRI were 89.1%, 86.1%, 87.8%, 89.1% and 86.0%, much superior to 81.8%, 67.8%, 76.5%, 77.6% and 75.0%(P＜0.05) by hrMRI. Conclusion Efficacy by deep learning-based hrMRI in differentiating quality of intrahepatic lesions is satisfactory, which warrants further investigation.

Key words: Hepatoma, Focal liver lesions, Deep learning, High resolution magnetic resonance imaging, Differential diagnosis