Abstract
OBJECTIVES: To explore the value of a T1 mapping-based radiomic model for evaluating liver function. METHODS: From September 2020 to October 2022, 163 patients were retrospectively recruited and categorized into normal liver function group, chronic liver disease group without cirrhosis, Child‒Pugh class A group, and Child‒Pugh class B and C group. Patients were randomly split into training and testing sets. Radiomic features were extracted from T1 mapping images taken both pre- and post-contrast injection, as well as during the hepatobiliary phase (HBP). Radiomic models were constructed to stratify chronic liver disease, cirrhosis and decompensated cirrhosis. Model performance was assessed with receiver operating characteristic curve analysis, and decision curve analysis. RESULTS: The K-Nearest Neighbors model demonstrated the best generalization across native T1 map, HBP T1 maps and HBP images. In the training set, based on native T1 maps, it achieved accuracies of 0.83, 0.86, and 0.86 in distinguishing chronic liver disease, cirrhosis, and decompensated cirrhosis, with corresponding AUCs of 0.92, 0.92, and 0.95. In the testing set, the accuracies were 0.75, 0.89, and 0.71, with AUCs of 0.79, 0.92, and 0.83, respectively. When using HBP images with T1 maps, the accuracies were 0.72, 0.90, and 0.72 in the testing set in identifying chronic liver disease, cirrhosis, and decompensated cirrhosis with AUCs of 0.82, 0.93, and 0.79, respectively. CONCLUSION: Radiomic analysis based on native T1 map, and HBP with or without T1 map images shows promising potential for liver function assessment, particularly in distinguishing cirrhosis.