Deep-learning model surpasses traditional methods in predicting HCC recurrence

A novel deep-learning model, Recurr-Net, showed superior accuracy in predicting hepatocellular carcinoma (HCC) recurrence post-surgery compared to histological microvascular invasion (MVI) and conventional clinical prediction scores. In a study of 1,231 patients, Recurr-Net achieved AUROC scores between 0.770 and 0.857 for internal validation, significantly better than MVI and various clinical risk scores. The model effectively stratified recurrence risks at two and five years, demonstrating its potential for improved preoperative prognostication in HCC cases.