3D Foundation Model Phenotypes HCC on MRI and Stratifies Recurrence Risk

A 3D foundation model automatically phenotyped hepatocellular carcinoma on preoperative MRI with AUC 0.932 vs 0.825 for senior radiologists; its imaging phenotype predicted recurrence (HR 2.16, 95% CI 1.20–3.87).

• Retrospective two-center study of 551 patients undergoing hepatectomy after gadoxetic acid–enhanced MRI; development n=291, validation n=192, external test n=68.
• 3D MorphStratify achieved micro-averaged AUC 0.932 for pathological gross classification, outperforming senior radiologists (AUC 0.825) with about 40-fold improved efficiency.
• Combined Morph-CIP model yielded 3-year recurrence-free survival AUCs of 0.765 (training), 0.711 (internal test), and 0.702 (external test).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.