Multimodal multiphasic preoperative image-based deep-learning predicts HCC outcomes after curative surgery

Area of Science:

• Hepatocellular Carcinoma (HCC) Research
• Artificial Intelligence in Oncology
• Medical Imaging Analysis

Background:

• Hepatocellular carcinoma (HCC) recurrence is common after curative surgery.
• Histological microvascular invasion (MVI) predicts recurrence but lacks preoperative utility.
• Existing clinical prediction scores for HCC recurrence show variable performance.

Purpose of the Study:

• To develop and validate Recurr-NET, a deep-learning model for predicting HCC recurrence.
• To compare the predictive accuracy of Recurr-NET against MVI and clinical risk scores.
• To assess the potential of Recurr-NET for preoperative prognostication in HCC.

Main Methods:

• Developed Recurr-NET, a multimodal random survival forest deep-learning model.
• Incorporated preoperative CT scans and clinical parameters from HCC patients.
• Validated the model using internal (Hong Kong) and external (Taiwan) cohorts.

Main Results:

• Recurr-NET demonstrated excellent accuracy in predicting HCC recurrence from 1 to 5 years (internal AUROC 0.770-0.857, external AUROC 0.758-0.798).
• Recurr-NET significantly outperformed MVI and clinical risk scores (p < 0.001).
• The model showed superior risk stratification for recurrence and mortality compared to MVI.

Conclusions:

• Recurr-NET provides accurate preoperative prediction of HCC recurrence.
• The model surpasses MVI and clinical scores in predictive performance.
• Recurr-NET holds significant potential for improving preoperative prognostication in HCC management.