Development and Validation of a Deep Learning Model for the Prediction of Hepatocellular Cancer Recurrence After Transplantation: The Time-Radiological Response- AlphafetoproteIN-Artificial Intelligence Model

In 1996, the introduction of the Milan criteria (MC) strongly modified the selection process of hepatocellular cancer (HCC) patients waiting for liver transplantation (LT). Many attempts to widen MC have been proposed. Initially, exclusively morphology-based (nodules number and target lesion diameter) criteria were created. In the last years, extended criteria also based on biological parameters have been added. Among the most adopted biology-based features, the levels of different tumor markers, liver function parameters like the model for end-stage liver disease (MELD), the radiological response after neo-adjuvant therapies, and the length of waiting-time (WT) can be reported.

Unfortunately, all the proposed models showed suboptimal prediction abilities for the risk of post-LT recurrence. Such impairment was derived from the limitations of the standard statistical methods to account for many variables and their non-linear interactions. Therefore, developing a model based on Artificial Intelligence (AI) represents an attractive way to improve prediction ability.

Thus, the investigators hypothesize that an AI model focused on an accurate post-transplant HCC recurrence prediction should improve our ability to pre-operatively identify patients with different classes of risk for HCC recurrence after transplant.

This study aims to develop an AI-derived prediction model combining morphology and biology variables. A Training Set derived from an International Cohort was adopted for doing this. A Test Set derived from the same International Cohort and a Validation Cohort were adopted for the internal and external validation, respectively. A user-friendly web calculator was also developed.