Refining Intra-Arterial Therapy Selection for Large Hepatocellular Carcinoma: A Deep Learning Approach Based on Covariate Interaction Analysis.

Abstract

Background: Hepatocellular carcinoma (HCC) is a major global health burden, with most patients presenting at advanced stages, limiting treatment options to intra-arterial therapy (IAT) such as transarterial chemoembolization (TACE) and hepatic arterial infusion chemotherapy (HAIC). However, optimizing IAT selection for large HCC remains challenging due to tumor heterogeneity and varying patient responses.

Aim: To develop and validate a deep learning (DL) model for guidance of decision-making between TACE and HAIC for unresectable HCC.

Methods: We conducted a retrospective, multi-center study involving 900 patients with large HCC treated with IATs. The DEep Learning for Interaction and Covariate Analysis in Intra-arterial Therapy SElection (DELICAITE) model integrates deep convolutional neural networks (DCNN) with covariate interaction analysis. The model was trained on dual-modal clinical and imaging data to predict treatment response and was validated using prospective and independent external validation cohorts.

Results: The DELICAITE model demonstrated superior discriminative ability and accuracy in predicting progressive disease (PD) in both internal and external test sets, with AUCs of 0.756, 0.664, and 0.701, respectively. Patients classified by the model into the "Maintain" group showed significantly longer overall survival (OS) compared to the "Alter" group (11.3 months vs 8.1 months, P < 0.001). The model's performance was further supported by its ability to stratify patients into subgroups most likely to benefit from TACE or HAIC.

Conclusion: The DELICAITE model provides a precise and innovative approach to refine IAT schemes for large HCC, offering clinicians a reliable tool to select the most suitable treatment option and potentially improve patient survival outcomes.