A Generative Model-based System for Predicting Survival and Guiding Treatment Decisions in Patients With Unresectable Hepatocellularcarcinoma Undergoing Transcatheter Arterial Chemoembolization in Combination With Immunotherapy and Targeted Therapy

The entry point of this study is the proposition of "generative longitudinal prediction," which utilizes only pre-treatment imaging to create high-fidelity predictions of post-treatment imaging. This approach effectively overcomes the clinical challenge of acquiring genuine longitudinal follow-up data. This paradigm shift not only tackles the scarcity of longitudinal data but also introduces an innovative method for treatment simulation using digital twins. Clinicians can intuitively assess the potential efficacy of various treatment plans before intervention through virtually generated multi-timepoint imaging, providing a visual foundation for personalized treatment decisions. This research merges generative AI with dynamic risk models to achieve: 1) a transition from static assessment to dynamic simulation; 2) earlier survival predictions; and 3) personalized optimization of treatment plans. By eliminating dependence on longitudinal data, we aim to deliver more precise and individualized treatment decision support for advanced liver cancer patients, ultimately enhancing survival outcomes and quality of life.