Thermal ablation is an important minimally invasive treatment for hepatocellular carcinoma (HCC), but local tumor progression (LTP) after ablation restricts the efficacy and status of ablation technology and seriously threatens patient survival. Insufficient coverage of thermal field is an important factor on the occurrence of LTP. Current thermal field planning relies on tumor contours and doctor experience, and the safety margin is uniform. Therefore, it cannot cope with the problem of insufficient coverage of thermal field caused by the different invasion capabilities of different tumors and different parts of the same tumor. This project intends to integratively analyze gray-scale ultrasound, contrast-enhanced ultrasound, magnetic resonance imaging and clinical information of HCC through deep canonical correlation analysis; summarize the prior knowledge of LTP risk factors in previous studies and perform conjoint analysis individual case data and common conclusions through knowledge graph; interpretatively predict the LTP risk and the high-risk LTP locations through link prediction; accurately predict the ablation safety margin required for different tumor parts through graph neural network, and achieve highly conformal thermal field planning based on different invasion capabilities to minimize the LTP risk of HCC. The project leverages tumor multi-modal imaging and prior knowledge as the entry point, performs highly conformal planning of the ablation thermal field through artificial intelligence technology, and provides a new method for precise ablation.

Inclusion Criteria:

1. Pathologically confirmed primary hepatocellular carcinoma
2. Undergo curative ablation
3. With complete clinical information and pre- and post-operative imaging information

Exclusion Criteria:

1. Undergo palliative ablation
2. Lack of clinical or imaging information
3. Age less than 18 years

HCC patients receiving ablation therapy