Non-invasive model for predicting future lung metastasis of hepatocellular carcinoma based on imaging heterogeneity analysis

BACKGROUND: Patients with hepatocellular carcinoma (HCC) are at high risk of lung metastasis, which is the only site significantly affecting survival among extrahepatic metastases. However, current HCC lung metastasis researches predominantly focus on biological or pathological heterogeneity through invasive biopsies, which are unsuitable for continuous and dynamic monitoring. Consequently, imaging heterogeneity should also be evaluated. METHODS: From five centers, 352 HCC cases were divided into training and test datasets. Combining backward stepwise hazard models with the least absolute shrinkage and selection operator (LASSO) method, a clinical imaging heterogeneity model (CIH-Model) was constructed using clinical indicators and imaging parameters (including tumor location, peritumor, intratumor, and enhancement pattern) and tested the model's discrimination and calibration. Patients were categorized according to CIH-Model scores, and the time to lung metastasis and overall survival (OS) were compared between categories. Multivariate regression analysis was conducted to determine whether subgroup was an independent predictor of OS. RESULTS: The areas under receiver operating characteristic curves (AUC) of the CIH-Model for predicting one-, two- and three-years lung metastasis free survival were 0.78, 0.92, and 0.87 (training) and 0.72, 0.72, and 0.71 (test), respectively, with sufficient calibration in both datasets. Subgroups separated according to the median CIH-Model score showed significant differences in time to lung metastasis (training, hazard ratio[HR]: 6.271, 95% confidence interval [CI]: 3.225-12.200, p < 0.001; test, HR: 2.862; 95% CI: 1.436-5.705, p = 0.002) and OS (training, HR: 3.723; 95% CI: 2.357-5.881, p < 0.001; test, HR: 4.003; 95% CI: 2.321-6.907, p < 0.001). Multivariate regression analysis showed that subgroup by lung metastasis risk was independently related to OS in the training (HR: 0.435; 95% CI: 0.259-0.730, p = 0.002) and test datasets (HR: 0.341; 95% CI: 0.178-0.653, p = 0.001). CONCLUSIONS: Assisted by the CIH-Model, patients with HCC at high risk of future lung metastasis could be identified, and survival risk could be distinguished for appropriate monitoring and timely prevention in these patients. CLINICAL TRIAL NUMBER: Not applicable.