Abstract
BACKGROUND: Besides tumorous information, synergistic liver parenchyma assessments may provide additional insights into the prognosis of hepatocellular carcinoma (HCC). This study aimed to investigate whether 3D synergistic tumor-liver analysis could improve the prediction accuracy for HCC prognosis. METHODS: A total of 422 HCC patients from six centers were included. Datasets were divided into training and external validation datasets. Besides tumor, we also performed automatic 3D assessment of liver parenchyma by extracting morphological and high-dimensional data, respectively. Subsequently, we constructed a tumor model, a tumor-liver model, a clinical model and an integrated model combining information from clinical factors, tumor and liver parenchyma. Their discrimination and calibration were compared to determine the optimal model. Subgroup analysis was conducted to test the robustness, and survival analysis was conducted to identify high- and low-risk populations. RESULTS: The tumor-liver model was superior to the tumor model in terms of both discrimination (training dataset: 0.747 vs. 0.722; validation dataset: 0.719 vs. 0.683) and calibration. Moreover, the integrated model was superior to the clinical model and tumor-liver model, particularly in discrimination (training dataset: 0.765 vs. 0.695 vs. 0.747; validation dataset: 0.739 vs. 0.628 vs. 0.719). The AUC of the integrated model was not influenced by AFP level, BCLC stage, Child-Pugh grade, and treatment style in training (6 months p value: 0.245-0.452; 12 months p value: 0.357-0.845) and validation (6 months p value: 0.294-0.638; 12 months p value: 0.365-0.937) datasets. With a risk score of 1.06, high- and low-risk populations demonstrated significant difference for progression-free survival (p < 0.001 in both datasets). CONCLUSIONS: Combined with clinical factors, 3D synergistic tumor-liver assessment improved the efficacy prediction of HCC.