Objective: Ki-67 is a well-established biomarker for tumor aggressiveness and poor prognosis in hepatocellular carcinoma (HCC). A reliable non-invasive method for preoperative Ki-67 assessment is clinically needed for risk stratification and individualized treatment. This study aimed to develop and validate a prediction model integrating triphasic contrast-enhanced CT radiomics with clinical features for preoperative Ki-67 expression status in HCC. Materials and Methods: This retrospective dual-center study enrolled 200 patients with 213 pathologically confirmed HCC lesions, Ki-67 expression was dichotomized as high (Ki-67 > 20%) and low (≤ 20%) based on established clinical criteria. Radiomic features were extracted from arterial, portal venous, and delayed phases. After rigorous feature selection, logistic regression was used to construct single-phase models, a multi-phase radiomics fusion model, a clinical model, and a combined clinical-radiomics fusion model. Performance was assessed by area under the curve, net reclassification index, integrated discrimination improvement, and decision curve analysis. Results: The combined fusion model showed robust discrimination, with AUCs of 0.866 and 0.824 in the training and internal test sets, respectively. In independent external validation (n=64), it achieved an AUC of 0.829 (95% CI: 0.709– 0.948), significantly outperforming the arterial phase model (AUC=0.713, P =0.031) and the radiomics-only fusion model (AUC=0.722, P =0.031). NRI and IDI confirmed significant incremental value (NRI=0.374, P =0.005; IDI=0.191, P < 0.001), and DCA demonstrated superior clinical net benefit. Conclusion: The fusion model integrating multi-phase CECT radiomic features with clinical indicators provides an effective, non-invasive tool for preoperative prediction of Ki-67 expression in HCC. It may facilitate risk stratification and inform individualized treatment planning in clinical practice. Plain Language Summary: Question : A reliable non-invasive method for preoperative Ki-67 assessment in HCC is needed for risk stratification and personalized treatment. Findings : A model combining triphasic CT radiomics and clinical features effectively predicted Ki-67 expression, showing robust performance (AUC=0.829) in external validation. Clinical Relevance Statement : This CT-based model provides an accessible, non-invasive tool to preoperatively assess tumor proliferation, potentially aiding individualized treatment planning and improving prognosis in HCC patients. Diagram of a five-step process: tumor segmentation, feature extraction, selection, model building, evaluation.The diagram outlines a five-step process for creating a model to predict Ki-67 non-invasively before surgery. Step 1 involves segmenting the tumor using liver images and CT scans. Step 2 extracts features like statistics, texture, shape and high-order elements, combined with clinical data. Step 3 selects features using methods like correlation coefficient, mMRMR, KBEST and LASSO. Step 4 constructs models integrating radiomics, clinical and clinical-radiomics data via logistic regression. Step 5 evaluates the model with graphs showing true positive rate vs. false positive rate and net benefit vs. threshold probability. Data from 200 patients with 213 HCC lesions, using triphasic CECT and radiomics plus clinical features, supports clinical-radiomics fusion. External validation shows AUC 0.829, surpassing AP with p=0.031, NRI=0.374 and IDI=0.191. The model aids in risk stratification and personalized treatment. Keywords: hepatocellular carcinoma, tomography, x-ray computed, radiomics, ki-67 antigen, predictive model, clinical features
Huang et al. (Fri,) studied this question.