Development and validation of a combined radiomic and clinical model based on contrast-enhanced ultrasound for preoperative prediction of CK19-positive hepatocellular carcinoma.

Abstract

Purpose: We aimed to develop and validate a combined model integrating radiomic features derived from Contrast-Enhanced Ultrasound (CEUS) images and clinical parameters for preoperative prediction of CK19-positive status in hepatocellular carcinoma (HCC).

Methods: A total of 434 patients who underwent CEUS and surgical resection from January 2020 to December 2023 were included. Patients were randomly divided into a training cohort (n = 304) and a validation cohort (n = 130). Radiomic features were extracted from multiphase CEUS images, including two-dimensional ultrasound (US), arterial, portal venous, and delayed phases, and combined to derive a Radscore model. Subsequently, a Combined Model was constructed using the Radscore and clinical parameters. Model performance was assessed using calibration, discrimination, and clinical utility.

Results: Multivariate logistic regression analysis identified Radscore (OR = 10.054, 95% CI: 5.931-19.120, p < 0.001) and AFP levels > 200 ng/mL (OR = 5.027, 95% CI: 2.089-12.784, p < 0.001) as significant predictors in the combined model. The AUC (Area Under the Curve) for the Combined Model was 0.954 in the training cohort and 0.927 in the validation cohort, compared to 0.939 and 0.917 for the Radscore Model alone. Calibration curves demonstrated strong concordance between predicted and actual outcomes. Decision curve analysis (DCA) showed that both the Radscore Model and the Combined Model exhibited good net benefits across a wide range of threshold values in both the training and validation cohorts.

Conclusion: The Radscore based on CEUS, combined with the serum markers AFP > 200 ng/L to construct a Combined Model, shows good predictive performance for CK19 + hepatocellular carcinoma (HCC).