Ki-67 is a critical prognostic marker for hepatocellular carcinoma (HCC), yet its clinical assessment relies on invasive biopsy. This study aimed to develop a deep learning framework using contrast-enhanced ultrasonography (CEUS) for non-invasive Ki-67 expression assessment and prognostic prediction in HCC. We retrospectively collected CEUS videos and clinical data of 456 HCC patients from 25 institutions, divided into a development cohort (288 patients, split into training and validation sets) and an external test cohort (168 patients with complete prognosis data). A channel-separated convolutional-based multimodal model (CECMM) integrating CEUS features and clinical characteristics was constructed, with its performance compared to alternative methods; the derived CECMMScore was used for prognostic stratification. The CECMM model outperformed comparative approaches, achieving accuracies of 89.50% (95% CI 85.50%-93.50%), 78.16% (95% CI 67.82%-86.21%), and 75.60% (95% CI 69.05%-82.16%), alongside AUCs of 0.93 (95% CI 0.89-0.96), 0.81 (95% CI 0.72-0.89), and 0.83 (95% CI 0.76-0.89) in the training, validation, and external test cohorts, respectively. Additionally, the CECMMScore was significantly associated with progression-free survival (log-rank p=0.0456), intrahepatic recurrence survival (p=0.0122), and early recurrence survival (p=0.0103) in the external test cohort. In conclusion, the proposed CEUS-based deep learning model achieves favorable performance in non-invasive Ki-67 quantification, providing a clinically valuable non-invasive indicator for HCC prognosis.
Zou et al. (2026) studied this question.