Can Machine Learning Predict Vessels Encapsulating Tumor Clusters Patterns and Histological Differentiation in Solitary Small (<5 cm) Hepatocellular Carcinoma?

Background: Hepatocellular carcinoma (HCC) is a major cause of cancer-related mortality worldwide. While systemic therapies have improved outcomes, predicting tumor histopathology remains difficult. Machine learning (ML) offers a promising strategy to uncover complex patterns in clinical and imaging data. This study aimed to evaluate the utility of ML in predicting histopathological features of HCC using clinical parameters and contrast-enhanced computed tomography (CT) images. Methods: We retrospectively analyzed 232 patients who underwent hepatic resection for solitary HCC (≤5 cm) without prior treatment. Histological features—vessels that encapsulate tumor clusters (VETC) and tumor differentiation—were determined. Enhanced CT images were processed using BiomedCLIP, a vision-language model pretrained on biomedical image-text pairs, to extract 512-dimensional image feature vectors. These were combined with clinical data and input into a support vector machine classifier. Five-fold cross-validation was used to evaluate performance via precision, recall, and F1-score. Results: VETC-positive tumors were significantly associated with worse disease-free and overall survival, identifying VETC as a poor prognostic factor in HCC. Clinical features alone yielded modest classification accuracy (F1 = 0.469 for VETC; F1 = 0.473 for differentiation). Incorporating image features modestly improved VETC prediction (F1 = 0.599), but did not enhance prediction of tumor differentiation. Image-based models provided limited additional value and did not outperform clinical models. Conclusion: VETC represents a histopathological marker of poor prognosis in HCC. Although ML models using routine clinical and imaging data showed limited predictive power, further refinement of image processing and data integration techniques may improve noninvasive histological prediction and support personalized treatment strategies.