Abstract
BACKGROUND: Hepatectomy is the primary curative treatment for early-stage hepatocellular carcinoma (HCC). However, post-operative early recurrence remains a significant challenge. The existing predictive approaches based on clinicopathological features lack sufficient accuracy. This study aimed to develop a deep learning (DL) framework using contrast-enhanced ultrasound (CEUS) to preoperatively predict early recurrence in patients with HCC undergoing hepatectomy. PATIENTS AND METHODS: In this retrospective study, a total of 115 patients with early-stage HCC who underwent preoperative CEUS were randomly divided into training (n=75) and validation (n=40) cohorts. Four DL models (CEUS-AP, CEUS-PP, CEUS-LP, and CEUS-MP) were developed using single-phase and multiphase CEUS cines. The CEUS-MP model, which integrated the arterial, portal venous, and late phases, was further combined with the clinical variables to construct a nomogram. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. RESULTS: The CEUS-MP model demonstrated superior predictive performance, with AUCs of 0.922 and 0.840 in the training and validation cohorts, respectively, significantly outperforming single-phase models (all P < 0.05). The combined nomogram achieved AUCs of 0.945 and 0.871 in the training and validation cohorts, respectively, with a high sensitivity (88.9% and 83.0%, respectively) and specificity (98.1% and 82.5%, respectively). Decision curve analysis confirmed the nomogram's clinical utility for threshold probabilities >30%. Visualization maps highlighted heterogeneous enhancement patterns as the key predictive features. CONCLUSION: The DL-based CEUS framework, particularly when integrated with clinical variables, provides a noninvasive and accurate tool for the preoperative prediction of early recurrence in patients with HCC undergoing hepatectomy.