A Deep Learning Nomogram for the Prediction of Early Recurrence in Hepatocellular Carcinoma After Curative Surgery

Purpose: The accurate prediction of post-hepatectomy early recurrence in patients with hepatocellular carcinoma (HCC) is crucial for decision-making regarding postoperative adjuvant treatment and monitoring. We aimed to develop and validate a deep-learning (DL) nomogram based on MRI for predicting early recurrence in HCC after curative resection. Methods: We retrospectively included 285 HCC patients who underwent Gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced MRI within one month before curative resection. Deep features were extracted from images of the arterial phase (AP), portal venous phase (PVP), and hepatobiliary phase (HBP) using VGGNet-19. Pearson’s correlation was firstly used to exclude redundant features. Three feature selection methods and five classification methods were combined to construct the DL signature. Univariate and multivariate logistic regression analyses were used to identify the independent risk factors for the early recurrence, which were incorporated into the DL nomogram. Results: Microvascular invasion (P = 0.039), tumor number (P = 0.001), and three-phase-based DL signature (P<0.0001) were independent risk factors for early recurrence. The DL nomogram integrating the DL signature and clinical risk factors outperformed the clinical nomogram which combined clinical risk factors, in the training set (AUC: 0.949 vs. 0.751; P<0.0001) and validation set (AUC: 0.908 vs. 0.712; P = 0.002). Excellent calibration was achieved for the DL nomogram in both training and validation sets. Decision curve analysis confirmed the clinical usefulness of the DL nomogram. Conclusions: The proposed DL nomogram was superior to the traditional clinical nomogram in predicting early recurrence for HCC patients after curative resection.