Deep Segmentation Feature-Based Radiomics Improves Recurrence Prediction of Hepatocellular Carcinoma

Objective and Impact Statement . This study developed validated a deep semantic segmentation feature-based radiomics (DSFR) model based on preoperative contrast-enhanced computed tomography (CECT) combined with clinical information to predict early recurrence (ER) of single hepatocellular carcinoma (HCC) after curative resection. ER prediction is great significance the therapeutic decision-making surveillance strategy HCC. Introduction important for However, it cannot currently be adequately determined. Methods Totally, 208 patients HCC resection were retrospectively recruited into model-development cohort ( n = 180 ) an independent validation 28 ). DSFR models different CT phases developed. The optimal was incorporated establish DSFR-C model. An integrated nomogram Cox regression established. signature used stratify high- low-risk groups. Results A portal phase-based selected as (area under receiver operating characteristic curve (AUC): development cohort, 0.740; 0.717). achieved AUCs 0.782 0.744 in cohorts, respectively. In C-index 0.748 0.741 time-dependent 0.823 0.822, respectively, recurrence-free survival (RFS) prediction. RFS difference between risk groups statistically significant P < 0.0001 0.045 respectively). Conclusion CECT-based can resection, its combination further improved performance