Deep Learning Predicts Overall Survival of Patients with Unresectable Hepatocellular Carcinoma Treated by Transarterial Chemoembolization Plus Sorafenib: A Multicenter Study

Abstract

Background: Biomarkers or models which provide accurate prognosis predictions in patients with unresectable hepatocellular carcinoma (HCC) are still lacking. The aim of this study was to develop and validate a deep learning-based overall survival (OS) prediction model in patients with unresectable HCC treated with transarterial chemoembolization (TACE) plus sorafenib. Methods: This retrospective multicenter study consisted of 201 patients with treatment-naive, unresectable HCC who were treated with TACE plus sorafenib from 2011 to 2016. The primary endpoint was overall survival (OS). Data from 120 patients were used as the training set for model development. A deep learning signature was constructed using the deep image features from preoperative contrast-enhanced computed tomography (CECT) images in the arterial and portal phase. An integrated nomogram was built using Cox regression by combining the deep learning signature and clinical features. For comparison, the clinical features were used to build a clinical nomogram. The deep learning signature and nomograms were also externally validated in an independent validation set of 81 patients. C-index was used to evaluate the performance of OS prediction. Findings: The median OS of the entire set was 19.2 months (95% CI: 17.7-20.7) and no significant difference was found between the training and validation cohort (18.6 months (95% CI: 16.2-21.2) vs. 19.5 months (95% CI: 17.8-21.9), P=0.45). The deep learning signature achieved good prediction performance with a C-index of 0.717 (95% CI: 0.709-0.726) in the training set and 0.714 (95% CI: 0.702-0.727) in the validation set. The integrated nomogram was built using the deep learning signature, alanine aminotransferase (ALT) values and Barcelona Clinic Liver Cancer (BCLC) stage, and it showed significantly better prediction performance than the clinical nomogram in the training set (C-index: 0.739, 95% CI: 0.731-0.748 vs. 0.664, 95% CI: 0.654-0.673, P=0.002) and validation set (C-index: 0.730, 95% CI: 0.717-0.742 vs. 0.679, 95% CI: 0.667-0.691, P=0.023). Interpretation: The deep learning signature could serve as a novel biomarker for OS prediction, and it provided significant added value to clinical features in the development of an integrated nomogram which may act as a potential tool for individual prognosis prediction and identifying potential patients with HCC who may benefit from the combination therapy of TACE plus sorafenib. Funding Statement: This study was supported by the National Natural Science Foundation of China (81901847) (81771945) (81971713) (61801474) (81871439), the Jiangsu Medical Innovation Team (CXTDB2017006), the Natural Science Foundation of Jiangsu Province (BK20190177), the Natural Science Foundation of Zhejiang Province (LZ18H180001), and Guangdong Provincial Key Research and Development Program (2019B010152001). Declaration of Interests: The authors disclose no conflicts. Ethics Approval Statement: The study was approved by the Institution Ethics Review Boards of the three mentioned centers. The need for informed consent was waived due to the retrospective nature of the study.