Prediction early recurrence of hepatocellular carcinoma eligible for curative ablation using a Radiomics nomogram

• Published in: Cancer imaging Vol. 19; no. 1; pp. 21 - 12
• Main Authors: Yuan, Chunwang, Wang, Zhenchang, Gu, Dongsheng, Tian, Jie, Zhao, Peng, Wei, Jingwei, Yang, Xiaozhen, Hao, Xiaohan, Dong, Di, He, Ning, Sun, Yu, Gao, Wenfeng, Feng, Jiliang
• Format: Journal Article
• Language: English
• Published: England BioMed Central Ltd 26.04.2019; BioMed Central; BMC
• Subjects: Ablation; Ablation (Surgery); Ablation techniques; Aged; Algorithms; Biomarkers; Cancer; Cancer recurrence; Carcinoma; Carcinoma, Hepatocellular - diagnostic imaging; Carcinoma, Hepatocellular - epidemiology; Carcinoma, Hepatocellular - pathology; Care and treatment; CAT scans; Colorectal cancer; Computed tomography; Datasets; Feature extraction; Female; forecasting; Hepatocellular carcinoma; Humans; Image contrast; Image enhancement; Liver cancer; Liver Neoplasms - diagnostic imaging; Liver Neoplasms - epidemiology; Liver Neoplasms - pathology; Male; Medical imaging; Medical imaging equipment; Medical prognosis; Medical research; Metastasis; Middle Aged; Nomograms; Pathology; Patients; Performance prediction; Prognosis; Radiomics; Recurrence; Redundancy; Regression models; Risk factors; Subgroups; Tomography; Tomography, X-Ray Computed; Training; Recurrence; Hepatocellular carcinoma; forecasting; Ablation techniques; Radiomics
• ISSN: 1470-7330; 1740-5025; 1470-7330
• DOI: 10.1186/s40644-019-0207-7

Predicting early recurrence (ER) after radical therapy for HCC patients is critical for the decision of subsequent follow-up and treatment. Radiomic features derived from the medical imaging show great potential to predict prognosis. Here we aim to develop and validate a radiomics nomogram that could predict ER after curative ablation. Total 184 HCC patients treated from August 2007 to August 2014 were included in the study and were divided into the training (n = 129) and validation(n = 55) cohorts randomly. The endpoint was recurrence free survival (RFS). A set of 647 radiomics features were extracted from the 3 phases contrast enhanced computed tomography (CECT) images. The minimum redundancy maximum relevance algorithm (MRMRA) was used for feature selection. The least absolute shrinkage and selection operator (LASSO) Cox regression model was used to build a radiomics signature. Recurrence prediction models were built using clinicopathological factors and radiomics signature, and a prognostic nomogram was developed and validated by calibration. Among the four radiomics models, the portal venous phase model obtained the best performance in the validation subgroup (C-index = 0.736 (95%CI:0.726-0.856)). When adding the clinicopathological factors to the models, the portal venous phase combined model also yielded the best predictive performance for training (C-index = 0.792(95%CI:0.727-0.857) and validation (C-index = 0.755(95%CI:0.651-0.860) subgroup. The combined model indicated a more distinct improvement of predictive power than the simple clinical model (ANOVA, P < 0.0001). This study successfully built a radiomics nomogram that integrated clinicopathological and radiomics features, which can be potentially used to predict ER after curative ablation for HCC patients.