A radiomics signature associated with underlying gene expression pattern for the prediction of prognosis and treatment response in hepatocellular carcinoma

Abstract

PurposeIdentifying robust prognosis and treatment efficiency predictive biomarkers of hepatocellular carcinoma (HCC) is challenging. The purpose of this study is to develop a radiomics approach for predicting the overall survival (OS) based on pretreatment CT images and to explore the radiomic-associated key genes. MethodsPatients with pathologically or clinically proven HCC from three data sets were retrospectively included in this study. The institute internal data that received transarterial chemoembolization (TACE) treatment was used as the training set to construct the radiomics signature to predict OS by the least absolute shrinkage and selection operator COX (LASSO-COX) regression algorithms. The model was externally tested in 41 patients from The Cancer Genome Atlas (TCGA) with available CT images. Area under the receiver operating characteristics curve (AUC) and the log-rank test were used for survival analysis based on high versus low radiomics score. RNA sequencing data of TCGA and Gene Expression Omnibus (GEO) public database were used for gene expression analysis. ResultsA total of 752 patients were divided into the Radiomics cohort (n = 267), the TCGA cohort (n = 338) and GEO cohort (n = 147). The rad-score divided patients into high and low risk groups, with significant survival differences (P < 0.0001 and P = 0.0055) in the training and external test set. The AUC for 5 years’ OS were 0.730 and 0.695, respectively. Seven OS-related genes (SPP1, GJA5, GJA4, INMT, PDZD4, ALDOA and MAFG) were identified, all of which were related with TACE efficiency, except for MAFG (P greater than 0.05). ConclusionsCT-radiomics signature could effectively predict the prognosis and treatment response of HCC, which were also associated with the tumor microenvironment heterogeneity.