Methods: A total of 269 patients treated with TACE between January 2016 and June 2020 were enrolled from two medical centers and divided into three cohorts: a training cohort (n = 126), an internal validation cohort (n = 84), an external test cohort (n = 59). Radiological features of the tumor area were extracted on multiple phases of CT before treatment, and progression-free survival (PFS) was predicted through the least absolute shrinkage and selection operator Cox (LASSO-Cox) regression algorithm. Overall survival (OS) was evaluated using the Kaplan-Meier curve. Genetic and tumor microenvironment differences related to the prognosis of TACE were analyzed in 23 patients with CT from The Cancer Genome Atlas (TCGA). Single-cell data from the comprehensive Gene Expression Omnibus (GEO) public database were used to explore the expression of different genes in cells. 

 Results: The area under the receiver operating characteristic curves of the delta radiomic model for predicting 2-year PFS in TACE-treated patients were 0.812, 0.720, and 0.807 in the training, internal validation, and external test cohorts, respectively. The Rad-score was calculated based on radiomic features selected through LASSO. The Rad-score stratified patients into high- and low-risk groups, with significant differences in OS across all cohorts. The high-risk group had a higher immune score. It is worth noting that in the immune infiltration analysis, there were significant differences in B cells, resting CD4 memory T cells, active natural killer cells, and mast cells among different risk groups. 

 Conclusions: The delta radiomics model based on multiphase enhanced CT accurately predicts prognosis and reflects tumor heterogeneity in HCC patients treated with TACE, and contributes to treatment planning.