Radiomics, a field that extracts quantitative features from medical images, plays a crucial role in predicting the efficacy of treatments like Transcatheter Arterial Chemoembolization (TACE) for hepatocellular carcinoma (HCC). Recent advancements have shown that combining radiomics with clinical and genetic data enhances predictive accuracy. This integration has significantly influenced current diagnostic and treatment strategies for HCC. Studies have demonstrated that these combined models provide more precise predictions, leading to improved patient outcomes. This review summarizes recent advances and current challenges in radiomics-based combined models for predicting outcomes after TACE in HCC. It systematically outlines key breakthroughs, including multimodal data fusion, improved methods for quantifying intratumoral heterogeneity, and enhanced model predictive performance. It also examines persistent bottlenecks: dataset-dependent feature standardization, limited model generalizability, clinical annotation bias, and high computational costs. The goal of this review was to guide researchers in addressing these technical barriers and optimizing model architectures, to provide evidence for individualized clinical decision-making, and to accelerate the translation of radiomics combined models from basic research into standardized clinical practice—ultimately improving post-TACE outcomes and long-term quality of life for HCC patients.