This research aims to critically appraise the foundations, advances, and challenges of radiogenomics and machine learning (ML) in hepatocellular carcinoma (HCC), with a focus on clinical translation and future directions. Radiogenomics enables non-invasive assessment of tumor biology and the tumor microenvironment by correlating imaging features with genomic data. For instance, one study utilized contrast-enhanced CT (CECT) radiomic features to predict genomic alterations in the PI3K signaling pathway, achieving an area under the curve (AUC) of 0.733 in external validation. Another model integrating MRI radiomics and exosomal miRNAs for predicting microvascular invasion reported an AUC of 0.900. ML, particularly deep learning, has significantly enhanced image analysis capabilities. In predicting response to transarterial chemoembolization (TACE), a radiomics model (AUC 0.813) outperformed traditional CT assessment. A random forest model combining ultrasound features and serum biomarkers to predict outcomes of targeted immunotherapy in advanced liver cancer demonstrated an external validation AUC of 0.899. Although these technologies show great promise in transforming HCC management towards precision medicine—facilitating early detection, risk stratification, treatment response prediction (e.g., using FDG-PET/CT features to predict mTOR pathway activation with an AUC of 0.733), and prognosis assessment (e.g., a radiogenomics model incorporating tumor microenvironment-related genes predicting overall survival with 1–3 year AUCs of 0.81–0.87)—significant challenges remain. These include methodological issues such as lack of standardization, small sample sizes, insufficient external validation, and the “black box” nature of models affecting interpretability. Furthermore, ethical considerations (e.g., data privacy and algorithmic bias) and barriers to clinical integration (e.g., workflow adaptation and regulatory approval) must be addressed. Future progress depends on conducting prospective multi-center trials, establishing standardized imaging and data analysis pipelines, developing explainable AI models, and creating robust ethical and regulatory frameworks to ultimately translate these innovative tools from research to routine clinical practice.
Ye et al. (Sat,) studied this question.