Radiomics-based Machine Learning to Predict the Recurrence of Hepatocellular Carcinoma: A Systematic Review and Meta-analysis

Rationale and Objectives

Recurrence of hepatocellular carcinoma (HCC) is a major concern in its management. Accurately predicting the risk of recurrence is crucial for determining appropriate treatment strategies and improving patient outcomes. A certain amount of radiomics models for HCC recurrence prediction have been proposed. This study aimed to assess the role of radiomics models in the prediction of HCC recurrence and to evaluate their methodological quality.

Materials and Methods

Databases Cochrane Library, Web of Science, PubMed, and Embase were searched until July 11, 2023 for studies eligible for the meta-analysis. Their methodological quality was evaluated using the Radiomics Quality Score (RQS). The predictive ability of the radiomics model, clinical model, and the combined model integrating the clinical characteristics with radiomics signatures was measured using the concordance index (C-index), sensitivity, and specificity. Radiomics models in included studies were compared based on different imaging modalities, including computed tomography (CT), magnetic resonance imaging (MRI), ultrasound/sonography (US), contrast-enhanced ultrasound (CEUS).

Results

A total of 49 studies were included. On the validation cohort, radiomics model performed better (CT: C-index = 0.747, 95% CI: 0.70–0.79; MRI: C-index = 0.788, 95% CI: 0.75–0.83; CEUS: C-index = 0.763, 95% CI: 0.60–0.93) compared to the clinical model (C-index = 0.671, 95% CI: 0.65–0.70), except for ultrasound-based models (C-index = 0.560, 95% CI: 0.53–0.59). The combined model outperformed other models (CT: C-index = 0.790, 95% CI: 0.76–0.82; MRI: C-index = 0.826, 95% CI: 0.79–0.86; US: C-index = 0.760, 95% CI: 0.65–0.87), except for CEUS-based combined models (C-index = 0.707, 95% CI: 0.44–0.97).

Conclusion

Radiomics holds the potential to predict HCC recurrence and demonstrates enhanced predictive value across various imaging modalities when integrated with clinical features. Nevertheless, further studies are needed to optimize the radiomics approach and validate the results in larger, multi-center cohorts.