Predicting the recurrence of hepatocellular carcinoma (HCC) is critical for effective clinical management and improved patient survival rates. Radiomics, a method leveraging advanced image analysis to extract quantitative features from medical images, has emerged as a promising tool in this domain. This meta-analysis investigated the efficacy of radiomics models in predicting HCC recurrence and rigorously assessed their methodological quality to determine avenues for improving their predictive performance, specifically focusing on enhancing the concordance index (C-index) as a key metric of model accuracy.
This study systematically reviewed major databases including Cochrane Library, Web of Science, PubMed, and Embase, encompassing research up to July 11, 2023. The analysis included 49 studies, each evaluated for methodological robustness using the Radiomics Quality Score (RQS). The predictive capabilities of radiomics models were compared against traditional clinical models and combined models that integrate both clinical parameters and radiomics signatures. The primary performance metric was the C-index, alongside sensitivity and specificity, across different imaging modalities: computed tomography (CT), magnetic resonance imaging (MRI), ultrasound/sonography (US), and contrast-enhanced ultrasound (CEUS).
The findings revealed that radiomics models generally outperformed clinical models in predicting HCC recurrence across most imaging techniques, as evidenced by superior C-index values on validation cohorts. Specifically, radiomics models demonstrated the following C-index ranges: CT (0.747, 95% CI: 0.70-0.79), MRI (0.788, 95% CI: 0.75-0.83), and CEUS (0.763, 95% CI: 0.60-0.93). In contrast, clinical models alone achieved a C-index of 0.671 (95% CI: 0.65-0.70). Ultrasound-based radiomics models were an exception, showing a lower C-index (0.560, 95% CI: 0.53-0.59). Notably, the integration of radiomics with clinical features in combined models further improved predictive performance. Combined models achieved even higher C-indices: CT (0.790, 95% CI: 0.76-0.82), MRI (0.826, 95% CI: 0.79-0.86), and US (0.760, 95% CI: 0.65-0.87), except for CEUS combined models which showed a C-index of 0.707 (95% CI: 0.44-0.97).
In conclusion, this meta-analysis underscores the potential of radiomics as a valuable approach for predicting HCC recurrence. Integrating radiomics signatures with clinical data significantly enhances predictive accuracy, demonstrating improved C-index values across CT, MRI, and ultrasound imaging. While these results are promising, further research is essential to refine radiomics methodologies, validate these findings in larger, multi-center studies, and ultimately optimize the clinical translation of radiomics-based HCC recurrence prediction models.
