Abstract Purpose Our review aims to assess the role of DCE-MRI in radiogenomics, with a focus on ML-based predictive models. Methods A comprehensive search was conducted in the PubMed, Scopus, Web of Science, Embase, and ProQuest databases via key terms related to “DCE-MRI”, “Machine learning”, “Deep learning”, “Glioblastoma”, and “Radiogenomics”. The extracted data include dataset descriptions, tumor classifications, MRI sequences, predicted genes, ML approaches, performance metrics, results, external validation status, and interpretability methods. Results Out of the 432 identified articles, five studies involving 621 GBM patients met the inclusion criteria. Studies have employed DCE-MRI alongside other sequences, including T1-weighted, T2-weighted, and FLAIR. The predicted genetic markers included IDH1 mutation, EGFR mutation, MGMT methylation, and key cell cycle regulatory molecules such as RTK, p53, and RB1. Forty percent of the studies utilized DL approaches, with only one study performing external validation. Performance metrics, such as the area under the receiver operating characteristic curve, were reported across studies. Conclusion Our findings underscore the promising role of DCE-MRI in GBM radiogenomics, particularly when combined with ML approaches. However, the lack of external validation limits the broader applicability of these models. Future research should focus on larger, multicenter studies with thorough validation methods to enhance the clinical relevance and advance precision medicine in GBM management.
Goodarzi et al. (Wed,) studied this question.