Detection of methylated MGMT using radiomic imaging: A systematic review and meta-analysis.

Abstract

e14036 Background: Gliomas is an aggressive brain tumour having a poor prognosis. The methylation status of the O6-methylguanine-DNA methyltransferase (MGMT) gene promoter serves the role of an essential biomarker for predicting patient outcomes in gliomas. Traditional methods involving methylation-specific techniques of PCR and pyrosequencing suffer from being invasive, time-consuming and expensive. Recent advancements in Artificial Intelligence (AI) and Machine Learning have facilitated a radical non-invasive prediction of MGMT methylation status. AI models, particularly deep learning algorithms, have shown promise in analysing medical imaging data, such as MRI scans, to infer MGMT methylation status. These models enable the extraction of complex features and pattern recognition from imaging data that may be inapparent. The integration of AI-based prediction methods with conventional clinical and molecular data has the potential to enhance personalized treatment strategies for GBM patients. Methods: A systematic search was conducted using PubMed, Google Scholar, and Scopus. PRISMA guidelines were followed. A boolean expression was constructed to retrieve and select articles from major medical databases. The studies that involved the application of Radiomic Imaging methods to detect potential MGMT mutation were considered for the final analysis. The R Studio package was used to evaluate the potential of the diagnostic test. The Pooled Sensitivity, Specificity, Accuracy and Area Under the Curve (AUC) were estimated to predict the diagnostic potential of Radiomic Imaging in the prediction of the MGMT mutation in Gliomas. The random effects model via the linear (mixed-effects) model framework was considered for statistical analysis. Results: A total of 15 studies with 7422 images assessed through various radiomic imaging modalities to detect MGMT mutation in gliomas were assessed through this meta-analysis. The Pooled Sensitivity and Specificity were estimated to be77.50% (71.25; 82.21, 95% CI, p< 0.0001, z = 22.918, SE = 0.03) and 71.35% ([61.72; 80.57], 95% CI, p<0.0001, z = 15.43, SE = 0.05) utilising the random effects model. The pooled area under the curve was estimated to be 0.7691(0.7097;0.8325, 95% CI, p<0.0001, z =23.48, SE = 0.0345). Radiomic imaging showed satisfactory accuracy(0.77 (0.7247;0.8275) 95% CI, p<0.0001, z =29.18, SE = 0.0245), indicating its potential as an important tool for non-invasive detection of MGMT mutations in gliomas. Conclusions: The clinical utility of the AI models has been successfully exhibited in regards to gliomas detection, as implied by the statistically significant accuracies, sensitivities and specificities. The high pooled accuracy of the AI models mirror their reliability in filling in the roles involving clinical applications. This meta-analysis has further consolidated AI’s stronghold in the realm of non-invasive cancer detection.