Comparison of machine learning classifiers to predict patient survival and genetics of GBM: towards a standardized model for clinical implementation

Radiomic models have been shown to outperform clinical data for outcome prediction in glioblastoma (GBM). However, implementation is limited by lack of parameters standardization. We aimed compare nine machine learning classifiers, with different optimization parameters, predict overall survival (OS), isocitrate dehydrogenase (IDH) mutation, O-6-methylguanine-DNA-methyltransferase (MGMT) promoter methylation, epidermal growth factor receptor (EGFR) VII amplification and Ki-67 expression GBM patients, based on radiomic features from conventional advanced MR. 156 adult patients pathologic diagnosis were included. Three tumoral regions analyzed: contrast-enhancing tumor, necrosis non-enhancing selected manual segmentation. extracted a custom version Pyradiomics, through Boruta algorithm. A Grid Search algorithm was applied when computing 4 times K-fold cross validation (K=10) get the highest mean lowest spread accuracy. Once optimal identified, model performances assessed terms Area Under The Curve-Receiver Operating Characteristics (AUC-ROC). Metaheuristic ensemble classifiers showed best performance across tasks. xGB obtained maximum accuracy OS (74.5%), AB IDH mutation (88%), MGMT methylation (71,7%), (86,6%), EGFR (81,6%). Best performing shed light possible correlations between MR tumor histology.