JMIR Publications

ABSTRACT

Background:

Radiomic models outperform clinical data for outcome prediction in high-grade gliomas (HGG). Many machine learning (ML) radiomic models have been developed, mostly employing single classifiers with variable results. However, comparative analyses of different ML models for clinically-relevant tasks are lacking in the literature.

Objective:

We aimed to compare well-established ML learning classifiers, including single and ensemble learners, to predict clinically-relevant tasks for HGG: overall survival (OS), isocitrate dehydrogenase (IDH) mutation, O-6-methylguanine-DNA-methyltransferase (MGMT) promoter methylation, epidermal growth factor receptor (EGFR) amplification and Ki-67 expression in HGG patients, based on radiomic features from conventional and advanced MRI. Our objective was to identify the best algorithm for each task in terms of accuracy of the prediction performance.

Methods:

156 adult patients with pathologic diagnosis of HGG were included. Three tumoral regions were manually segmented: contrast-enhancing tumor, necrosis and non-enhancing tumor. Radiomic features were extracted with a custom version of Pyradiomics, and selected through Boruta algorithm. A Grid Search algorithm was applied when computing 4 times K-fold cross validation (K=10) to get the highest mean and lowest spread of accuracy. Model performance was assessed as Area Under The Curve-Receiver Operating Characteristics (AUC-ROC).

Results:

Ensemble classifiers showed the best performance across tasks. xGB obtained highest accuracy for OS (74.5%), AB for IDH mutation (88%), MGMT methylation (71,7%), Ki-67 expression (86,6%), and EGFRvIII amplification (81,6%).

Conclusions:

Best performing features shed light on possible correlations between MRI and tumor histology.