Automated grade prediction of glioma patients based on magnetic resonance imaging and a random forests approach

An automated brain tumour classification method is presented which is able to distinguish between low-grade and high-grade glioma on conventional MRI scans. Per patient, 208 quantitative features are extracted from a manually annotated brain tumour database of 274 patients. These features were then used to train a Random Forests classification algorithm. We achieved a high-grade prediction sensitivity of 85.5% and specificity of 83.3%, with a global accuracy of 85.0%