Machine-learning-based classification of Glioblastoma in multiparametric MRI

Glioblastoma (GBM) is the most frequent and lethal primary brain cancer. Due to its therapeutic resistance and aggressiveness, clinical management is challenging. This study aims to develop a machine-learning-based classification method using radiomic features of multiparametric MRI to correctly identify high grade (HG) and low grade (LG) GBMs. Multiparametric MRI of 50 patients with GBM, 25 HG and 25 LG, were used. Each patient has T1, contrast-enhanced T1, T2 and FLAIR MRI, as well as provided tumor contours. These tumor contours were used to extract features from the multiparametric MRI. Once these features have been extracted, the most significant and informative features were selected to train random forests to differentiate HG and LG GBMs while varying feature correlation limits were applied to remove redundant features. Then leave-one-out cross-validation random forests were applied to the dataset to classify HG or LG GBMs. The prediction accuracy, receiver operating characteristic (ROC) curves, and area under the curve (AUC) were obtained at each correlation limit to evaluate the performance of our machine-learning-based classification. The best performing parameters predicted on an average, a prediction accuracy of 0.920 or 46 out of 50 patients, 22/25 for HGG and 24/25 for LGG, consistently with an AUC of 0.962. We investigated the process of distinguishing between HG GBM and LG GBM using multiparametric MRI, radiomic features, and machine learning. The result of our study shows that grade of GBM could be predicted accurately and consistently using the proposed machine-learning-based radiomics approach.