Molecular Subtyping of Diffuse Gliomas Using Magnetic Resonance Imaging: Comparison and Correlation Between Radiomics and Deep Learning

Abstract

The 2016 WHO classification of diffuse gliomas is significant. The aim of this study was to establish comprehensive predictive models based on preoperative multi-parametric MRI. A total of 1,016 diffuse glioma patients were retrospectively collected from Beijing Tiantan Hospital, China. The patients were randomly divided into a training set (n = 780) and an internal validation set (n = 236). According to the 2016 WHO classification, the diffuse gliomas can be categorized according to four binary classification tasks (tasks I—IV). Predictive models based on radiomics and deep convolutional neural networks (DCNN) were developed respectively, and their performances were compared using receiver operating characteristic (ROC) curves. Additionally, the radiomics and DCNN features were visualized and compared based on t-distributed stochastic neighbor embedding technique and Spearman’s rank-correlation test. In the training set, areas under the curves (AUCs) of the DCNN models (ranging from 0.99—1.00) outperformed those of the radiomics models in all conducted tasks. In the independent validation set, the AUCs of the DCNN models outperformed those of the radiomics models in tasks I, II, and III. DCNN features demonstrated superior discriminative capability compared to radiomics features based on feature visualization analysis, and their general correlations were weak. In conclusion, both the radiomics and DCNN models could preoperatively predict the molecular subtypes of diffuse gliomas, but the latter models performed better in most circumstances. Funding Statement: This study was supported by the National Natural Science Foundation of China (No. 81601452). Declaration of Interests: No conflict of interest to declare. Ethics Approval Statement: Ethical approval for this retrospective study was received from the institutional review board of the Beijing Tiantan Hospital.