Predicting IDH Mutation and MGMT Methylation Status in Glioma Patients at the Voxel Level using CEST-Based Deep Learning

Motivation: Predicting glioma subtypes based on molecular profiles is crucial for treatment decisions and predicting survival rates. Goal(s): We proposed a CEST-based deep learning method to predict IDH mutation and MGMT methylation status in glioma patients at the voxel level. Approach: 86 patients were recruited for CEST experiments on 3T MRI scanner. A CEST-based deep learning method, composed of a 1D convolutional neural network, was proposed for different types of status prediction at the voxel level. The confusion matrix and ROC were conducted to evaluate the performance of the proposed method. Results: Our method achieves higher accuracy compared to existing CEST-based prediction methods. Impact: The proposed method may facilitate the application of CEST MRI in the diagnosis of glioma.