NIMG-55. A COMPARISON OF THREE DIFFERENT DEEP LEARNING-BASED MODELS TO PREDICT THE MGMT PROMOTER METHYLATION STATUS IN GLIOBLASTOMA USING BRAIN MRI

Abstract GBM is the most common primary malignant brain tumor in adults. O6-methylguanine DNA methyltransferase (MGMT) promoter methylation status an important prognostic biomarker that predicts response to temozolomide and guides treatment decisions. At present, only reliable way determine MGMT through analysis of tissues. Given limitations complications histology-based methods, imaging-based approach for non-invasively prediction beneficial. This study aimed develop compare three different deep-learning-based approaches predicting non-invasively. We obtained 576 T2 weighted images (T2WIs) with their corresponding masks from, The Brain Tumor Segmentation (BraTS) 2021 datasets. Dataset was split into five folds at patient's level stratified by perform a 5-fold cross-validation. developed models: voxel-wise, slice-wise, whole-brain. For voxel-wise classification, methylated unmethylated were made 1 2 0 background, respectively. converted each T2WI 32x32x32 patches. trained 3D-Vnet model segmentation. After inference, we constructed whole volume based on patch's coordinates. final majority voting between predicted voxel values biggest connected component. slice-wise object detection prediction; then, used prediction. whole-brain approach, 3D Densenet121 Whole-brain, accuracy 65.42%(SD 3.97%), 61.37%(SD 1.48%), 56.84%(SD 4.38%) found across whole-brain, deep learning approaches, effective BraTS dataset.