Classification of MGMT Promoter Methylation in Glioblastoma Patients Using EfficientNet-RNN

EasyChair Preprint 9612

Abstract

Glioblastoma (GBM) is a deadly malignant brain tumor. The biggest threat is the very low survival time, as it reduces the chances of administering the right treatment. For patients with GBM, time is extremely precious and an accurate prognosis is essential. The problem that is prevalent is the invasiveness in the procedures required to identify if the standard treatment will be effective for the patient. Due to lack of a direct indication in the MRI, doctors are forced to perform surgery to identify the genotypic indication of the prognosis. Promoter methylation of the MGMT biomarker indicates a better response to chemotherapy and longer survival times in patients with GBM. While MGMT itself is not identifiable from images, this project attempts to predict its presence from multimodal MRI data acquired from Radiological Society of North America (RSNA). The proposed method trains an EfficientNet-RNN model for each of the 4 modalities of the MRI and fuses their individual outputs to produce an AU-ROC score of 0.5876, which is an improvement on the individual values.

Keyphrases: Glioblastoma, MGMT, MRI, brain, deep learning, genotype, multimodal

@booklet{EasyChair:9612,
  author    = {S Manoj and T V Raghavasimhan and S Umamaheswari and D Sangeetha},
  title     = {Classification of MGMT Promoter Methylation in Glioblastoma Patients Using EfficientNet-RNN},
  howpublished = {EasyChair Preprint 9612},
  year      = {EasyChair, 2023}}