Abstrakt

Glioblastoma Image Analysis using Convolutional Neural Networks to Accurately Classify Gene Methylation and Predict Drug Effectiveness

Viraj Mehta

Glioblastoma multiforme is a deadly brain cancer with a median patient survival time of 18-24 months. A single biopsy cannot provide complete assessment of the tumor’s microenvironment, making personalized care limited. 50% of the patients do not respond to the anti-cancer drug Temozolomide (TMZ) because of the over-expression of MGMT gene. Epigenetic silencing of the MGMT gene by methylation results in decreased MGMT expression, increased sensitivity to TMZ, and longer survival. The purpose of this research is to use artificial intelligence (AI) to design a low-cost platform to determine the MGMT’s methylation status and suggest non-invasive treatment plan. An AI platform is developed that uses a U-Net architecture for tumor identification in the brain MRI scans, and a ResNet-50 architecture for methylation prediction using MRI scans from the TCIA (The Cancer Imaging Archive) along with genetic data from TCGA (The Cancer Genome Atlas). The foundational software is written using Python, math libraries and TensorFlow. Image segmentation of 5000 patient brain MRI scans using a U-Net model revealed an accuracy of 90% for tumor segmentation. ResNet50 image classifier model was used for MGMT methylation status prediction. The web- platform quickly uploads the MRI scans and provides MGMT status in few seconds. The platform allows oncologists to recommend personalized treatment plans, eliminating huge time/cost investments of invasive biopsies. Patients with Positive/methylated MGMT will be receptive to chemotherapy with TMZ. Patients with unmethylated MGMT will not be sensitive to TMZ and would need additional MGMT modulation with miRNAs.