G.L.O.W.: A Novel Hybrid Neural Network Approach for Glioblastoma Localization Using Carcinogenic Oxidative Stress Biomarkers

Voiceover

Presenter(s)

Julia Gao, Om Vegesna

Abstract or Description

Oxidative stress is a precursor to numerous malignant cancers, such as glioblastoma multiforme, the most dangerous brain cancer. However, precise identification of oxidative stress biomarkers remains incomplete.

We originally sought to map the glycolysis enzyme and oxidative stress marker, glyceraldehyde-3-dehydrogenase (GAPDH), in the brain after simulated radiation effects. We obtained high Optical Density readings post-radiation which suggest rapid cell proliferation. In addition, GAPDH expression was 311% higher, indicating the cells had become cancerous. We then developed a hybrid neural network for glioblastoma localization using our identified carcinogenic oxidative stress biomarkers for early detection of tumors as small as 8 µm. With multi-modal imaging from QUEST MRIs, the model uses 3D rendering to identify tumor location, size, and growth rate.

Our hybrid architecture processes 100 epochs of 64 batches efficiently and accounts for genomic data to uncover new pathways for development. Additionally, this approach of inferring cancer from oxidative stress allows us to detect smaller or minuscule tumors, granting strong functionality for this new method of cancer diagnosis. Our model, the Glioblastoma Localization and Optimization Workbench (G.L.O.W.) is 96% accurate and consistent across all brain regions according to the Precision Report. G.L.O.W. could be adapted for the whole body and be refined with clinical trials and more training. Our approach enables us to locate tumors with high accuracy for early treatment, all while being cost-effective and time-efficient. Our results are encouraging, potentially exhibiting a new method of glioblastoma diagnosis.