Rapid Intraoperative Glioma Profiling for MGMT Promotor Methylation Prediction using Stimulated Raman Scattering Microscopy and Deep Neural Networks

• Applicant: Dr. David Reinecke
• Subject Area: Clinical Neurology; Neurosurgery and Neuroradiology
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 521771484

This project aims to develop and evaluate a combined deep neural network that can predict the MGMT promoter methylation status in patients with diffuse glioma within a few minutes intraoperatively, on which the further treatment concept may depend. Using a portable fiber laser-based stimulated Raman scattering microscope, small tissue samples obtained from stereotactic biopsy or surgical resection will be collected intraoperatively and converted into hematoxylin and eosin-like image. After the development of a neural network specialized in image features these generated images will then be analyzed in the first step. In a second step, a genetic embedding model will be developed using publicly available genomic databases of diffuse gliomas and taking into account the WHO CNS5 molecular taxonomy with typical co-occurrences (e.g. IDH status, ATRX loss, 1p/19q codeletion and CDKN2A/B deletion). In the next step, the learned information from the images and the genomic data will be merged into a combined transformer-based model. On this basis it aims to develop a color-coded overlapping probability heatmap, on the acquired histological image previously, showing the areas relevant to methylated or non-methylated MGMT promoter status in terms of the WHO CNS5 molecular taxonomy. Finally, the project aims at clinical evaluation of the developed combined deep-learning model to verify its reproducibility.

• DFG Programme: WBP Fellowship
• International Connection: USA
• Host: Professor Dr. Daniel A. Orringer