Identifying Glioblastoma Imaging Signatures using a multiscale approach

• Applicant: Dr. Salvador Castaneda Vega
• Subject Area: Nuclear Medicine, Radiotherapy, Radiobiology ; Hematology, Oncology ; Medical Informatics and Medical Bioinformatics ; Molecular and Cellular Neurology and Neuropathology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 552093537

Differentiating between tumor tissue and post-radiation inflammatory changes using combined positron emission tomography and magnetic resonance imaging (PET/MRI) after radiotherapy is a critical and challenging task that very often determines the course of treatment and outcome of glioma patients. Machine learning models have been developed to identify these tissue characteristics. However, the complexity of the tumor microenvironment and limitations in clinical practice make it difficult to train these models accurately on a voxel-by-voxel basis using biopsy samples. We have previously shown that preclinical PET/MRI data can be used to predict imaging phenotypes in humans. For this reason, we aim to develop a machine learning workflow that uses specifically designed ground truth of the whole tumour at micro and macro scales. Our first goal is to successfully transfect tumor cells with two reporter genes to express two specific imaging biomarkers: one for tumor cell-membrane identification using PET in vivo and one for tumor nucleus identification using light sheet microscopy ex vivo. Light-sheet microscopy allows verification and exact localization of the tumor at the cellular level. The reporter gene for in vivo PET on the other hand, expresses a protein on the surface of tumor cells that can be specifically targeted by a radiotracer. Our second goal is to develop two machine learning models. One is specifically focused on identifying tumours and inflammation microscopically. The targets found by this first cellular identification will be used to train a second machine learning model using preclinical in vivo PET/MRI to identify tumor and inflammatory regions following ablative radiotherapy. We are confident that this combined approach will not only push the boundaries of current clinical diagnostic capabilities but also equip us with a machine learning methodology that excels in tissue identification with unparalleled accuracy.

• DFG Programme: Research Grants
• Co-Investigators: Professor Dr. Stefan Bauer; Professorin Dr. Bettina Weigelin