Standard histopathological evaluation is no longer considered the end-point for patients' treatment decision. Genetic and molecular evaluation has shown that tumors with the same histopathological features can behave differently clinically. Some recent studies have found that imaging signatures can predict underlining genomic features such as gene expression and molecular biomarkers and also treatment allocation and treatment outcome. Other results have found correlation between radiomic features and overall survival and progression-free survival. In spite of these advancements in molecular and genetic biology, treatment of pediatric brain tumors remains with low survival rates and modest response to first line therapy. Thus, novel models are needed for better treatment of brain tumors, and new biomarkers that enable in vivo treatment follow-up. Our goal is to develop a multi-scale model to better understand pediatric brain tumors, and which data types are most predictive for diagnosis, and treatment.
What are the goals of this project?
We will use clinical, genomic, and imaging data and develop models for each data source, and then integrate each model to eventually develop a multi-scale model of pediatric brain tumors. For each separate model, we will evaluate the use of adult brain tumor data from Stanford and publicly available sources to pre-train the models.
The Children's Brain Tumor Network contributed to this project by providing access to the Pediatric Brain Tumor Atlas.