Radiopathomic Modeling of Glioma Heterogeneity Throughout a Patient's Disease Trajectory - PROJECT ABSTRACT The infiltrative nature of gliomas makes it difficult to define and treat the full extent of tumor using conventional magnetic resonance imaging (MRI). Both standard of care and experimental treatment can further cloud traditional imaging signatures, hindering accurate treatment delivery and response assessment. Accurate methods for delineating the true extent of tumor are therefore desperately needed for directing treatments and evaluating their effectiveness. Radio-pathomic mapping algorithms have emerged as promising tools to identify microscopic infiltrating tumor cells and within lesion pathological heterogeneity. This proposal aims to externally validate and integrate radio-pathomic models developed using autopsy and biopsy samples from two different institutions, the Medical College of Wisconsin (MCW) and the University of California, San Francisco (UCSF). The combined models will be further refined and developed with the inclusion of clinical information, novel deep learning approaches, and finally the completion of a prospective multi-site clinical validation study. In this proposal, aim 1 will focus on the external validation of radio-pathomic maps and their ability to generalize when applied to external datasets acquired at various time points during a patient’s therapeutic trajectory. Performance for identifying pathological measures of proliferation, cellularity, and molecular subtype will be assessed in newly diagnosed patients. At tumor recurrence, each model will be evaluated to determine how well they differentiate treatment effects from true recurrent tumor. Additionally, we will compare radio- pathomic maps to metabolic measurements obtained with MR spectroscopy. Aim 2 will unify and refine the radio- pathomic models by leveraging comprehensive datasets from both institutions. The impact of time between the last MRI scan and death, timing since initial diagnosis, and other clinically prognostic variables on model performance over time and generalizability to a patient's full trajectory will be investigated. Different strategies for retraining models and incorporating data acquired from over the disease trajectory including deep learning will also be explored. Aim 3 will test the ability of optimized radio-pathomic models to correctly identify regions of glioma infiltration at diagnosis and tumor recurrence in a multi-site prospective clinical validation study, a critical step for broader adoption and clinical translation. Successful completion of this proposal will advance the field of radio-pathomics for glioma characterization, offering non-invasive tools to aid in treatment planning, monitoring treatment response, and guiding personalized therapies. The outcomes will contribute to the development of a robust and clinically applicable framework for glioma management, ultimately improving patient outcomes and quality of life.
