Multimodal Imaging Biomarkers for Early Detection of Glioblastoma Progression - Project Summary: Glioblastoma (GBM) is the most common malignant primary brain tumor in adults and remains highly lethal, with poor long-term survival despite current clinical management. GBM is usually detected after neurologic symptoms prompt brain MRI, and definitive diagnosis requires histopathologic and molecular classification after biopsy or resection. GBM progression has also been strongly linked to abnormal oxygen delivery, vascular remodeling, invasion, and tissue structural changes; however, by the time GBM is radiographically apparent, many functional and morphological abnormalities may already be established. Because untreated newly diagnosed tumors can grow rapidly, there is a need to identify earlier biomarkers of GBM-associated abnormal change. Current clinical imaging relies primarily on MRI to detect suspicious brain lesions and monitor progression. MRI provides a global view of tumor burden, diffusion, perfusion, permeability, and oxygenation-related physiology, but it does not directly define local tumor biology at the microscopic level. Given the vascular and oxygenation abnormalities that have been studied in GBM progression, the retina may provide an accessible site for longitudinal assessment of related neurovascular changes. Retinal imaging has identified early structural and microvascular abnormalities in other central nervous system diseases, and emerging GBM studies suggest that retinal thinning, enlarged foveal avascular zone, and reduced microvascular density are associated with reduced survival. However, it remains unclear whether retinal structural, vascular, and oxygenation changes emerge early during GBM progression or correspond to brain MRI findings and local tumor biology. My F99 work established a foundation in multimodal optical imaging, image-based feature extraction, and machine learning analysis of cancer-associated functional and morphological features. The proposed studies require an orthotopic patient-derived mouse model because longitudinal relationships among retinal imaging, brain MRI, microscopic oxygenation, vascular remodeling, and tissue-level biology must be evaluated within an intact neurovascular system over time. Although ex vivo analyses will provide essential biological validation of imaging-derived biomarkers, they cannot capture the dynamic physiological changes in blood flow, oxygenation, metabolism, and vascular remodeling that require repeated in vivo measurements. These integrated physiological processes cannot be adequately reproduced in cell culture, organoids, computational models, or human studies alone. My F99 work established a foundation in multimodal optical imaging, image-based feature extraction, and machine learning analysis of cancer-associated functional and morphological features. In the K00 phase, I will extend this framework into neuro-oncology under the mentorship of Dr. Sava Sakadzic and Dr. Hiroaki Wakimoto, by combining retinal imaging, brain MRI, nonlinear optical microscopy, and AI-based multimodal modeling. The longitudinal and multiscale data generated in this project require data-driven approaches to identify the most informative features associated with early GBM progression. Therefore, I will integrate retinal, brain MRI, nonlinear optical microscopy, and histology-based validation features into multimodal datasets and apply unsupervised machine learning and AI-based fusion models to identify predictive biomarkers of early GBM-associated abnormal change. This K00 training will support my transition to independence in AI-enabled functional and morphological imaging biomarkers that guide future clinical workflows and potential devices for cancer detection.
