Abinand Rejimon
BME PhD Defense Presentation

Date: 2026-03-27
Time: 2:00 pm 
Location / Meeting Link: Emory Clinic Building A Room A1303 / https://zoom.us/j/5582282841?omn=93492209298

Committee Members:
Hyunsuk Shim, PhD (Advisor); Lee A.D. Cooper, PhD (Co-Advisor); Brent D. Weinberg, MD/PhD; Candace Fleischer, PhD; Bree Eaton, MD; John Oshinski, PhD


Title: An Integrated Approach for Optimizing Spectroscopy-Guided Radiotherapy Strategies in Pediatric High-Grade Glioma

Abstract:
Pediatric high-grade glioma (pHGG) is the leading cause of brain tumor-related death in children. Standard radiotherapy (RT) planning relies on contrast-enhanced T1-weighted and T2-FLAIR MRI, but many pHGG are non-enhancing, forcing clinicians to depend on nonspecific T2-FLAIR signal alone. This limitation produces treatment volumes that miss infiltrative tumor margins while exposing healthy brain to unnecessary radiation, contributing to high recurrence rates. Whole-brain spectroscopic MRI (sMRI) maps endogenous metabolites, including elevated choline and reduced N-acetylaspartate, that reveal tumor infiltration independent of contrast enhancement. Two barriers have limited clinical adoption: insufficient validation of sMRI in pediatric tumor biology and the burden of expert artifact review required before sMRI-derived volumes can be used clinically. This dissertation addresses both barriers. A retrospective analysis of pHGG patients treated at Emory University established that standard imaging is unreliable for post-treatment response assessment. sMRI-guided stereotactic biopsies then validated metabolite maps against pediatric tumor biology in cases where conventional imaging was insufficient to guide clinical decision-making. An ongoing prospective trial, NCT04908709, integrated sMRI into proton beam RT planning with serial acquisitions across the treatment course. Longitudinal metabolic maps detected recurrence before standard MRI changes were apparent, and preliminary results showed twice the median overall survival of a matched standard-imaging cohort. To address the artifact review burden, NN-Artifact, a mixture-of-experts deep learning pipeline, was developed by coupling automated T2-FLAIR lesion segmentation with two spatially specialized convolutional neural networks targeting the tumor and surrounding brain regions. Applied without retraining to the prospective pHGG cohort, NN-Artifact produced spatially coherent metabolic maps across all patients, eliminating up to four hours of expert review per case. Together, these studies establish a clinically validated and computationally scalable framework for sMRI-guided RT planning and treatment monitoring in pHGG, and provide the foundation for deploying this technology across institutions and national clinical trials.