TMOD-41. Evaluating Response to Mutant IDH Inhibition in Lower Grade Gliomas Using Artificial Intelligence-Based Brain Tumor Volumetrics

Orally available inhibitors of mutant IDH (mIDHis) have emerged as a promising new therapy for patients with lower grade mIDH gliomas, yet the therapeutic response is challenging to assess objectively. Longitudinal analysis of tumor volume on MRI can provide objective metrics of tumor response, but automated tools for such analyses are lacking. Here we present an artificial intelligence (AI)-based brain tumor MRI segmentation method focused on post-treatment monitoring for lower grade gliomas. Using this approach, we demonstrate volumetric treatment response patterns in a real-world cohort of patients with lower grade gliomas treated with mIDHis. This retrospective analysis included 46 prospectively identified patients with grade 2 and 3 mIDH gliomas treated with mIDHi at a single center. We trained an AI-based brain tumor MRI segmentation model on an internal and the BraTs postoperative glioma dataset. We applied the trained model to extract tumor sub-compartment volumes from 1,323 MRI exams. We compared the average annualized tumor growth rate before and after the initiation of mIDHi. A linear mixed effect model was employed to evaluate mIDHi and other potential effects of: 1) prior radiation therapy (RT), 2) tumor grade, 3) tumor type (oligodendroglioma versus astrocytoma), and 4) the presence of measurable enhancing disease per RANO criteria. Our findings revealed that the growth rate was significantly reduced from +8.5% to +1.5% (p <0.05) in response to mIDHi. The linear mixed effect model demonstrated a significant mean reduction (p<0.05) in the growth rate in the post-treatment group, whereas no significant difference was observed in all other effects. This work demonstrates the utility of AI-based automated brain tumor MRI segmentation for capturing subtle volumetric changes in MRIs of patients with lower grade mIDH gliomas. Our initial findings suggest that mIDHis can significantly reduce tumor growth rate regardless of tumor type, grade, enhancement, or prior RT.

Duke Scholars

Author Katherine Barnett Peters Neurosurgery, Neuro-Oncology

Author Evan Calabrese Radiology, Neuroradiology