Abstract 3683: AI assisted MRI volumetrics improve recurrent glioblastoma patient stratification following immunotherapy treatment

Abstract Background: MRI-based monitoring of tumor progression after treatment plays a critical role in patient care decisions and clinical trial monitoring for glioblastoma patients. However, traditional MRI assessments of glioblastoma tumor burden (i.e. RANO-based criteria) can be significantly confounded by inflammatory pseudo-progression—especially in the context of immunotherapy treatment. Methods: In an effort to improve MRI assessments of tumor burden in glioblastoma patients, we performed volumetric tumor segmentation of >1000 MRI timepoints from 41 recurrent high-grade glioma patients who underwent oncolytic herpesvirus treatment. Volumetric assessments were correlated with patient outcomes and clinical covariates to assess their utility in stratifying patient outcomes relative to RANO analysis. Furthermore, to reduce the cost and time required for tumor segmentation, we leveraged our dataset and the Brain Tumor Segmentation (BraTS) 2023 challenge dataset to train a convolutional neural network to automate glioblastoma tumor segmentation. Automated segmentation volumes were validated against manually segmented tumor volumes and optimized for high-throughput processing. Results: While routine RANO assessment of tumor progression in the months following therapy failed to separate short- and long-surviving patients, volumetric assessment enabled clear stratification of patients into short, medium, and long-surviving groups using only a few months of post-treatment imaging follow up. Furthermore, automated segmentation volumes showed a high-concordance with manual segmentation volumes in most patients, with deviations between the two being largely restricted to misclassification of resection cavities. In most cases, such errors could be easily detected and corrected, enabling rapid and accurate segmentation. Conclusions: Volumetric tumor segmentation significantly improves stratification of recurrent glioblastoma patients following immunotherapy treatment versus traditional RANO classification. While such segmentation is time and skill intensive when performed manually, AI assisted segmentation can result in accurate segmentation volumes with greatly reduced time and skill. Surgical artefacts remain challenging for automated segmentation approaches; however, we expect these limitations to be overcome as we improve our training datasets. We expect this to enable rapid and cheap volumetric analyses capable of improving glioblastoma patient monitoring and, ultimately, providing clinicians with more accurate information when making patient care decisions. Citation Format: Alexander L. Ling, Joshua D. Bernstock, Erickson Torio, Naoyuki Shono, Junfeng Liu, Ana Montalvo Landivar, Nafisa Masud, Ethan Chen, E. Antonio Chiocca. AI assisted MRI volumetrics improve recurrent glioblastoma patient stratification following immunotherapy treatment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3683.