Optical Microscopy Predictions of Focal Recurrence in Glioblastoma

Area of Science:

• Neuro-oncology
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Glioblastoma (GBM) frequently recurs after initial treatment, posing a significant clinical challenge.
• Current management of recurrent GBM lacks a standard of care.
• Predicting recurrence location is crucial for optimizing advanced-stage therapies.

Purpose of the Study:

• To develop and validate an AI-based model for predicting the risk of focal glioblastoma recurrence.
• To assess the utility of AI-estimated tumor infiltration in predicting recurrence at resection margins.

Main Methods:

• An AI model was developed using label-free optical microscopy to quantify tumor infiltration (AI-infiltration) in surgical margin samples.
• A random forest classifier integrated AI-infiltration with clinical, radiographic, and molecular data.
• The model was trained and validated on a cohort of 80 patients with glioblastoma.

Main Results:

• Higher glioblastoma infiltration was observed in margin samples from recurrent tumors compared to non-recurrent ones (p = 0.026).
• The AI-driven random forest model achieved high prediction accuracy for focal recurrence (86.6% training, 80.3% validation AUC).
• AI-infiltration emerged as the strongest predictor, outperforming molecular features, and maintained performance across tumor locations.

Conclusions:

• AI-based prediction of tumor infiltration can accurately identify sites at high risk for glioblastoma recurrence.
• This approach holds potential for guiding precision, multimodal therapies to specific high-risk areas.
• The findings suggest a novel strategy for managing recurrent brain tumors by predicting recurrence patterns.