Leica Proficinet Surgical Tumor Award - Optical Imaging-based Neural Boosting Predicts Recurrence in Glioblastoma

Introduction: Glioblastoma is a universally fatal diagnosis, and local recurrence remains a hallmark of this disease. The extent of tumor resection is the most significant predictor of recurrence. Stimulated Raman histology (SRH) has previously been trained to define glioma within the infiltrative margins, utilizing artificial intelligence (AI). It remains unknown if machine learning is able to predict glioblastoma recurrence. This study aims to evaluate a predictive model of focal recurrence in patients with glioblastoma using SRH and AI-generated cellularity scores for tissue samples taken at the resection cavity margins.

Methods: A multi-center, retrospective cohort study was conducted on patients diagnosed with glioblastoma who underwent resection followed by acquisition of spatially annotated tissues from the resection cavity margins. Tissues were analyzed using SRH optical imaging, and histopathology analysis was performed using confocal microscopy. An AI-generated cellularity score called FastGlioma was generated.

Results: Over 250 patients and 1,000 specimens were analyzed, of which a nested subset of 32 patients were selected for the selection criteria. Using preoperative and postoperative imaging, each margin sample was determined to be in an area of recurrence (n=37) or nonrecurrence (n=104). FastGlioma scores were found to be significantly higher in the recurrent margin sample group when compared to the nonrecurrent group (p = 0.038), which was further confirmed by a pathologist-determined cellularity score (0-3) that demonstrated similar findings (p = 0.024). Six machine learning classifiers were then trained for recurrence prediction. Extreme gradient-boosted trees (XGB) performed best when using the thirteen most predictive variables with an AUC of 0.872. XGB screening for the minimum practical number of variables demonstrated an AUC of 0.759 when using only FastGlioma and methylation scores as variables.

Conclusion : FastGlioma has the potential to predict focal recurrence of glioblastoma, allowing for more tailored approaches to surgical resection and radiotherapy to increase progression-free survival.