SURG-57. AI-based quantification of tumor infiltration as a major determinant of glioblastoma recurrence and predictive modeling

A hallmark of glioblastoma (GBM) is disease recurrence, which occurs in all patients despite tumor resection, radiation, and chemotherapy. A critical challenge in glioblastoma treatment is the management of recurrent disease for which no standard of care exists. Predicting the location of glioblastoma recurrence may improve the efficiency of advanced stage therapies. Here we present a predictor/classifier of glioblastoma recurrence using an artificial intelligence (AI)-based tumor infiltration open-source model acquired from unprocessed surgical tissues at initial resection. Model predictions were trained using label-free optical microscopy to generate a normalized tumor infiltration value within whole-slide optical images. In a prospective testing cohort of 80 patients, 367 samples, and 133,454 unique images, glioblastoma infiltration was significantly higher in margin samples from recurrent tumors (p = 0.026) compared with non-recurrent tumors. Notably, our AI-based infiltration metric emerged as a major driver of both glioblastoma recurrence and accurate recurrence prediction over clinical, radiographic and molecular features. A random forest (RF) machine learning classifier was able to predict site recurrence with an average area under the receiver operating characteristic curve (AUC) of 85.1% ± 11.0 for a training cohort and 80.9% (95% CI: 0.680–0.940; bootstrapped, 2000 resamples) for a testing cohort. Model performance remained high regardless of tumor location, resulting in regression model predictions of recurrence within 6 months of diagnosis at 5 and 10 millimeters of each sampled site. These findings represent the potential for AI estimates of glioblastoma infiltration to predict sites of tumor recurrence, thereby improving accessibility for targeted precision multimodal therapies to highest-risk areas of disease.