755 Pathology in the Digital Era of Neuro-oncology: A Pathomics-Based Approach to Modeling Outcomes in Medulloblastoma

Abstract

INTRODUCTION: Pathomics is an emerging data science technique that may be used to extract intricate, sub-visual histopathologic features using high-throughput digital image analysis. METHODS: Using the Children’s Brain Tumor Network database, pediatric medulloblastoma patients were evaluated for inclusion in the study. After tumor segmentation, 49 quantitative pathomic features were extracted using a digital pathology software package (Qupath). LASSO cox proportional hazard model with stratified 5-fold cross-validation was used to identify high-performing features and create a predictive survival model by identifying low-, medium- and high-risk OS cohorts. A classification model was developed using histogram gradient boosting classifier to perform binary classification of Group3/4 vs. SHH/WNT with leave-one-subject-out cross validation. RESULTS: 84 patients with median age at diagnosis of 8.61 years (range 0.31-21.72), and median OS of 45.6 months (range 0.76-195.87) were included in the study. A survival model built using only clinical features yielded a concordance index (c-index) of 0.76 compared to actual outcomes (stratification: p < 0.001). However, a survival model built by combining clinical and high-performing pathomic features yielded the best performance, showing disparate outcomes between low-, medium- and high-risk groups (p < 0.001) with resultant c-index of 0.85. Furthermore, pathomic features classified subjects into Group 3/4 or SHH/WNT with classification accuracy of 71% (SEM = 0.06), and t-test against stratified chance accuracy (57%) showed significantly above-chance performance (p = .019). CONCLUSIONS: This study represents the largest pathomic-based machine learning analysis for predicting survival and molecular subtypes in medulloblastoma. The results demonstrate success in pathomic analysis using machine learning-based modeling to predict survival and molecular subtypes for patients with medulloblastoma. Further work will aim to perform our analysis on an external validation dataset to enhance model performance and improve its clinical applicability.