NIMG-08. AN INTEGRATED INFORMATICS MODEL COMBINING CLINICAL FACTORS, RADIOMICS AND A NOVEL CONNECTOMICS FRAMEWORK TO DISTINGUISH PATHOLOGICALLY-PROVEN RADIONECROSIS FROM PROGRESSION IN TREATED BRAIN METASTASES

Abstract PURPOSE/OBJECTIVE(S) To distinguish radionecrosis (RN) from true progression (TP) in brain metastases treated with stereotactic radiosurgery (SRS), we apply machine learning to create a multi-domain model that incorporates clinical factors, multiparametric radiomics(mpRads), and tumor connectomics, novel MRI-based complex graph theory framework describes the intricate network of relationships within surrounding tissue. MATERIALS/METHODS Metastases SRS had pathologic confirmation RN vs. TP after imaging were included single institution. Regions interest manually segmented using largest diameter T1 post-contrast(T1C) lesion plus corresponding area T2 FLAIR hyperintensity. We developed an Integrated Radiomics Informatics System (IRIS) based on isomap support vector (IsoSVM) classify leave-one-out cross-validation (LOOCV). Class imbalance was resolved differential misclassification weighting during training IRIS. Area under receiver operating characteristic (AUC-ROC) AUC-PR (precision recall) analysis performed. RESULTS analyzed 135 lesions 110 patients. There 43 cases (31.9%) 92 (68.1%) TP. The top-performing connectomics features degree centrality (increased RN) average path length (decreased RN), suggesting greater “connectivity” increased similarity intralesional between T1C signal regions cases. radiomics feature multidimensional entropy TP), demonstrating heterogeneity Finally, prior RT before SRS, histology, size. LOOCV IsoSVM successfully classified AUC-ROC 0.84 (95% CI: 0.77-0.90) 0.90 0.82-0.95). F1 score 0.89. CONCLUSION Our machine-learning able efficiently combine multiple domains (i.e., radiomics, factors) pathologically-proven excellent discrimination.