Abstract 132: Clot Radiomics Features From Ct Imaging Predict Treatment Outcomes From Stroke Thrombectomy

Background: Radiomics features (RFs) extracted from CT images have the potential to provide valuable biological and structural information about ischemic stroke blood clots, which may inform mechanical thrombectomy (MT) outcome. Hypothesis: We aimed to identify RFs that predict MT outcome and interpret these findings using paired histological data. Methods: We extracted 293 RFs from co-registered non-contrast CT (nCCT) and CT angiography (CTA) images. RFs predictive of revascularization outcome (defined by first pass effect [FPE], which is near complete removal of occlusive clot in one pass of MT) were selected. We then trained and cross-validated (5-fold) a balanced logistic regression model to assess the RFs’ ability in FPE prediction. In a subset of cases, we performed digital histology on clots retrieved during MT. We computed 227 engineered clot histomic features from whole slide histopathology that represented clot structure and texture. Results: Following univariate feature selection and multi-variate feature ranking, we identified 6 RFs significantly associated with FPE (Figure, top rows). Continuity in lower pixel intensities, scattered higher intensities, and intensities with abrupt changes in texture were associated with successful revascularization outcome. The multi-variate model trained using the 6 RFs could predict FPE well, with AUC=0.832±0.031 and accuracy=0.760±0.059 in training and AUC=0.787±0.115 and accuracy=0.787±0.127 in cross-validation. Each of the 6 RFs were related to clot component organization in terms of red blood cell (RBC) and fibrin/platelet (FP) distribution. Clots with more diversity of components and with varying sizes of RBC and FP regions in the clot section were positively associated with RFs predictive of FPE. Conclusion: Clot RFs are potential structurally- and biologically-interpretable candidate biomarkers for accurate FPE prediction.