Abstract P315: Radiomics Signature of DWI-FLAIR Mismatch Correlates With Clinical Phenotype of Patients With Large Vessel Occlusion Stroke

Abstract

Introduction: For patients presenting with large vessel occlusion for endovascular thrombectomy (EVT), FLAIR hyperintensity within ischemic lesions may be inversely related to parenchymal viability but relies on subjective grading. Radiomics is an emerging image quantification methodology that may more objectively represent continuous image characteristics. We propose a novel radiomics approach to describe infarct FLAIR positivity and evaluate its correlation with clinical presentation. Methods: For patients with pre-EVT MRI, ischemic lesions were manually segmented on DWI, co-registered to FLAIR and visually graded for FLAIR positivity (0, +, ++). Radiomics were extracted within the ischemic lesion outlines on FLAIR. LASSO regression was used to select features for the DWI-FLAIR mismatch radiomics signature. Canonical correlation analysis was used to relate this signature to clinical features. Results: We identified 103 patients with mean age 68±16 years and 63% female. The radiomics signature of DWI-FLAIR mismatch included FLAIR histogram kurtosis and gray level cluster shade; both correlated with visual grading ( Figure 1A ). The first of the estimated 3 canonical pairs was statistically significant (canonical correlation=0.50, corrected p=0.009); kurtosis was positively associated with atrial fibrillation and age, while negatively associated with last known well, smoking, and diabetes ( Figure 1B ). Conclusion: The radiomics signature of DWI-FLAIR mismatch before EVT correlates with visual grading and may provide a continuous metric to describe infarct evolution. Further exploration of larger datasets is required to determine additional granular relationships with clinical features.