Abstract TMP59: Automated Analysis of Dynamic Computed Tomographic Angiography Perfusion Maps Accurately Identifies Regional Hypoperfusion in Minor Stroke Patients

Abstract

Background: Multi-phase computed tomographic angiography (mCTA) perfusion maps can help confirm the diagnosis of stroke which can be especially challenging when the symptoms are minor. The current study proposes an automated stroke detection method for identifying patient stroke status based on volumetric analysis of mCTA perfusion maps applied to minor stroke patient populations. Methods: Minor stroke patients were acutely imaged with mCTA and CTP. mCTA perfusion maps were created using an extreme gradient boosting trees classifier with a CTP TMAX>6s ground truth. Volumes for each patient for each hemisphere were calculated from ten perfusion thresholds equally spaced from the minimum and maximum of the scaled perfusion map. The ten volumes form the variables in a logistic regression algorithm trained on the known stroke status of the hemisphere. 10-fold cross validation was implemented in addition to receiver-operating characteristic (ROC) analysis to produce accuracy, sensitivity, specificity, and area-under-curve (AUC). Results: In total 82 minor stroke patients (median age: 71, 48% female, median NIHSS: 3). 78% had identifiable intracranial occlusions. The analysis generated an ROC curve with an AUC of 81%. Cross-validation produced accuracy, specificity, and sensitivity of 76%, 87%, and 65%, respectively. Conclusion: The stroke detection model developed in the current study accurately categorized stroke and healthy hemispheres in minor stroke. Based on these results dynamic CTA perfusion could help diagnose stroke when the symptoms are mild thus providing accessible and accurate stroke diagnosis in primary stroke centres.