Abstract TP95: Utility Of Automated Artificial Intelligence Software In The Analysis Of Ct Perfusion With Acetazolamide Challenge For The Assessment Of Cerebrovascular Reserve In Patients With Cervico-cranial Arterial Stenoses And Occlusions

Abstract

Background: Cerebrovascular reserve (CVR) assessment is valuable in predicting stroke risk in patients with cranio-cervical stenoses and occlusions. CTP with Acetazolamide challenge (AC) allows a qualitative and quantitative assessment of CVR but is limited by ROI positioning, motion and software. Artificial intelligence (AI) automated software with delay in arrival correction has been validated for the processing of CTP data for acute stroke and may more easily detect abnormal CVR. We sought to compare the results of image-processing with a standard platform (Vitrea™) and AI platform (RAPID™). Methods: Retrospective analysis of an IRB approved prospective database of patients who had CTP with acetazolamide (1g IV) who also had a stenosis/occlusion of the internal carotid or middle cerebral arteries was performed. Symptomatic status was defined as recurrent symptoms related to the vessel of interest within 6 months of scan. Two readers, blinded to the clinical scenario, independently reviewed the CTP datasets and identified CBF regional abnormalities and changes post AC using the 2 different platforms. Abnormal CVR was defined as an increase of CBF post AC of <25% of the baseline in the region of interest with Vitrea™, and an expansion of the volume of tissue with CBF<30% using RAPID™. CTP analyses were correlated with the clinical presentation. Results: Thirty patients fulfilling criteria received CTP before and after AC. 24 patients (80%) were symptomatic. Vitrea™ detected decrease in CBF pre-AC in 27 patients and worsening CBF value post-AC in 19 patients. 10 Patients had an increase >25% in CBF post AC. Rapid™ detected 5 patients with CBF lesion <30% pre-AC. Expansion of the CBF lesion occurred in 8 patients post-AC with a mean change in volume of 7.7 ml (P-value=0.002). The sensitivity for the detection of poor CVR in patients with recurrent symptoms was 80% (95% CI 59.30 - 93.17) with Vitrea™ and 45% (95% CI 25.55%-67.18%) with RAPID™. Conclusion: Automated post processing software with delay in arrival correction algorithm is less sensitive to abnormal CVR and has lower correlation with clinical presentation in this small cohort. Further study is warranted to identify thresholds that may better correlate.