323 QAS.AI: AI Based Software to Intra-Operatively Provide Intracranial Aneurysm Occlusion Prognosis

Abstract

INTRODUCTION: Most AI applications provide features that optimize workflow in clinical settings (diagnoses, reconstruction etc.), however few applications provide intra-operative guidance to surgeons. In several completed studies, AI based backend algorithms were developed to intra-operatively segment intracranial aneurysms (IA) on digital subtraction angiograms (DSA), extract angiographic parametric imaging (API) features from the IAs, and provide the 6-month occlusion prognosis. METHODS: Upon login, the user is presented with a dashboard with a case history, here users can delete, rerun, add comments, and add ground truth outcomes for each case. A case can be initiated in retrospective or prospective modes, retrospective mode will allow the user to upload pre- and post-device placement DSAs from the local file system. Prospective mode will “listen” to the connected angiographic imaging system, show all DSAs acquired and add DSAs as they are generated by the scanner. Users can preview the imaging sequences and select appropriate ones. User must then view, and either approve or modify AI generated IA and inlet vessel masks. This is followed by the results page displaying the 6-month occlusion prediction as well as pre- and post-device API maps. RESULTS: Application segmented IAs with Dice coefficient of 0.903 (95% CI 0.867-0.937), and accuracy of 82.3% (78.3%-86.3%). IA occlusion prognosis was accomplished with AUROC of 0.77 (0.74–0.80) and accuracy of 77.9% (76.2%-79.6%). Identical performance was achieved in both, retrospective and prospective modes. CONCLUSIONS: QAS.AI successfully developed a web-based application that packages several algorithms into a user-friendly method for clinicians to obtain intra-operative prognosis of 6-month IA occlusion. The application is flexible and can be used live while connected to an angiographic scanner or with existing completed cases.

Ähnliche Arbeiten