A Retrospective Evaluation of the Potential Impact of an Artificial Intelligence-based Software on the Detection and Follow up of Unruptured Intracranial Aneurysms

Introduction: A patient with an unruptured intracranial aneurysm that is undetected in routine clinical care or not referred to follow up is a missed opportunity to mitigate the risk of subarachnoid hemorrhage. In a large, metropolitan healthcare system, we retrospectively investigated the ability of an artificial intelligence (AI)-based software to detect aneurysms and the potential impact on patient referral.

Methods: Across 11 hospitals, a total of 1777 consecutive CTAs from patients who underwent evaluation for stroke from 05/01/2021 - 10/20/2021 were processed by Viz ANEURYSM (Viz.ai, Inc.), which uses a convolutional deep neural network trained to flag patients with suspected aneurysms on CT angiograms (CTAs). Ground truth was provided by two expert neuroradiologists. The radiology reports and clinical records of the suspected true positive cases at least 2 mm in size were reviewed to determine if the aneurysm was detected and followed up within 4 months of CTA.

Results: A total of 20 suspected aneurysm patients were confirmed positive. The median age was 65 [IQR 14], and 55% of patients were female. The median maximum aneurysm dimension was 4.7 mm [IQR 2.5 mm], and the most common location was the posterior communicating artery (35%) followed by the middle cerebral artery (15%) and anterior communicating artery (15%). There were 11/20 (55%) cases that were not detected in the radiology report, and none of them were referred for follow up. Of the 9 cases detected in the radiology report, 4 were not referred to follow up.

Conclusion : In a retrospective review of consecutive CTAs, the algorithm identified cases that were not detected in routine clinical care, and there were detected cases that were not referred to follow up. These results suggest that AI-based detection and referral software may improve clinical care.