Abstract TP10: Artificial Intelligence Algorithm as a Diagnostic Tool for Aneurysms

Abstract

Viz Aneurysm is an artificial intelligence engine created to detect aneurysms from CT-Angiography (CTA) scans. As with any new technology, it should be measured against the current standard of care. Here, we compared Viz.ai ‘flagging’ of an aneurysm against interpretation reports by faculty radiologists in order to understand accuracy and precision of the technology as a diagnostic tool. In this study, CTA scans were analyzed by Viz Aneurysm from October 2022 to June 2023 at a single Comprehensive Stroke Center. Information regarding each subject was collected through Epic, and the presence of an aneurysm was verified by a radiologist. Although Viz.ai states that the algorithm is limited to detecting aneurysms that are 4 mm in diameter or greater, we have included all aneurysms discovered by radiologists. These ranged in diameter from 0.8 to 22 mm. A total of 963 scans were analyzed, with 70 aneurysms detected total. Approximately 51.10% scans were performed on male patients, and our population ranged in age from 23 to 94 years. The algorithm performed well in flagging scans containing definite aneurysms, with a specificity of 96.78% (95% CI 0.95-0.98) and a negative predictive value of 97.23% (95% CI 0.96-0.98). It was less accurate in excluding scans without aneurysms, showing a sensitivity of 63.64% (95% CI 0.51-0.75) and a positive predictive value of 60.00% (95% CI 0.50-0.69). When aneurysms smaller than 4 mm were excluded, the sensitivity and specificity increased to 81.25% (95% CI 0.64-0.93) and 96.90% (95% CI 0.96-0.98), respectively. The algorithm was successful in identifying all cases (n = 5) of ruptured aneurysms. In several cases (n = 27), radiologists were unable to definitively diagnose an aneurysm, frequently citing the possibility of an infundibulum. It should be noted that there was one case where the algorithm detected an aneurysm, but the patient was diagnosed with a 23-mm pituitary macroadenoma. This suggests that the algorithm may be susceptible to neuroanatomical mimics. We have shown that Viz Aneurysm is a highly specific diagnostic tool that may help medical professionals to identify definite aneurysms. More studies of this nature should be conducted in a variety of populations to confirm these results.