471 Automated Detection and Analysis of Cerebral Aneurysms With the Viz.ai ANX Algorithm

Abstract

INTRODUCTION: Machine learning algorithms have shown groundbreaking results in neuroimaging. We evaluate the performance of a convolutional neural network (CNN) to detect and analyze intracranial aneurysms (IAs) from computed tomography angiography (CTA). METHODS: Consecutive patients CTA were identified from a single center between January 2015 and July 2021. The ground truth determination of cerebral aneurysm presence or absence was made by the neuroradiology report. The primary outcome was performance of the CNN at detecting IAs in an external validation set, measured using area-under-the-curve (AUC) receiver-operator curve statistics. Secondary outcomes included accuracy for location and size measurement. RESULTS: Among 400 patients with CTA, 150 (37.5%) were male, median age was 39 years (SD 21), and 195 were diagnosed with IAs on neuroradiologist evaluation. Median IAs maximum diameter was 4.6 mm [IQR 2 mm]. In the independent validation imaging dataset, the CNN performed well with 87.6% sensitivity (95%-CI [0.81, 0.92]), 94.0% specificity (95%-CI [0.90, 0.97]) and positive predictive value of 90.9% (95%-CI [0.84, 0.95]) in the subgroup with diameter >=3 mm. CONCLUSIONS: The described Viz.ai ANX CNN performed exceptionally well at identifying presence or absence of intracranial aneurysms in an independent validation imaging set.

Ähnliche Arbeiten