Machine Learning–Enabled Detection of Unruptured Cerebral Aneurysms Improves Detection Rates and Clinical Care

Abstract

Background Unruptured cerebral aneurysms (UCAs) have a relatively low prevalence of ≈3%, but detection can prevent devastating consequences of subarachnoid hemorrhage. Here, we assess the performance of a machine learning algorithm to identify UCAs and determine whether routine use of the algorithm improves detection and patient care. Methods From a prospectively maintained multicenter registry across 8 certified stroke centers (1 comprehensive and 7 primary), we identified patients who underwent computed tomography angiography for evaluation of possible stroke from March 14, 2021, to November 31, 2021. A convolutional deep neural network (Viz ANEURYSM) trained to identify UCAs at least 4 mm in size analyzed the images, and ground truth was provided by a blinded expert neuroradiologist. The primary outcome was rate of clinical follow‐up for UCAs detected by the machine learning algorithm. Results Among 1191 computed tomography angiograms performed during the study period, 50 (4.2%) were flagged by the machine learning algorithm as possibly demonstrating a UCA, of which 31 cases were confirmed as true positive (positive predictive value, 62%). There were a total of 36 true aneurysms with 4 cases of multiple aneurysms. Overall, the most common locations included internal carotid artery (42%). Of these cases, 10 (27.8%) were not noted in the clinical radiology report or clinical notes, with a median size of 4.4 mm (interquartile range, 1.6 mm), and 24 (67%) were not referred for follow‐up, with median size of 4.4 mm (interquartile range, 4.2 mm). Of the 24 aneurysms not referred for follow‐up, 15 (62.5%) had been noted in the radiology report. A total of 33.3% (5/15) of the detected but not referred cases had a diameter >7 mm, with median PHASES score of 7. Conclusions UCAs of sizes and intradural locations that require attention and may warrant treatment are frequently missed in routine clinical care. A machine learning algorithm that flags studies and notifies clinicians may minimize missed care opportunities.