AI-based detection and sizing of saccular intracranial aneurysms: a single-center retrospective validation study using computed tomography angiography

基于人工智能的囊状颅内动脉瘤检测和尺寸测量:一项采用计算机断层扫描血管造影的单中心回顾性验证研究

阅读:1

Abstract

BACKGROUND: Imaging advantages have raised intracranial aneurysm (IA) detection rates but have also increased radiologists' workloads. Coupled with visual fatigue, this heavier burden heightens the risk of missed or erroneous diagnoses. Concurrently, artificial intelligence (AI) has shown great promise for analyzing medical images. This study aims to evaluate the diagnostic performance of AI software for IAs and to provide an initial, single-centre validation of its potential as a supportive tool in future deployment. METHODS: Between January 2019 and September 2023, 452 patients with 544 IAs diagnosed by head and neck computed tomography angiography (CTA) who also underwent digital subtraction angiography (DSA) were included. The AI's ability to detect the presence, location, and size of IAs was recorded. Its results were compared with DSA, and the agreement between AI and radiologists in measuring IA size was evaluated. RESULTS: The AI software demonstrated a sensitivity of 88.97% (95% CI [0.861-0.963]) and an accuracy of 75.04% (95% CI [0.715-0.783]) for detecting IAs. Specifically, the accuracy and sensitivity of AI in detecting IAs that are smaller than three mm, between 3-5 mm, and larger than five mm are 58.46% (95% CI [0.462-0.698]) and 66.67% (95% CI [0.537-0.775]), 76.68% (95% CI [0.714-0.813]) and 88.93% (95% CI [0.843-0.924]), 77.10% (95% CI [0.720-0.816]) and 94.24% (95% CI [0.906-0.966]), respectively. There was good agreement between radiologists and DSA, between AI and DSA, and between radiologists and AI for identifying the location of IAs, with kappa values all greater than 0.75. The radiologists and AI also showed good consistency in measuring depth, width, height, and maximum diameter, with intraclass correlation coefficients (ICCs) all greater than 0.75, except for the neck width, which had an ICC of 0.492. CONCLUSIONS: The AI software performs well in terms of detecting IAs that are larger than three mm and shows good agreement with radiologists in localizing positions and extracting their morphometric parameters (except for neck width measurements). The AI software has proved to be a reliable adjunct for IA detection and measurement tasks.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。