A deep-learning model for intracranial aneurysm detection on CT angiography images in China: a stepwise, multicentre, early-stage clinical validation study

Bin Hu ¹, Zhao Shi ¹, Li Lu ²

¹Department of Radiology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, China.
²Department of Radiology, the Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu, China.
³Department of Medical Imaging, the First People's Hospital of Lianyungang, Lianyungang, Jiangsu, China.
⁴Deepwise Artificial Intelligence (AI) Lab, Deepwise, Beijing, China.
⁵Department of Medical Imaging, Guizhou Province People's Hospital, Guiyang, Guizhou, China.
⁶Department of Radiology, Ma'anshan People's Hospital, Ma'anshan, Anhui, China.
⁷Department of Medical Imaging, the Affiliated Suqian First People's Hospital of Nanjing Medical University, Suqian, Jiangsu, China.
⁸Department of Radiology, People's Hospital, Hubei University of Medicine, Shiyan, Hubei, China.
⁹Department of Medical Imaging, the Affiliated Wuxi People's Hospital of Nanjing Medical University, Wuxi, Jiangsu, China.
¹⁰Department of Radiology, General Hospital of Ningxia Medical University, Yinchuan, Ningxia, China.
¹¹Department of Medical Imaging, the First Affiliated Hospital of Xi'an Jiaotong University, Xi'an, Shaanxi, China.
¹²Department of Radiology, the First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
¹³Department of Radiology, People's Hospital of Sanya, Sanya, Hainan, China.
¹⁴Department of Radiology, Lanzhou University Second Hospital, Lanzhou, Gansu, China.
¹⁵Department of Radiology, Tangdu Hospital, Air Force Medical University (Fourth Military Medical University), Xi'an, Shaanxi, China.
¹⁶Medical Imaging Center, Shenzhen Hospital of Southern Medical University, Shenzhen, Guangdong, China.
¹⁷Institute of Artificial Intelligence, Beihang University, Beijing, China.
¹⁸Department of Radiology, Medical Imaging Center, Peking University Shougang Hospital, Beijing, China.
¹⁹Medical Big Data Center, Guangdong Provincial People's Hospital, Guangzhou Guangdong, China.
²⁰Department of Radiology, Ningbo First Hospital, Ningbo, Zhejiang, China.
²¹Department of Radiology, Third Center Medical Center, Chinese PLA General Hospital, Beijing, China.
²²Department of Radiology, General Hospital of Southern Theater Command, PLA, Guangzhou, Guangdong, China.
²³Department of Radiology, Central Hospital of Jilin City, Jilin, China.
²⁴Division of Cardiovascular Imaging, Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA.
²⁵Department of Computer Science, The University of Hong Kong, Hong Kong Special Administrative Region, China.

⁰Department of Radiology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, China.

The Lancet. Digital Health +

|

March 22, 2024

View abstract on PubMed

Summary

A new artificial intelligence (AI) model significantly improves detection of intracranial aneurysms on CT angiography (CTA) scans. This AI tool enhances clinician diagnostic performance and shows high acceptance in real-world clinical settings.

Area Of Science

Medical Imaging
Artificial Intelligence in Healthcare
Neurosurgery

Background

Real-world implementation of artificial intelligence (AI) models remains limited.
Developing and validating AI for clinical use is crucial for advancing healthcare.
Intracranial aneurysms pose significant health risks, necessitating accurate and efficient detection methods.

Purpose Of The Study

To develop a CT angiography (CTA)-based AI model for detecting intracranial aneurysms.
To evaluate the AI model's impact on clinician diagnostic performance.
To validate the AI model's practical application in real-world clinical scenarios.

Main Methods

A deep learning model was trained on 16,546 head and neck CTA images from 14,517 patients.
The model underwent multi-stage external validation involving 120 clinicians across 15 hospitals.
Prospective validation was conducted on 1,562 real-world clinical CTA cases.

Main Results

The AI model achieved high sensitivity (0.957) and outperformed clinicians in detecting aneurysms.
AI assistance significantly improved diagnostic performance (AUC 0.878 vs 0.795) and reduced reading time.
The model demonstrated high acceptance (92.6% adoption) and improved prospective diagnostic accuracy (AUC 0.909).

Conclusions

The developed AI model shows substantial clinical potential for intracranial aneurysm detection.
AI integration enhances clinician diagnostic capabilities and workflow efficiency.
The model is validated for real-world clinical implementation, improving patient care.