Automated Echocardiographic Measurement of Left Ventricular Ejection Fraction Based on Foundation Model in Computer Vision
10.3969/j.issn.1000-3614.2024.11.006
- VernacularTitle:基于计算机视觉大模型的超声心动图左心室射血分数自动化测量
- Author:
Xintong WU
1
;
Xiaolin DIAO
;
Qi ZHAO
;
Jiahui GENG
;
Xiaoyuan GAO
;
Zixing WANG
;
Xin QUAN
;
Zhenhui ZHU
;
Wei ZHAO
Author Information
1. 中国医学科学院 北京协和医学院 国家心血管病中心 阜外医院 信息中心,北京 100037
- Keywords:
echocardiogram;
left ventricular ejection fraction;
computer vision;
foundation model
- From:
Chinese Circulation Journal
2024;39(11):1092-1097
- CountryChina
- Language:Chinese
-
Abstract:
Objectives:To examine the feasibility of using foundation model in computer vision for echocardiographic left ventricular ejection fraction measurement. Methods:Based on the most extensive publicly accessible repository of echocardiographic loops,EchoNet-Dynamic,featuring 10024 recordings from individual patients,a foundation model in computer vision,VideoMAE V2,was fine-tuned,validated,tested using 7460,1288,and 1276 echocardiographic loops,respectively. Results:The mean absolute error between left ventricular ejection fraction measurements of VideoMAE V2 and expert's measurements was 3.94％ (95％CI:3.79％-4.11％).The Pearson's correlation coefficient was 0.91 (95％CI:0.89-0.92).Additionally,VideoMAE V2 demonstrated exceptional accuracy in identifying patients with a left ventricular ejection fraction below 50％,achieving an AUC of 0.96 (95％CI:0.95-0.97). Conclusions:This study validates the feasibility of using foundation model in computer vision for measuring left ventricular ejection fraction in echocardiographic loops and lays the foundation for the development of a generalized multimodal automated interpretation system for echocardiography.
