AI-Assisted Echocardiographic Prescreening of Heart Failure With Preserved Ejection Fraction on the Basis of Intrabeat Dynamics

JACC-CARDIOVASCULAR IMAGING

卷:14

Article

Chiou, Yu-An¹  Hung, Chung-Lieh^2,3  Lin, Shien-Fong^1,4 

[1] Natl Chiao Tung Univ, Coll Elect & Comp Engn, Dept Elect & Comp Engn, Hsinchu, Taiwan

[2] MacKay Mem Hosp, Dept Internal Med, Div Cardiol, Taipei, Taiwan

[3] Mackay Med Coll, Inst Biomed Sci, New Taipei, Taiwan

[4] Natl Chiao Tung Univ, Coll Elect & Comp Engn, Inst Biomed Engn, 1001 Univ Rd, Hsinchu 30010, Taiwan

关键词: artificial intelligence;    chronic obstructive pulmonary disease;    convolutional neural network;    heart failure with preserved ejection fraction;    U-n et;   

DOI  :  10.1016/j.jcmg.2021.05.005

来源: Elsevier

【 摘 要 】

OBJECTIVES The aim of this study was to establish a rapid prescreening tool for heart failure with preserved ejection fraction (HFpEF) by using artificial intelligence (AI) techniques to detect abnormal echocardiographic patterns in structure and function on the basis of intrabeat dynamic changes in the left ventricle and the left atrium. BACKGROUND Although diagnostic criteria for HFpEF have been established, rapid and accurate assessment of HFpEF using echocardiography remains challenging and highly desirable. METHODS In total, 1,041 patients with HFpEF and 1,263 asymptomatic individuals were included in the study. The participants' 4-chamber view images were extracted from the echocardiographic files and randomly separated into training, validation, and internal testing data sets. An external testing data set comprising 150 patients with symptomatic chronic obstructive pulmonary disease and 315 patients with HFpEF from another hospital was used for further model validation. The intrabeat dynamics of the geometric measures were extracted frame by frame from the image sequence to train the AI models. RESULTS The accuracy, sensitivity, and specificity of the best AI model for detecting HFpEF were 0.91, 0.96, and 0.85, respectively. The model was further validated using an external testing data set, and the accuracy, sensitivity, and specificity became 0.85, 0.79, and 0.89, respectively. The area under the receiver-operating characteristic curve was used to evaluate model classification ability. The highest area under the curve in the internal testing data set and external testing data set was 0.95. CONCLUSIONS The AI system developed in this study, incorporating the novel concept of intrabeat dynamics, is a rapid, time-saving, and accurate prescreening method to facilitate HFpEF diagnosis. In addition to the classification of diagnostic outcomes, such an approach can automatically generate valuable quantitative metrics to assist clinicians in the diagnosis of HFpEF. (J Am Coll Cardiol Img 2021;14:2091-2104) (c) 2021 The Authors. Published by Elsevier on behalf of the American College of Cardiology Foundation. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

【 授权许可】

Free