Work

Artificial intelligence-enhanced electrocardiography improves the detection of coronary artery disease

Year: 2024

Type: article

Abstract: Background: Most patients with myocardial infarction have significantly stenotic coronary arteries. Our aim is to develop an artificial intelligence (AI)-assisted algorithm to improve the detection of significant coronary artery disease (CAD) among high-risk subjects with normal electrocardiograms (ECGs).Methods: We conduct a retrospective study of ECGs among patients (age ≥18 years) receiving coronary angiography (CAG) that provides a clinical diagnosis. The training (n=5,376) and validation (n=7,163) cohorts are collected from Chang Gung Memorial Hospital at Keelung and Linkou, Taiwan, respectively. Five types of features are used in the construction of the algorithm and these are obtained from 12-lead ECG datasets: (1) time interval; (2) amplitude; (3) difference in amplitudes; (4) slope between peaks; (5) various common features such as heart rate. A total of 561 features are identified. The XGBoost model is used since it gives the best performance.Results: Several findings can be pinpointed. Firstly, the AI-enhanced ECG algorithm can be an effective approach to screening for CAD. In subjects with normal ECGs, the sensitivity is 0.82-0.84 and the prediction rate is 45%-47%. In subjects with abnormal ECGs with or without ischemia, the prediction rates are 92%-95% and 80%-83%, respectively. Secondly, most of the top ranked features are associated with slope and amplitude difference between peaks, which are not easily recognized by human clinicians. Thirdly, sex differences are highlighted and have a tremendous effect on the prediction and ranking of features. Lastly, the performance of the AI-enhanced ECG algorithm when detecting CAD is comparable with that of myocardial perfusion scintigraphy-Thallium scanning, a nuclear medicine test. Thus, the AI-enhanced ECG approach should be a simpler alternative.Conclusions: An AI-enhanced ECG algorithm provides a noninvasive and effective method for the detection of significant CAD in high-risk subjects with normal ECGs and also in those subjects with abnormal ECGs without ischemic description.Funding: This research was funded by grants from the Ministry of Health and Welfare (Smart Healthcare for Obesity Therapeutics; PD-109-GP-02, MG-110-GP-03 and MG-111-GP-03 to HKS) and Chang Gung Memorial Hospital (CRRPG2H0191-0193 to CTM; CRRPG2H0181-0183 to NIY, CRRPG2H0171-0173 to CHY; CORPG2K0011-0013 to CCL).Declaration of Interest: YJC, CTM, TPS, NIY, CCL, CTC, HKS, and SFT do not have competing interests to disclose. CHY, THT, CHC, and TFT are inventors on the Taiwan patent I768624.Ethical Approval: The present study was approved by the Institutional Review Board of Chang Gung Medical Foundation (IRB No: 201800724B0, 202001605B0C101, and 202000077B0A3).

Source: Computational and Structural Biotechnology Journal

DOI: 10.1016/j.csbj.2024.12.032