Development and validation of a deep-learning algorithm for rule-in and rule-out coronary artery disease based on electrocardiogram without evidence of myocardial ischemia

Abstract

Background

Current coronary artery disease (CAD) guidelines recommend to rule-out or rule-in patients for further examination by assessing a pretest probability (PTP) ≤ 5 % or ≥ 15 %. We developed and validated a deep-learning algorithm for rule-in or rule-out based on electrocardiogram (ECG) without myocardial ischemia evidence.

Methods

Between October 2019 and June 2022, data from two centers (Fuwai Hospital [Beijing] and Yunnan Fuwai Hospital) of CAD-suspected patients undergoing either coronary angiography or coronary computed tomography were used. Data from the Fuwai Hospital (Beijing) were used to train (randomly 90 %) and internally validate (randomly 10 %) a deep-learning algorithm to detect CAD (≥ 70 % stenosis) based on 12-lead ECGs. An algorithm-based decision-making protocol was established for rule-out or rule-in based on a predefined threshold allowing for a 95 % negative predictive value (NPV). Data from the Yunnan Fuwai Hospital were used to externally validate the performance of the decision-making protocol. The CAD prevalence was calculated in patients who were recommended to rule-in or rule-out.

Results

In internal validation set, area under the receiver operating characteristic curve (AUC) was 0.81 and the CAD prevalence of patients who were recommended rule-out and rule-in were 5 % (40/790) and 23 % (527/2253), respectively. In external validation set, the CAD prevalence of patients who were recommended rule-out and rule-in were 0 % (0/661) and 15 % (255/1699), respectively.

Conclusions

Our algorithm based on ECG without myocardial ischemia evidence performed good in CAD detection. An algorithm-based decision-making protocol could achieve the guideline-recommended performance in guiding rule-out or rule-in for further examination.