# ECG trained artificial intelligence for the detection of patients with inducible myocardial ischemia

## TL;DR

This study developed an AI model using resting ECGs to detect inducible myocardial ischemia, showing strong performance across diverse patient groups.

## Contribution

The novel contribution is an AI model trained on resting ECGs to identify patients with inducible myocardial ischemia, enabling early screening.

## Key findings

- The AI model achieved an AUROC of 0.90 and AUPRC of 0.87 in detecting inducible myocardial ischemia.
- Model performance remained robust across age, sex, comorbidities, and clinical diagnoses.
- Validation on 35,898 patients confirmed consistent results in age- and sex-matched datasets.

## Abstract

Myocardial ischaemia is associated with adverse prognosis. Identifying high-risk individuals who require a stress test is challenging, and a practical screening tool to detect these patients, especially in asymptomatic individuals, is lacking. We aimed to develop an artificial intelligence (AI) model based on a resting 12-lead electrocardiogram to detect patients with inducible myocardial ischaemia.

An AI model was developed using 12 074 resting 12-lead ECGs from 11 700 patients and tested on 1342 patients at two hospitals. Patients with inducible ischaemia were defined as those who received revascularisation for silent ischaemia, stable angina, or unstable angina between 2004 and 2020 (n = 6070). No ischaemia group included patients with 0% stenosis in all epicardial coronary arteries and coronary artery calcium score of ≤100 in coronary computed tomography angiography (n = 7346). The primary outcome was the model performance categorising patients with inducible myocardial ischaemia. We further validated the model through multiple reference and external validation datasets encompassing 35 898 patients. The model showed an area under the receiver operating characteristic curve (AUROC) of 0.90 (95% CI 0.88─0.92), and an area under the precision-recall curve (AUPRC) of 0.87 (95% CI 0.84─0.89). The model performance was robust regardless of age, sex, comorbidities, clinical diagnosis, or culprit vessels. Consistent results were demonstrated in an age- and sex-matched dataset (n = 7414; AUROC 0.85, 95% CI 0.83─0.87 and AUPRC 0.84, 95% CI 0.82─0.87), as well as in reference and external cohorts.

Electrocardiogram-trained AI demonstrated favourable performance in detecting inducible myocardial ischaemia. It may enable screening and risk stratification of high-risk patients.

Graphical Abstract