A research team announced on Mar. 10 the development of an artificial intelligence-powered wearable electrocardiogram (ECG) system that can detect myocardial ischemia significantly earlier than current methods. The new technology aims to address delays in diagnosis, which are linked to increased heart muscle damage and higher mortality rates.
Myocardial ischemia is a leading cause of death and disability worldwide, and early detection is critical for improving patient outcomes. Traditional 12-lead ECGs are considered the clinical gold standard but often miss transient or unpredictable episodes during continuous monitoring. While existing wearable ECG devices have proven effective at detecting arrhythmias such as atrial fibrillation, their ability to identify ischemia has been limited by the subtle changes in ECG waveforms that occur over time.
The research team developed a hierarchical temporal fusion transformer architecture that analyzes cardiac signals across three timescales: intra-beat morphological features, inter-beat variability, and long-term trends. This approach allows the system to predict impending ischemia and assess post-reperfusion injury risk using dual-task learning. The device pairs with an FDA-cleared, chest-worn single-lead ECG patch capable of 14-day continuous monitoring with high signal quality during daily activities.
Validation across four large-scale datasets involving more than 108,000 patients showed strong performance: an area under the receiver operating characteristic curve (AUROC) of 0.947 for ischemia detection and sensitivity between 84.1% and 87.3% at 90% specificity across all cohorts. The model maintained high positive predictive value up to 20 minutes before events and demonstrated consistent results across different demographic groups without evidence of bias.
The system provides an average early warning window of over 18 minutes, allowing clinicians time to initiate emergency protocols before irreversible heart damage occurs. Its attention patterns closely match those identified by cardiologists, supporting its clinical interpretability. However, researchers note limitations due to study populations being primarily from Chinese hospitals; further validation in diverse settings is needed along with prospective trials to confirm real-world effectiveness.
Future work will focus on expanding predictions to other cardiovascular events, integrating electronic health records for personalized risk assessment, and developing privacy-preserving federated learning approaches.
