AI Detects Hidden Heart Disease on Simple ECG

A groundbreaking artificial intelligence model, EchoNext, can now identify silent structural heart disease using only a standard electrocardiogram (ECG). This innovation promises earlier diagnoses, more efficient patient care, and addresses a critical gap in current screening methods.

Revolutionary AI Outperforms Experts

Researchers have developed EchoNext, an AI capable of reliably detecting a range of structural heart diseases (SHDs) across diverse clinical settings. Published in Nature, the study reveals the AI model surpasses the accuracy of traditional physician reviews in identifying these often-undetected conditions.

The Silent Threat of SHD

Millions of individuals, particularly older adults, live with undiagnosed structural heart disease, placing a significant burden on healthcare systems. The economic impact already exceeds $100 billion annually in the U.S. While echocardiography is a vital diagnostic tool, its accessibility is limited by cost and infrastructure, leaving many patients without necessary scans.

ECG as a New Diagnostic Frontier

This AI advancement offers a low-cost alternative, leveraging vast digital ECG archives. The potential for a ten-second ECG to reveal hidden heart conditions could allow for the precise allocation of scarce imaging resources to patients who need them most.

Study Design and Methodology

The research team analyzed over 1.2 million paired ECG and echocardiogram records from 230,000 adults. The AI, a convolutional neural network named EchoNext, processed raw ECG waveforms along with routine ECG parameters and patient age/sex data. Its performance was rigorously tested on internal and external patient cohorts, assessing generalization across various demographics and clinical contexts.

EchoNext’s Superior Performance

In retrospective analyses, EchoNext demonstrated impressive accuracy, achieving an area under the receiver operating characteristic (AUROC) of 85.2% in detecting composite SHD. Its performance remained consistent across different hospital settings and when test sites were swapped, indicating strong generalization capabilities. External validation across multiple medical centers yielded AUROC values between 78% and 80%.

Disease-specific detection was also strong, with an AUROC of 90.4% for identifying reduced left ventricular ejection fraction (LVEF). In a comparative reader study, EchoNext outperformed thirteen cardiologists, correctly identifying SHD in 77% of cases compared to the physicians’ 64%. When clinicians were provided with the AI’s risk score, their accuracy improved slightly to 69%, highlighting the AI’s ability to uncover subtle prognostic patterns.

A recent report indicates that approximately 1 in 8 adults in the United States have some form of cardiovascular disease, underscoring the widespread need for effective screening tools. (Source: CDC, 2024)

Potential for Widespread Impact

A silent deployment of EchoNext on over 84,000 ECGs from patients without prior echocardiograms revealed that 9% were flagged as high risk. The analysis suggested that nearly 2,000 cases of silent SHD might have been identified if the AI alerts had been active, based on current care pathways.

A prospective pilot study, DISCOVERY, further supported the AI’s utility. Participants classified as high-risk by the AI showed a significantly higher prevalence of previously unrecognized structural heart disease (73%), demonstrating the model’s potential to effectively triage patients for further imaging.

The research team has made their code and data publicly available, encouraging independent validation and further development of AI in cardiovascular diagnostics.

Future Directions and Considerations

EchoNext’s ability to enhance ECG screening for SHD, including conditions affecting LVEF and heart valves, shows significant promise. By prioritizing high-risk individuals for echocardiography, this AI can reduce diagnostic delays and the substantial economic burden of heart disease. However, the authors acknowledge potential risks such as patient anxiety from false positives and biases in clinical adoption, stressing the need for ongoing research and pragmatic trials to confirm improvements in survival and healthcare value.