AI Model Spots Heart Disease in Two Seconds Using Standard ECGs

Cardiologists at Imperial College London have developed an artificial intelligence model capable of identifying major heart diseases from a routine electrocardiogram (ECG) in less than two seconds. Presented at the European Society of Cardiology annual congress in Munich, the system extracts subtle diagnostic signals invisible to the human eye, offering a potential solution to long diagnostic wait times.

Bridging the Diagnostic Gap

Standard ECGs have been used for over a century to record the heart's electrical rhythm, making them essential for diagnosing acute conditions like heart attacks or arrhythmias. However, standard ECGs cannot directly detect structural conditions such as heart failure or valve disease. Diagnosing these conditions typically requires an echocardiogram—an ultrasound scan of the heart—for which patients often wait several months.

The new AI tool bridges this gap by identifying early markers of heart failure and valve disease directly from standard ECG data. Because ECGs are among the most widespread diagnostic tools in medicine, with roughly one billion performed globally each year, integrating AI analysis could significantly increase the value of existing clinical workflows.

Trial Performance and Triage Potential

In a clinical trial analyzing data from 67,000 patients in the United States, the model identified up to 81% of patients with heart failure and 90% of those with heart valve disease. Funded by the British Heart Foundation (BHF), the project was led by clinical research fellow Dr. Ahmed El-Medany and Professor Fu Siong Ng of Imperial College London.

While the system is not designed to provide a definitive standalone diagnosis, clinicians emphasize its value as a rapid triage mechanism. Patients flagged as high risk by the algorithm can be prioritized for immediate echocardiograms, drastically reducing the time required to initiate treatment. Furthermore, because ECGs are frequently performed during routine checkups or hospital visits for unrelated symptoms, running the model automatically across all hospital readings could enable opportunistic screening for previously unsuspected heart conditions.

Next Steps for AI Diagnostics

Following the initial trial success, researchers are working toward translating the algorithm into practical point-of-care hardware, including handheld AI-driven ECG readers for frontline healthcare workers.

The breakthrough was showcased alongside other digital health developments at the Munich conference, such as research from the University of Tokyo utilizing short facial videos to screen for hypertension and type 2 diabetes. Together, these innovations reflect a growing movement toward deploying machine learning models to accelerate early detection across primary care settings.