AI ECG detects heart disease in under two seconds

What this is about

A research team led by Imperial College London has presented an AI model that derives signs of heart failure and heart-valve disease from a standard electrocardiogram. According to the researchers, the analysis takes under two seconds. Results from a study involving 67,000 patients were presented at the European Society of Cardiology congress in Munich on 31 August 2026.

This matters because an ECG is cheap, quick and widely available. A reliable diagnosis of these conditions still requires echocardiography, an ultrasound examination of the heart. Patients can wait several months for that appointment. The model is therefore not intended to decide who is ill. It is designed to help move urgent cases into ultrasound diagnostics sooner.

What the AI ECG actually does

A conventional ECG records the heart's electrical activity. Clinicians use it to identify issues including abnormal rhythms or signs of a heart attack. The new model also searches for subtle patterns associated with heart failure or diseased heart valves that are difficult for people to see in the trace.

In the reported study, the system detected up to 81% of people with heart failure and up to 90% of those with heart-valve disease. Those detection rates do not mean that every positive result is correct or that every condition is found. The output is a risk signal. A person with a concerning result would still need a cardiac ultrasound.

Why it matters

The researchers say roughly one billion ECGs are performed worldwide each year. If an existing ECG could also serve as a triage tool, hospitals might identify high-risk people sooner without first introducing another scarce diagnostic procedure for everyone.

There are two potential groups of beneficiaries. People with symptoms could be prioritised more quickly. Signs could also be found in people whose ECG was taken for another reason. An earlier diagnosis can lead to earlier appropriate treatment. The British Heart Foundation, which funded the study, also stresses that the system will not detect every heart condition.

In plain language

The system works like an extra sorting aid in a crowded emergency department. It treats nobody and makes no final diagnosis. It only marks the ECG records that clinicians and ultrasound teams should review first — much like a smoke alarm warns of danger without determining exactly what is burning.

A practical example

A hospital records 1,000 ECGs in one day. Suppose 40 belong to people who are later diagnosed with heart failure. At 81% sensitivity, the model would flag about 32 of those 40 cases; roughly eight could be missed. It could also incorrectly flag ECGs from people without the condition.

The hospital must therefore treat neither a flag as a diagnosis nor an unflagged result as reassurance. A sensible workflow would prioritise clinical review: symptoms, medical history, laboratory results and the ECG would be considered together before an ultrasound appointment is assigned.

Scope and limits

First, the reported performance numbers come from a large study, but the announcements do not yet provide a complete peer-reviewed evidence base for routine use in German hospitals. Results may vary by device, patient population and care setting.

Second, false alarms and missed cases remain unavoidable. Without details on specificity, positive predictive value and disease prevalence, it is impossible to know how many additional ultrasound examinations the system would produce.

Third, the model must not replace clinicians or ultrasound. Broad deployment would require external validation, regulatory clearance, continuous quality checks and clear accountability. Processing sensitive health data must also be secured technically and legally.

SEO & GEO keywords

AI ECG, electrocardiogram, heart failure, heart-valve disease, echocardiography, Imperial College London, British Heart Foundation, European Society of Cardiology, medical AI, early diagnosis