An AI reads an ECG in two seconds and finds what cardiologists cannot

An AI reads an ECG in two seconds and finds what cardiologists cannot

Introduction

An AI system trained by researchers at Imperial College London can read an ECG (Electrocardiogram) in under two seconds, identifying signs of heart failure and valve disease that cardiologists might miss. These results, presented at the European Society of Cardiology congress, show promising potential for improving heart health care.

The AI’s Superhuman Abilities

Dr. Ahmed El-Medany describes the AI as "superhuman" due to its ability to detect subtle signals in ECGs that are often overlooked by human clinicians. The AI successfully identified up to 81% of heart failure cases and nearly 90% of valve disease cases in a trial involving 67,000 patients in the US.

Addressing Capacity and Cost Issues

The technology aims to address capacity and cost challenges in heart health care. Heart failure diagnoses often involve lengthy waiting times for echocardiogram scans, while ECGs are performed more frequently but lack the same level of diagnostic capability. The AI could be used for triage, prioritizing patients for further imaging based on the AI’s analysis of their ECGs.

Large-Scale Potential

The current trial is part of a broader research program. The researchers trained their models on 1.6 million ECGs from Brazil, linked to patient records, and several million additional recordings from the US. The models have shown promising accuracy in detecting not only heart failure and valve disease but also heart attacks, arrhythmias, diabetes, and kidney disease.

Future Directions

The Brazilian dataset, with its linked patient records, enabled the models to learn patterns associated with diagnoses, improving their accuracy. The researchers aim to continue refining their models and exploring their potential across various medical fields.