Imperial Researchers Develop AI That Flags Heart Disease From ECGs
London, UK
RALPH ORLOWSKI/Reuters
Source Analysis
What Happened
What Happened
Where Sources Agree
- arrows_inputAI Training Dataset Composition: Reports consistently note that the AI models were trained on 1.6 million ECGs from Brazil, with several million additional recordings from the United States, according to Imperial researchers.
- arrows_inputClinical Triage Potential: Reports broadly predict the technology will improve patient triage as the Imperial College London study of 67,000 US patients demonstrated the AI can identify 81% of heart failure cases and 90% of valve disease cases.
- arrows_inputIntended Clinical Use: Sources verify the AI is designed as a triage tool to help prioritize echocardiograms rather than as a replacement for clinical diagnosis, addressing the issue where patients often wait months for these scans, according to Imperial College London researchers.
Where Sources Disagree
- arrows_outputDiagnostic Accuracy Reporting: Some outlets report the AI's diagnostic accuracy for heart disease as ranging from 83% to 93%, while others emphasize specific detection rates of 81% for heart failure and 90% for valve disease.
- arrows_outputAI Application Scope: While some reports highlight the AI's ability to identify conditions outside cardiology, including diabetes and kidney disease, other outlets focus exclusively on the system's capacity to detect heart failure and valve disease from routine ECGs.
Timeline
September 1, 2026
Validation and Deployment Next Steps: Researchers emphasise substantial further validation, regulatory approval, and testing in real-world clinical workflows are required before routine use; they see the tool as a triage/prioritisation aid rather than a diagnostic replacement and are exploring handheld ECG readers and wider clinical studies. Experts warned the system will not pick up all cases and noted practical motives such as long waits for echocardiograms.
September 1, 2026
Results Presented and Commercialised: The findings were presented at the European Society of Cardiology annual congress in Munich, and Imperial's team has established a spinout, Cardiovolt.ai, (with BHF-funded work) to support development toward clinical applications and commercialisation. The research has also begun to move beyond the university into broader development efforts.
September 1, 2026
Trial Shows High Detection Rates: In testing covering about 67,000 US patients, the system identified roughly 81% of people with heart failure and 90% of those with valve disease; across the wider programme reported accuracies ranged from about 83% to 93% for heart disease in testing datasets. These results came from trials and validation datasets cited by the researchers.
Summaries by Ground AI
Sources
See All 31Most Written About
The stories getting the most coverage from the last 24 hours
Source Analysis
Timeline
September 1, 2026
Validation and Deployment Next Steps: Researchers emphasise substantial further validation, regulatory approval, and testing in real-world clinical workflows are required before routine use; they see the tool as a triage/prioritisation aid rather than a diagnostic replacement and are exploring handheld ECG readers and wider clinical studies. Experts warned the system will not pick up all cases and noted practical motives such as long waits for echocardiograms.
September 1, 2026
Results Presented and Commercialised: The findings were presented at the European Society of Cardiology annual congress in Munich, and Imperial's team has established a spinout, Cardiovolt.ai, (with BHF-funded work) to support development toward clinical applications and commercialisation. The research has also begun to move beyond the university into broader development efforts.
September 1, 2026
Trial Shows High Detection Rates: In testing covering about 67,000 US patients, the system identified roughly 81% of people with heart failure and 90% of those with valve disease; across the wider programme reported accuracies ranged from about 83% to 93% for heart disease in testing datasets. These results came from trials and validation datasets cited by the researchers.
