An AI analyzes an ECG in two seconds and detects what cardiologists may overlook.

      An ECG takes only a few seconds to perform and is inexpensive. Researchers at Imperial College London have developed an AI system that can analyze one in less than two seconds, detecting signs of heart failure and valve disease that clinicians might miss from the same trace. These findings were shared at the European Society of Cardiology congress in Munich and reported by the Guardian on Monday; the trial involved 67,000 patients in the United States.

      The AI tool was able to identify as much as 81% of heart failure cases and up to 90% of valve disease cases, even though it was not originally designed to diagnose these conditions. Dr. Ahmed El-Medany, a clinical research fellow at the British Heart Foundation at Imperial, described the system as “superhuman AI.”

      The claim emphasizes that while the signal is present in the ECG recording, clinicians are unable to consistently extract it on their own. The main issue it aims to address is capacity. "Patients can often wait several months for a heart ultrasound after being referred by their doctor," noted Professor Fu Siong Ng, a cardiology professor at Imperial. Delays in diagnosing heart failure can have severe implications, as the disease worsens over time, and earlier treatment generally yields better results, leaving patients waiting while their condition deteriorates.

      In contrast to an echocardiogram, which requires a trained sonographer, specialized equipment, and an appointment, an ECG can be executed with just ten electrodes and a nurse, making it more common. The goal is to utilize the AI for triage rather than to replace clinicians. If software can analyze ECGs already being conducted and pinpoint patients more likely to have structural heart issues, those individuals could be prioritized for ultrasounds instead of following the usual waiting list order.

      "Technology like the AI ECG in this research could help expedite patients who are most likely to have a heart abnormality," remarked Dr. Sonya Babu-Narayan, a consultant cardiologist at the British Heart Foundation. Cost is another factor that could allow this approach to be implemented widely. ECGs are among the most affordable and frequently used medical tests, so incorporating software to analyze existing recordings would require minimal additional infrastructure or costs.

      This current trial is part of a larger research initiative. Ng’s team trained its models on 1.6 million ECGs from Brazil linked to patient records, along with several million more from the United States. The broader project includes investigations beyond heart failure and valve disease; the models have also been employed to detect heart attacks, arrhythmias, and conditions outside cardiology, such as diabetes and kidney disease. The researchers have reported accuracies ranging from 83% to 93% for heart disease and 70% to 80% for other conditions.

      The Brazilian dataset is crucial because the recordings are linked to subsequent patient outcomes, allowing the models to learn which ECG patterns were associated with later diagnoses, rather than merely recognizing previously identified conditions at the time of testing. Additionally, the research has expanded beyond academia. The BHF-funded work is being commercialized through a spinout called Cardiovolt.ai, with Ng as chief medical officer and Dr. Arunashis Sau as chief scientific officer.

      The team plans to integrate hardware with software, intending to develop handheld ECG devices that incorporate the AI into the workflow, which could extend the technology beyond hospitals to areas where ECGs are easier to perform. However, this would also shift the system's role from assisting clinicians in prioritizing existing patients to potentially identifying individuals who have not yet been referred for further examination.

      Britain faces an unresolved regulatory challenge concerning such technology. The country boasts a robust academic pipeline for cardiovascular deeptech, with research yielding new methods for addressing cardiovascular disease. However, transitioning from academic findings presented at medical conferences to routine hospital use necessitates a separate process of clinical validation and regulatory approval. The existing system has not undergone this process yet. An AI tool that identifies patients needing referral for further investigation makes a clinical claim, thus needing to meet relevant medical device standards before widespread deployment in the UK.

      There is also a distinction between what has been shared at a medical congress and what has been shown in standard clinical practice. Although the trial indicates that information regarding heart failure and valve disease can be derived from an ECG not specifically designed to detect those conditions, it has yet to be proven whether utilizing that information in real clinical environments leads to earlier treatments or improved patient outcomes.