AI Unlocks Hidden Biomarkers in Routine ECGs, Detecting Disease and Predicting Mortality Beyond Human Capability
Key Insights
Imperial College London spinout Cardiovolt.ai (search) has raised £1.4 million to commercialize AI models that extract hidden diagnostic signals from standard 10-second ECGs.
The AI achieves 83–93% diagnostic accuracy for heart disease (search) and 70–80% for non-cardiac conditions like diabetes (search) and kidney disease (search), validated across international datasets.
A separate Nature study from UC Berkeley used 440,000 ECGs to discover a previously unrecognized biomarker for sudden cardiac death (search) invisible to even expert cardiologists.
A routine electrocardiogram takes just ten seconds to record and is one of the most common tests in medicine. For decades, clinicians have used it primarily to check heart rate and rhythm, regarding it as a relatively blunt instrument compared with more sophisticated techniques such as echocardiography. Now, two independent research efforts are demonstrating that artificial intelligence can extract far more information from those ten seconds of electrical data than any human—no matter how expert—can perceive.
The Imperial College London Breakthrough
A spinout from Imperial College London's National Heart and Lung Institute (NHLI), Cardiovolt.ai (search), has developed a suite of AI models capable of diagnosing hidden heart conditions, flagging non-cardiac diseases such as diabetes (search) and kidney disease (search), and predicting a patient's risk of death—all from a standard ECG trace. The company has just closed a pre-seed funding round led by Twin Path Ventures, with additional grant funding from Innovate UK, prize money, accelerator support from Imperial Enterprise, and investment from Imperial's DT Prime, bringing total funding to £1.4 million.
"We looked at the ECGs to see if we could do things that are superhuman," said Dr Arunashis Sau, Chief Scientific Officer of Cardiovolt.ai (search), lecturer at NHLI, and cardiology registrar at Imperial College Healthcare NHS Trust. "Not things that clinicians can already do, but things that no cardiologist, no matter how expert, can do."
To train the models, Dr Sau and colleagues sourced over 1.6 million ECGs linked to patient medical histories from a research group in Brazil, supplemented by several million additional ECGs from the United States. The team first asked whether AI could stratify patients into high-risk and low-risk groups and estimate mortality timing. "That's not something we want to use in a general sense," Dr Sau noted, "but in a hospital you might use it to identify people at high risk and intervene early."
The models performed strongly across the board: diagnostic accuracy for heart disease (search) reached 83–93%, while accuracy for non-cardiovascular conditions such as diabetes (search) and kidney disease (search) ranged from 70–80%, all validated across international datasets.
What the AI detects is, in effect, a digital biomarker—a signal that disease processes are underway, embedded in the heart's electrical trace. To investigate the biological underpinnings, the team turned to the UK Biobank, where ECGs are accompanied by imaging data, genetic profiles, and protein measurements not typically available in larger collections. "For example, are there genetics that determine you are at risk of a certain disease as picked up by this biomarker? Are there protein changes? Are there structural changes in the heart?" said Dr Libor Pastika, Chief Technology Officer of Cardiovolt.ai (search).
From Lab to Clinic
Cardiovolt.ai (search) is commercializing AI models developed in Professor Fu Siong Ng's group at NHLI with longstanding support from the British Heart Foundation. "We have chosen this path because it is the one most likely to see the technology used in hospitals," said Professor Ng, who serves as Cardiovolt.ai's Chief Medical Officer alongside his role as consultant cardiologist at Chelsea and Westminster Hospital NHS Foundation Trust and Imperial College Healthcare NHS Trust. "We are the people who are happy to take the risk, and devote our time and energy to making that happen."
The initial clinical focus is on diagnosis of conditions that would otherwise go unnoticed. "If someone comes in to have an ECG, we want to pick up underlying heart failure or valve disease that would never be picked up by a human doctor," Professor Ng explained. "If there is any suspicion, then we will do an echocardiogram to confirm it right away."
The company's immediate priority is securing regulatory approval in the UK, EU, and US. "Our primary market is healthcare providers—hospitals, health systems and cardiology practices that already perform millions of ECGs every year," said Boroumand Zeidaabadi, Chief Executive Officer of Cardiovolt.ai (search). Development has been supported by Imperial's enterprise ecosystem, including the AI SuperConnector accelerator programme and the 2025 Venture Catalyst Challenge, where the team won the AI and robotics track prize.
A New Biomarker for Sudden Cardiac Death (search)
In parallel, researchers at the University of California, Berkeley and collaborating institutions published a groundbreaking study in Nature demonstrating that a deep learning model trained on more than 440,000 ECGs linked to national death records in Sweden identified a previously unrecognized biomarker for sudden cardiac death (search)—one of the leading causes of mortality worldwide. Remarkably, many of the patients flagged as high-risk would not have been identified using current clinical guidelines, despite decades of research devoted to understanding electrocardiography.
The significance extends well beyond cardiology. Electrophysiologists have devoted entire careers to studying every electrical waveform produced by the human heart. The information was always present within the ECG, yet the mathematical relationships predicting sudden cardiac death (search) remained invisible to generations of highly trained physicians. Artificial intelligence did not simply recognize a known pattern more efficiently; it identified a new biomarker that physicians themselves did not know existed.
Redefining the Physician's Role
These developments raise profound questions about the future of medicine. Artificial intelligence excels at quantitative intelligence—processing enormous datasets, discovering unknown biomarkers, identifying subtle abnormalities, and estimating risk with extraordinary precision. Patients, however, experience illness through fear, uncertainty, family responsibilities, financial concerns, and deeply personal definitions of hope and quality of life.
This distinction highlights the difference between disease and illness. Disease is measurable and exists in laboratory values, imaging studies, and ECG waveforms. Illness is experienced and exists in conversations, anxiety surrounding a diagnosis, and the personal values that shape every medical decision. AI may calculate the probability of sudden cardiac death (search), but it cannot determine whether a patient should undergo an invasive procedure or what tradeoffs they consider meaningful.
The physician's role therefore becomes even more important as AI becomes more capable. Rather than spending time searching for disease hidden within complex data, physicians will increasingly interpret what those discoveries mean for individual patients—contributing qualitative intelligence through empathy, ethical reasoning, contextual understanding, and shared decision-making.
As Professor Ng concluded: "A ten-second ECG, once a blunt screening tool, may soon become one of the most informative tests in medicine."