Scientists use smartwatch data and AI to detect heart damage
Researchers say advance could make early screening for structural heart disease possible on a large scale
Scientists have uncovered structural heart problems by using an artificial intelligence tool to vet smartwatch data, in the latest advance in the fast-evolving field of cardiology diagnostics.
The new technique found conditions including weakened pumping ability, damaged valves and thickened muscle, promising to expand the potential of wearable devices beyond detecting rhythm disorders.
The work harnesses simple readouts of heart electrical activity, known as single-lead electrocardiograms (ECGs), that can be recorded by smartwatches without the need for extra equipment. The preliminary study was unveiled at the American Heart Association’s annual scientific sessions in New Orleans.
The innovation has the “potential to transform structural heart disease screening in communities”, says the research, which hasn’t yet been peer reviewed.
“This could make early screening for structural heart disease possible on a large scale, using devices many people already own,” said Rohan Khera, the study’s senior author and director of the Cardiovascular Data Science Lab at Yale School of Medicine.
The study used single-lead ECGs that Apple Watch owners can take using sensors on the device’s back and digital crown. Smartwatch heart data is already used to spot warning signs for atrial fibrillation, or irregular rhythm.
The scientists assembled 266,000 more sophisticated heart activity measurements, known as 12-lead ECGs, gathered from 110,006 patients at Yale New Haven Hospital between 2015 and 2023. They used these to develop an AI algorithm to detect heart conditions, which they then validated externally in more than 45,000 patients.
Their idea was that the AI boost would enable use of the single-lead ECG to screen for some conditions previously identifiable only with the 12-lead measurements. They made the AI model more resilient by adding in “noise” during its training, to simulate the imperfect signals and data it would be likely to receive from the watches.
The researchers then tested the AI tool’s performance on 600 Yale outpatients. The participants all used the same type of Apple Watch to measure single lead ECGs for 30 seconds on the same day they received a heart ultrasound.
The AI algorithm accurately identified the people with structural heart disease 86 per cent of the time. It correctly ruled it out 99 per cent of the time in the participants who didn’t suffer from it.
The researchers acknowledged limitations in their work, including the small sample size of patients with structural heart problems, and the occurrence of some false positive results.
The study is part of a growing effort to widen the use of AI analysis of ECGs beyond clinics and hospitals and into the general population, said Fu Siong Ng, cardiology professor at Imperial College London.
“There is a currently [a] lot of interest in developing and testing AI-enhanced electrocardiogram — heart tracing — algorithms to diagnose hidden disease, such as heart failure and valve disease,” Ng said. “This work is very much in line [with] ongoing prospective clinical studies testing our own such models on the ability of smartwatches to detect hidden heart disease.”