AI finds hidden ECG signal that predicts sudden cardiac death risk

Sudden cardiac death claims over 300,000 lives annually in the U.S., despite the availability of implantable defibrillators that can prevent many fatal heart rhythms. The challenge lies in identifying who needs these devices. A study led by Ziad Obermeyer at the University of California, Berkeley, used artificial intelligence to address this issue. By training a neural network on a 10-second electrocardiogram (ECG), the team aimed to predict cardiac arrest risk. A second network was used to interpret these predictions into visible ECG features for cardiologists. Currently, cardiologists rely on left ventricular ejection fraction (LVEF) from ultrasounds, which is not always accurate. Obermeyer’s team sought a better risk marker using ECGs, which are more accessible. They employed a 64-layer neural network trained on a large dataset from Sweden, successfully identifying high-risk patients. This approach was validated with data from the U.S. and Taiwan, showing its broader applicability. QUESTION: How might advancements in AI technology change the way doctors predict and prevent health issues in the future?