Stanford University Finds Deep Learning Can Accurately Diagnose Arrhythmia

Science and Technology The researchers at Stanford University in the United States have recently developed a deep learning algorithm that diagnoses 13 different types of arrhythmias by analyzing the ECG data generated by wearable monitoring devices. The accuracy is even higher than that of cardiologists. This result can be used to improve the diagnosis and treatment of arrhythmia in remote areas.

Patients with potential arrhythmias usually visit a doctor and are examined by an electrocardiographer. If there is no problem with the electrocardiograph, the doctor may allow the potential patient to use the wearable device for a two-week continuous monitoring of the heart rhythm. The device generates data for more than 300 hours, and doctors need to analyze every second of the data to find signs of arrhythmia. Harmonic rhythm data and non-hazard rhythm data are often extremely difficult to distinguish.

A news bulletin from Stanford University stated that Wu Enda, the head of the school's machine learning team and a well-known artificial intelligence expert, discovered that this is a data issue. Researchers have developed a deep learning algorithm that can diagnose different types of arrhythmia based on ECG signals. Working with companies that provide wearable heart rate monitoring devices, they acquired about 36,000 people's ECG data samples to train a deep neural network model. After 7 months, this neural network model was more accurate than a cardiologist in diagnosing arrhythmia, and in many cases even exceeded the doctor's rate. Related research papers have been published on the online open database arXiv, which includes preprints of scientific literature.

According to the researchers, there are many types of arrhythmias, and the differences are very subtle, but they have a great influence on how to deal with the arrhythmia. For example, there are two types of arrhythmias known as second-degree atrioventricular block, which look similar, but one of them does not require treatment, while the other requires immediate observation. Their research results can not only find signs of arrhythmia, but also can find the different types of arrhythmia with high accuracy. This accuracy is unprecedented. In addition, the advantage of this algorithm is that it will not fatigue, and it can continuously diagnose arrhythmia.

Researchers hope that their algorithm can be used to provide expert arrhythmia diagnosis in remote areas or in developing countries who cannot see cardiologists. This algorithm can also be used in conjunction with a wearable heart rate monitoring device that is routinely used by high-risk groups to inform emergency personnel when a potentially fatal abnormal heart rhythm is detected.