The algorithm gathers data from wearable monitors to find life-threatening irregular heartbeats and allows for data to be sorted through in remote areas where there is a scarcity of cardiologists.
“One of the big deals about this work, in my opinion, is not just that we do abnormality detection but that we do it with high accuracy across a large number of different types of abnormalities,” said Awni Hannun, a graduate student and co-lead author of the paper, in a press release. “This is definitely something that you won’t find to this level of accuracy anywhere else.”
Electrocardiograms (ECGs) are usually given to patients who are suspected of having an arrhythmia, but those have to be done in a doctor’s office. And most of the time, the ECG isn’t able to detect the problem fully and the patient has to wear an ECG monitor continuously for two weeks. The results include hours of data where every second matters when determining if the patient has a problematic arrhythmia.
Stanford Machine Learning Group researchers developed a deep learning algorithm to detect 14 types of arrhythmia for ECG readings. In collaboration with heartbeat monitor company iRhythm, a massive data set was collected and used to train a deep neural network model. It could diagnose arrhythmia as accurately as cardiologists, according to the researchers.
Participants in the study wore an iRhythm wearable ECG monitor, which is a small chest patch, for two weeks and were asked to go about their days like normal while the device measured heart activity. The research group took about 30,000, 30-second clips from different patients that represented a variety of arrhythmia.
“The differences in the heartbeat signal can be very subtle but have massive impact in how you choose to tackle these detections,” said Pranav Rajpurkar, a graduate student and co-lead author of the paper. “For example, two forms of the arrhythmia known as second-degree atrioventricular block look very similar, but one requires no treatment while the other requires immediate attention.”
Three expert cardiologists tested the accuracy of the algorithm by analyzing 300 undiagnosed clips of data and were asked to find arrhythmia within the recordings. Using the cardiologists’ annotated clips, the algorithm could give a diagnosis.
“There was always an element of suspense when we were running the model and waiting for the result to see if it was going to do better than the experts,” said Rajpurkar. “And we had these exciting moments over and over again as we pushed the model closer and closer to expert performance and then finally went beyond it.”
This is not the first instance of arrhythmia diagnoses in a non-clinical setting. Researchers in the Massachusetts Institute of Technology’s Media Lab recently created a smartphone app that could detect the presence of an arrhythmia based on how light reflects off of the skin in a selfie. Apple also has jumped on board and is using its built in heart rate sensor and artificial intelligence to detect atrial fibrillation.
The researchers suggest that this algorithm could create cardiologist-level arrhythmia diagnosis and treatment accessible to people in rural areas who don’t have access to cardiologists in person. They also suggest that it could be used as part of a wearable device for at-risk people that can alert emergency services in the event of a potentially deadly irregular heartbeat.
[Want to stay more on top of MDO content? Subscribe to our weekly e-newsletter.]