A continuing study into the medical potential of consumer wearables has confirmed devices like Apple Watch are sensitive enough to detect abnormal heart rhythms with a 97 percent accuracy, a performance that beats out add-on ECG accessory KardiaBand.
Apple Watch, and other consumer wearables including Android Wear devices, were used in a study that applies a neural network called DeepHeart to recorded heart rate data in a bid to detect atrial fibrillation, a common heart arrhythmia that can lead to stroke in some patients.
As AppleInsider detailed last May, the DeepHeart study is a joint effort between the University of California, San Francisco's Health eHearth Study and digital health startup Cardiogram, the latter of which markets an eponymous Apple Watch app.
Comprised of both online and in-hospital components, the study incorporated data from a group of 9,750 Cardiogram users, a subsection of which contributed 129 million heart rate measurements and 6,338 ECGs to train the DNN. Over time, DeepHeart learned to transform raw sensor measurements, specifically hart rate and step counts, into fibrillation risk scores, says Cardiogram cofounder Brandon Ballinger.
The resulting model is capable of achieving a c-statistic, or accuracy, of 97 percent when compared to a traditional 12-lead ECG. That same figure was presented at the Heart Rhythm Society's Heart Rhythm 2017 conference last year, but has now been confirmed in a peer review process. Results of the study, "Passive Detection of Atrial Fibrillation Using a Commercially Available Smartwatch," were be published in JAMA Cardiology on Wednesday.
Of note, the now confirmed 97 percent accuracy rate is higher than that of AliveCor's FDA-certified KardiaBand, an Apple Watch ECG peripheral that launched in the U.S. in November. Apple Watch with DeepHeart attained a 98 percent sensitivity and 90 percent specificity compared to respective Kardia Band rates of 93 and 84 percent.
Unlike Apple Watch's built-in optical heart rate sensor, the $199 KardiaBand takes a 30-second ECG reading by measuring electrical impulses collected at a user's extremities. To conduct measurements, users wearing Apple Watch on their left arm place their right thumb on a sensor pad, which presses an inner electrode onto their left wrist, and vice versa for right arm wearers.
DeepHeart is the first large-scale, peer-reviewed study to appear in a medical journal demonstrating consumer wearables' ability to detect a major health condition, Ballinger says. Further, DeepHeart's semi-supervised deep learning algorithm and heuristic pre-training realized a significant increase in accuracy compared to similar artificial intelligence research.
For example, Google Brain's research on Diabetic Retinopathy required 128,175 training labels, while Stanford University's skin cancer detection study used 129,450 labels. DeepHeart needed only 6,338 ECGs to attain a high level of accuracy thanks to a special pre-training procedure that approximated the average difference between successive heart rate measurements across different window sizes, the study says.
As detailed in previous reports, validation was performed on a test sample of 51 patients scheduled to undergo treatment. Each person wore an Apple Watch for 20 minutes before and after cardioversion, a procedure that restores normal heart rhythms to patients with arrhythmia.
The study's results arrive as Apple dips its toe into the waters of investigational medicine. Last year, the company partnered with Stanford to deliver its own Apple Watch-based atrial fibrillation detection program called the Heart Health Study.
Cardiogram plans to expand on its research initiatives in 2018 as it looks to apply deep neural network technologies to real-world scenarios, including diagnosis of previously unidentified atrial fibrillation. In February, the firm, in collaboration with UCSF, released a study showing Apple Watch and its sensors can detect early signs of diabetes with 85 percent accuracy.
"Our ultimate goal is to turn the wearable on your wrist into Baymax -- a personal healthcare companion," Ballinger told AppleInsider, referring to the superhero appearing in various Marvel Comics titles and the movie Big Hero 6.
Interestingly, Apple is also rumored to be working on non-invasive glucose sensors for Apple Watch, technology many consider to be a "holy grail" of modern medicine. The project is not expected to bear fruit for at least a few years, if it even survives early development.
Cardiogram is a free 55.2MB download from the iOS App Store.