AI Case Study
Cardiogram detects atrial fibrillation with 97% accuracy surpassing FDA-cleared wearable ECG devices using the Apple Watch and machine learning
Atrial fibrillation, the most common abnormal heart rhythm, causes 1 in 4 strokes and frequently goes undiagnosed. In the clinically validated mRhythm study by Cardiogram and the University of California San Fransisco (UCSF), DeepHeart, which is a a semi-supervised deep neural network, detected atrial fibrillation with 97% accuracy (c-statistic) in a hospital environment, using optical heart rate sensors, setting the stage for cost-effective, broadly-deployed AF screening.
Healthcare Providers And Services
According to Cardiogram's blog post, in 2015, "Cardiogram and UCSF created the mRhythm study to collect and analyze user heart rate data. They trained trained DeepHeart, their deep learning architecture, on 139 million heart rate measurements contributed by 9,750 users. The study's results, published in JAMA Cardiology, show that DeepHeart can detect atrial fibrillation with accuracies higher than FDA-cleared wearable ECG devices.
DeepHeart has high accuracy on detecting atrial fibrillation in a hospital environment. The real world, however, is very different from a hospital bed. Motion, sweat, and sunscreen can cause inaccurate optical heart rate readings. Alcohol consumption and exercise can mask or be mistaken for arrhythmias. The task of detecting atrial fibrillation is much harder."
R And D
According to Apple's press release, "Atrial Fibrillation (AFib), the leading cause of stroke, is responsible for approximately 130,000 deaths and 750,000 hospitalizations in the US every year. Many people don’t experience symptoms, so AFib often goes undiagnosed."
Researchers report 97% accuracy in detecting atrial fibrillation, higher than FDA-cleared wearable ECG devices.
"DeepHeart is a semi-supervised deep neural network that accurately predicts cardiovascular risk, requiring 10 times less labeled data than conventional deep learning techniques."
"Training deep neural nets on small datasets is an area of active research. At Cardiogram, we have an abundance of unlabeled user data, and applied unsupervised training to boost the accuracy of our model.
In one attempt, we pretrained the model with a noisy auto-encoder: Given an input of user data with gaussian noise, predict the true, unadulterated data. The learned model weights were then used as initialization parameters in supervised training.
Our second method of unsupervised pretraining outperformed the first. Here, we pretrained the network to predict a set of hand-engineered features inspired by the medical literature: average absolute difference between successive heart rate measurements in window sizes of 5 seconds, 30 seconds, 5 minutes, and 30 minutes.
This model obtained a 93% c-statistic (95% confidence interval: 91%-95%) on the tuning set."
Heart rate measurements and step counts