The Stanford Machine Learning Group, along with digital health company iRhythm Technologies, have developed a deep learning algorithm for the detection and diagnosis of cardiac arrhythmias, or abnormal heart rhythms.
According to officials, the algorithm is capable of expert level detection of 14 cardiac output classes, including 12 arrhythmias as well as sinus rhythm and noise from artifact. The collaboration leveraged the iRhythm data science and clinical teams’ knowledge in electrocardiogram (ECG) analysis, as well as iRhythm’s ECG data set to produce an arrhythmia detection algorithm.
Because deep learning models are dependent upon vast amounts of reliable data, the company provided an annotated data set of about 30,000 unique patients, 500 times larger than standards-based databases utilized in previous studies, officials noted. This enabled the Stanford researchers, in collaboration with iRhythm machine learning specialist Masoumeh Haghpanahi, Ph.D., to develop the 34-layer convolutional neural network, comparable to artificial intelligence (AI) models used in computer vision and speech recognition, officials stated.
iRhythm already uses a wearable biosensor—called Zio —worn for up to 14 days on a patient in order to gather data and diagnose cardiac arrhythmias. Now, speculatively, by employing this algorithm, diagnosing arrhythmias could become much more affordable, simpler, and feasible for physicians.
The Stanford Statistical Machine Learning Group is a blend of faculty, students, and post-docs spanning AI, systems, theory, and statistics. “The group’s work spans the spectrum from answering deep, foundational questions in the theory of machine learning to building practical large-scale machine learning algorithms which are widely used in industry,” it states.
According to a paper published by Haghpanahi and others, the researchers stated, “We develop an algorithm which exceeds the performance of board-certified cardiologists in detecting a wide range of heart arrhythmias from electrocardiograms recorded with a single-lead wearable monitor. We build a dataset with more than 500 times the number of unique patients than previously studied corpora… We exceed the average cardiologist performance in both recall (sensitivity) and precision (positive predictive value).”
Get the latest information on Healthcare Analytics and attend other valuable sessions at this two-day Summit providing healthcare leaders with educational content, insightful debate and dialogue on the future of healthcare and technology.