Deep learning for Arrhythmia detection

Artificial Intelligence has been revolutionizing every field of medicine and Cardiology is no exception.The applications of use of artificial intelligence range from detecting subtle patterns in ECG to use of data to predict the site of catheter ablation sites in management of arrhythmia.

A recent study by Department of Cardiology at Mayo clinic used an AI-enabled EKG that used a convolutional neural network to detect electrocardiographic signature of atrial fibrillation during a normal sinus rhythm using a standard 10-second, 12-lead ECG. The neural network was trained on about 450,000 ECGs to identify subtle changes in heart structure in those who developed atrial fibrillation.When the neural network was then tested on a set of patients, a few of whom were diagnosed with atrial fibrillation, it correctly identified subtle signs of atrial fibrillation with about 96 % accuracy.

Atrial fibrillation is associated with an increased risk of stroke and heart failure which highlights the importance of early detection and management.. The use of AI enables screening for early signs of atrial fibrillation in ECGs and hence early detection of the condition.It can also guide treatment of atrial fibrillation.Currently when a person presents with stroke it is important to understand if they had atrial fibrillation before as it guides the treatment plan and the usage of blood thinners for management. AI can provide crucial evidence in these decisions. As described by Dr Paul Friedman, M.D., chair of the Department of Cardiovascular Medicine at Mayo Clinic “AI is like a flashlight that helps you see at night—you still use your eyes, but you can see farther ahead to know what is coming. AI-enabled tools will do the same for physicians, enabling us to detect silent, impending or future disease by interpreting the previously invisible signals our bodies give off all of the time.”

Wearables like iWatch and Kardia have integrated the recording of ECG in the device. The integration of deep learning into wearable technology for intermittent screening for silent AF may be cost effective by preventing sequelae such as stroke. Data from these sources are promising and studies have been conducted on detection of Atrial Fibrillation on these recordings. Further advances in analytic algorithms and sensor technology are needed for their automatic use.

Deep learning has promising applications in detection of electrical abnormalities of the heart.However it is often criticized in the clinical context as a black box. It may also cause over fitting if the data is limited. Poor quality data may also lead to biased prediction.

The challenges of obtaining extensive labelled data and achieving interpretable models must also be addressed.