Real-time Atrial Fibrillation Detection Using Artificial Neural Network on a Wearable Electrocardiogram

Abstract

Providing equal healthcare quality on heart diseases are an issue in developing countries, especially in Indonesia, due to is wide-spread areas. It is founded that the heart diseases occur not only in big cities but also in rural areas, that is caused by unhealthy lifestyle and foods. Heart disease itself is a disease with gradually symptoms changes that can be seen based on the hearts' electrical activity or electrocardiogram signals. Now, wearable medical devices are capable to be worn daily, so that, it can monitor our heart condition and alert if there is an abnormality. An embedded device worn on the chest can be used to perform a real-time data acquisition and processing of the electrocardiogram, that consists of a 1-lead ECG, an ARM processor, a Bluetooth module, an SD card, and rechargeable batteries. Also, by performing a digital filter and Tompkins algorithm, we obtain the P-wave presences and the heart rate variability values (heartbeat, average heartbeat, standard deviation, and root mean square) then by using an artificial neural network with 4 input, 6 hidden, and 1 output layers that has multi-layer perceptrons and backpropagation. We are able to perform a pre-diagnosis of atrial fibrillation, that is one of the common arrhythmias, from 41 recorded training samples (Physionet MIT/BIH AFDB and NSRDB) and 6 healthy subjects as test samples. The neural network has 0.1% error rate and needed 31548 epochs to train itself for classification the heart disease. Based on the results, this prototype can be used as a medical-grade wearable device that
can help cardiologist in giving an early warning on the user's heart condition, so that it can prevent sudden death due to heart diseases in rural areas.