A Novel Approach for Identification of Arrhythmia using Transfer Learning Approach

Classification of cardiac issues has been a major topic of research over the decades. But, since 2000s due to the scope of implementation using the concepts of Machine learning, Artificial intelligence that came into persistence due to growth of computational power, lot of scope has been detected for computer automated disease detection, early diagnosis, personalized healthcare. Arrhythmia is one such heart related disease where it would lead to abnormalities in the functioning of heart either due to fast heartbeats often called the Tachycardia or the slow heartbeats often called the Bradycardia. The novel method of arrhythmia detection is using the ECG (Electrocardiogram) which studies the condition of the heart using the electric impulse. In this research we have considered the popularly known MIT-BIH dataset which contains the 48hr continuous diagnosis. Initially we have considered the ECG images, divide them under Train, test, validation splits and have implemented a classifier called EfficientNetB4 using Deep neural networks which attained a training accuracy of 98% and validation accuracy of 91% where the classifier could correctly classify the ECG images under appropriate classes.