A Novel Application for Autonomous Detection of Cardiac Ailments using ECG Scalograms with Alex Net Convolution Neural Network

Abstract

Cardiac Ailments are most wide spread diseases across the world that affects the functioning of the heart. Automatic detection of these diseases plays vital role for immediate diagnosis. Many state-of-the-art algorithms were proposed for automatic classification of cardiac diseases which involve hand crafted features extraction. A Convolutional Neural Network which is pre-trained especially AlexNet with transfer learning model is used in this paper to classify four cardiac ailments i.e., Arrhythmias, Congestive heart failure, Atrial Fibrillationand Normal sinus rhythm which eliminates the needfor feature detection manually.The Electrocardiogram (ECG) signals were considered from MIT-BIH database, 300 signals for each class with a period of 500 samples, a total of 1200 signals for four classes is formed into a single dataset. Scalograms of thesedatasets were obtained using Continuous wavelet transform which were converted to RGB images. These images were resized and given to train transferred AlexNet with appropriate training options. The study evaluates that this method outperforms on test images and gives an overall model accuracy as 93.3%, mean MCC as 91.17% and weighted F₁ scoreas 93.40% with less complexity and time consumption.