Heart Diseases Classification by ECG Trace Images Using Deep Learning

Recently, many deep learning algorithms have emerged as advanced techniques in the medical field for diagnosing diseases, including heart disease. In this study, an approach was followed that is based on electrocardiogram (ECG) images to detect different heart diseases. Pre-processing was performed for the data images using morphology technology to remove lines from the background ECG paper image to obtain an image containing only the changes of electrical activity for the potion’s heart. The pre-processed data images are trained at a rate of 80% of each class data image in the training stage and 20% of each class image used for the testing stage in the efficiency evaluating stage of each model. Seven classification models have been proposed in binary classification. Models 1–7 have been trained to classify the natural ECG case (Nrm) with the other diseases. Models’ efficiency is calculated using four measures, where the accuracy reaches 100%, the precision reaches 100%, the specificity is 100%, and the f1-score is 100%. For models 6 and 7, the results of the accuracy reached (88.1366 and 91.0978)%, precision (80.7443 and 91.0834)%, specificity (79.1734 and 88.8665)%, and f1-score (79.4476 and 89.8999) %. The proposed diagnostic system is fast, accessible, more sensitive, and harmless. It is also more cost-effective than any other diagnostic method.