PS-B07-7: MACHINE LEARNING CLASSIFICATION OF CORONARY HEART SYNDROME BASED ON ELECTROCARDIOGRAM IMAGE USING CONVOLUTIONAL NEURAL NETWORK ALGORITHM

Abstract

Background and Aim: Electrocardiogram (ECG) is one of the most important examination models used to diagnose various abnormal heart rhythms. An EKG records the electrical activity of the heart through wires and electrodes that are attached to the skin of the arms, legs, and chest wall. Early examination of heart defects can improve quality of life through appropriate treatment. One of the abnormalities in the heart is coronary heart disease. CHD is a heart disease that is mainly caused by the narrowing of the coronary arteries due to atherosclerosis or spasm or a combination of both. In this study, a system was designed to classify coronary heart disease from the results of ECG images. Methods: The study consisted of several stages, starting with the collection of ECG image data for cardiac abnormalities which were used for training images and testing images. The next step is pre-processing using a grayscale process to show the intensity value. Furthermore, at the segmentation stage, a thresholding process is used which aims to divide the image into two parts, namely between the object and the background by forming a binary image. The next stage is the process of taking feature values from the segmentation results using Invariant Moment. The final stage is classification using the Convolutional Neural Network. The data used in this study are images obtained from Beecardia physiobank. There are four classifications of images processed in this system, namely Normal, Asymptomatic, Angina Pectoris, and Acute Myocardial Infarction. The amount of data used is 126 where the image is divided into two, namely, 101 training data and 25 testing data. Results: The Convolutional Neural Network method is able to classify coronary heart disease from the results of ECG images very well. The results of the classification process for coronary heart disease from the results of ECG images are able to achieve an accuracy of 92%. Conclusion: The classification system can perform its performance well, as evidenced by the percentage obtained, including a height exceeding 90%. Suggestions for the future, more training data can be used so that at the time of testing, the test data can get higher accuracy.