Determination of stenotic coronary artery infarction localization with artificial intelligence techniques

According to the World Health Organization (WHO), cardiovascular diseases caused about 17,1 million people to lose their lives in 2012 worldwide. It is estimated that this number will reach 22,2 million by 2030. According to the cause of death statistics for the Turkey Statistical Institute in 2016, our country circulatory system caused 162 876 deaths had occurred. Among the causes of death, circulatory system diseases are the leading cause with 39.8%. In this study, it is aimed to develop a physician assisted specialist system in order to detect coronary artery occlusions in early stage and decrease myocardial infarction (MI) risk by applying digital signal processing and artificial intelligence techniques on 12 lead electrocardiogram (ECG) signals. MI, which means that part of the heart muscle is necrosis, the formation of cholesterol and fat plaques on the inner walls of the MI coronary arteries results in a region of the myocardium that can not be adequately fed to the bloodstream. The diagnosis of MI is based on changes in typical chest pain, ECG, and plasma enzymes. The initial findings of necrosis beginning within the first 40 minutes after the narrowing and occlusion of the heart vessels can be followed by changes in the ECG. Depression or elevation of the ST segment in the ECG reflects the infarction. In the study, different heart rate, amplitude and ECG signals with ST segment elevation/depression were recorded with different infarction severity and localization for MI early diagnosis and infarction localization. 12 channel ECG signals were normalized and power spectral density values were obtained with Welch Method. The first 15 parameters of the power spectral density are specified as the feature vector. The attribute vector is taken as the input parameters of the artificial neural network. The power spectral density values of 340 12-channel ECG data were used for learning in the artificial neural network and the network was tested with 60 data that were not introduced to the network. These 60 test data classified 99.85% correctly by the artificial neural network. The completed artificial neural network was tested with a total of 1600 data sets, 200 normal for each infarct location (Anterolateral, Anteroseptal, Anterobasal, Posteroinferior, Posteroseptal, Posterolateral and Posterobazal) and 200 for normal ECG data from the scenarios formed from normal, ST segment elevation and ECG signals with collapsed ECG signals. As a result of the test, 1600 data sets were successfully classified with 99.94% success.