Myocardial Ischemia Detection with ECG Analysis, Using Wavelet Transform and Support Vector Machines Asie Bakhshipour, Alireza Fallahi, Mohammad Pooyan, Hojjat Mohammad Nejad Biomedical Engineering Department , Shahed University Tehran, Iran afallahi@shahed.ac.ir, pooyan@shahed.ac.ir Abstract— In this paper, we propose a novel method for the detection of myocardial ischemic events from electrocardiogram (ECG) signal, using the Discrete Wavelet Transform (DWT) technique and Support Vector Machines (SVM). The ST-T Segment is obtained based on the detection of R peak location based on the well-known Pan-Tompkins method. Then ratio of energy in the DWT approximation coefficients rather than detail coefficients calculated as the features. SVM is used to build classifiers for ischemic and normal ECG signals. The proposed method achieved correct rate of 98.2%, sensitivity of 98.43% and specificity of 99.45%. Keywords- ECG, Myocardial Ischemia, Wavelet transform, SVM. I. INTRODUCTION Myocardial Ischemia is one of the most common causes of death in the world and early diagnosis and treatment of it has a critical importance. Ischemia is caused by a blockage in the arteries leading to the heart [1]. This type of blockage deprives the cardiac tissue of necessary oxygen. Without oxygen, the cardiac tissue begins to die leading to a myocardial infraction or heart attack. Decrease cardiac oxygenation, effect ventricular repolarization. The damaged cardiac tissue does not depolarize as quickly as the healthy tissue. This causes some of the depolarizing wave to appear during the normally isoelectric ST segment (the time between the S wave and the T wave). If the damage is severe enough, it may even affect the T wave. By using features derived from the modified ST segment and T wave it may be possible to determine if a patient is experiencing ischemia [2,3]. As the ECG analysis is the most accurate, safe and none invasive method in myocardial ischemia detection, several works has been reported in this area. Badilini et. al [4] used frequency domain analysis. They found that ischemic episodes have a lower frequencies term than normal episodes. Stamkopoulos et.al [5] used MLPNN for detection of Ischemia episodes. P.Ranjith et.al [6] used quadratic spline wavelet transform for extraction features of ECG signal and [7] used DCT coefficients and MLPNN for classifying ischemic episodes. In this paper we used ratio of energy in the approximation coefficients of ECG signal as the features. Then we used SVM classifier for classifying normal and ischemic episodes. Figure 1 shows ECG signal analysis flowchart. In the following Figure 1: block diagram of the system sections of this paper, we present the method that are used for ST-T segment detection. Third section describes feature extraction procedure. The classification techniques are discussed in section 4. We argue analysis results in the fifth section. II. ST-T SEGMENT DETECTION Figure 2 shows a complete ECG wave and ST-T segment. In order to detect ST-T segment, first we found peak of the R location. There are several methods [8] have been proposed for detecting QRS complex. We used the well-known Pan&Tompkins method [9] for detection of QRS complex. This method is based on analysis of the slope, amplitude and width of QRS complexes. The algorithm includes a series of filters and methods that perform lowpass, highpass, derivative, squaring, integration, adaptive thresholding and search procedures . Figure.3 illustrates the steps of the algorithm in the schematic form. In the first step the algorithm passes the signal through a low pass and a high pass filter in order to reduce the influence of the muscle noise, the power line interference, the baseline wander and the T-wave interference. After filtering, the signal is differentiated to provide the QRS slope information using the following formula: )] 4 ( 2 ) 3 ( ) 1 ( ) ( 2 [ 8 1 ) ( − − − − − + = n x n x n x n x n y (1) Then the signal is squared point by point making all data point positive and emphasizing the higher frequencies. After squaring, the algorithm performs sliding window integration in order to obtain waveform feature information. A temporal location of the QRS is marked from the rising edge of the integrated waveform. In the last step two thresholds are