Electrocardiogram Feature Recognition Algorithm with Windowing and Adaptive Thresholding

Heartbeat abnormalities of human body can be diagnosed by observing electrocardiogram (ECG) signal. Traditional methods of analyzing ECG signals to determine a person's abnormality are based on the expertise of cardiologists, where sometimes multiple interpretations or misinterpretations of the disorder occur. The development of pattern recognition methods nowadays have rapidly advanced so that make it possible to be applied to ECG signal. Certain feature of ECG required for pattern recognition are P, Q, R, S and T signals. In this paper, we propose a pattern recognition method for ECG features by using adaptive threshold to find P, Q, R, S, and T position. First, we find R signal defined by local peaks, P and T signal which are defined by maximum value from a specific window, S signal defined by local valleys of the ECG signal and the rest Q signal which is defined by minimum value between P and R signal. Then based on those information, we use 48 ECG signals that contain abnormality and 18 normal ECG signals from physionet database. Experimental results show that the accuracy level of our method to recognize P, Q, R, S and T signals are 96,52%, 95,88%, 96,56%, 98,35%, and 95,88% respectively for both normal and abnormal ECG signal.