earticle

영어

This paper develops a novel framework for feature extraction based on a combination of Linear Discriminant Analysis and cross-correlation. Multiple Electrocardiogram (ECG) signals, acquired from the human heart in different states such as in fear, during exercise, etc. are used for simulations. The ECG signals are composed of P, Q, R, S and T waves. They are characterized by several parameters and the important information relies on its HRV (Heart Rate Variability). Human interpretation of such signals requires experience and incorrect readings could result in potentially life threatening and even fatal consequences. Thus a proper interpretation of ECG signals is of paramount importance. This work focuses on designing a machine based classification algorithm for ECG signals. The proposed algorithm filters the ECG signals to reduce the effects of noise. It then uses the Fourier transform to transform the signals into the frequency domain for analysis. The frequency domain signal is then cross correlated with predefined classes of ECG signals, in a manner similar to pattern recognition. The correlated co-efficients generated are then thresholded. Moreover Linear Discriminant Analysis is also applied. Linear Discriminant Analysis makes classes of different multiple ECG signals. LDA makes classes on the basis of mean, global mean, mean subtraction, transpose, covariance, probability and frequencies. And also setting thresholds for the classes. The distributed space area is divided into regions corresponding to each of the classes. Each region associated with a class is defined by its thresholds. So it is useful in distinguishing ECG signals from each other. And pedantic details from LDA (Linear Discriminant Analysis) output graph can be easily taken in account rapidly. The output generated after applying cross-correlation and LDA displays either normal, fear, smoking or exercise ECG signal. As a result, the system can help clinically on large scale by providing reliable and accurate classification in a fast and computationally efficient manner. The doctors can use this system by gaining more efficiency. As very few errors are involved in it, showing accuracy between 90% - 95%.