Arrhythmia Classification Based On ECG Analysis

Arrhythmia is a very common phenomenon in our society, serious arrhythmia will immediately threaten to human life. Therefore, timely checking out the arrhythmia is of great significance for prevention of heart disease and sudden cardiac death. ECG directly records the regularly changing of weak current, in the process of heartbeat, and it is the important tools of diagnosing cardiovascular disease. Because of its outstanding characteristics, non-invasive, rapid and accurate, ECG plays an irreplaceable role in the clinical diagnosisIn recent years, using computer to process data and realizing the rapid automatic analysis of ECG to enhance the efficiency and accuracy of clinical diagnosis of arrhythmia is a hot issues of medical research field. However, due to several reasons, such as noise interference, individual differences, many classifications of arrhythmia, this technique still fails to meet clinical needs recently. This paper does further research on the identification and classification of various abnormal heartbeats, introducing of the abnormal cardiac rhythm analysis, combining wavelet transform analysis and independent component analysis, realizing the automatic classification of nine kinds of heartbeats.First of all, aiming at the frequency interference and baseline drift of ECG signal, median filter and comb filter are used for denoising respectively in this paper. Filter-based method is more stable, which calculates quicker and can get a better result in the real-time processing. Secondly, based on the ECG feature extraction methods in recent years (including the direct methods, the parameter methods, the trans formation methods and the statidtical methods), this paper tries to synthesize three methods:the parameter methods, the transformation methods and the statidtical methods, to extract the feature parameters in different domains respectively and constitute a high-dimensional feature space, in order to express the character of the signal as fully as possible. Thirdly, the high-dimensional feature space can improve the classification accuracy but it also increases the computation of the classifier. Genetic algorithm is used in this paper for optimize the feature space, in which the distance between and within classes is taken for the evaluation function of the fitness. The vectors which could represent the character of signal best would be chosed while the redundancy would be removed. So an optimized feature space with lower dimension but higher resolution is formed. Finally, the LibSVM is used to design a classifier, which would train and test the signal simples of MIT-BIH Arrhythmia Database, and farther classify among the normal ECG and 8 arrhythmias effectively. Compare to the time consumed and classification results before, the dimension is reduced whereas the classification accuracy and calculation speed of the classifier are improved to some extent.