Research On Methods Of Analysis And Recognition Of Electrocardiosignal

With people living a better life, the incidence of cardiovascular diseases, which is named as “the first killer”, is getting higher and higher. While, to classify the electrocardiogram(ECG) in the traditional way, can make a great of mistakes, as result of the tiredness of medical workers. Automatic diagnosis of electrocardiogram(ECG), can relief medical workers, thus efficiency of arrhythmia classification can make a improvement. In this paper, a new method of arrhythmia classification has been proposed, which is based on the higher order statistics(HOS) and decision table classification.According to the procedure of automatic diagnosis, this paper introduce each of these steps, from preprocessing to arrhythmia classification. The electrocardiogram signal extracted from patient, need preprocessing to cancel power frequency interference and the baseline drift. As the electrocardiogram used in this paper comes from the MIT-BIH arrhythmia stander database, this paper doesn’t introduce the procedure of preprocessing and the detection of QRS in detail. The research this paper do are as follows:First, there are two principles in the feature extraction, for one is to make the individual difference between the same type of data small, for another is to make the individual difference between the different type of data big. The feature based on the original ECG individual difference is big, this is not good for classification. Thus, this paper adopt the method of the higher order statistics, which transform the original ECG signal to the three cumulants(the second cumulant、the third cumulant and the fourth cumulant). After plotting at matlab, it is easy can see from the drawing that the individual difference getting smaller between the same type of data.Second, the feature extracted by the way of using HOS, combine the feature of RR interval of the original ECG signal, a feature vector which contain 18 elements is acquired. In this paper, a method named decision table, which is based on the rough set theory, has been proposed. Using the 18 elements, which is acquired in the procedure of feature extraction, as the condition attribute of the decision table. As the decision attribute of the decision table is the 5 types of ECG signal, according to the standard of AAMI. Using the data from the MIT-BIH database as the training sample, to simplify the decision table. The decision table after simplification, is used to classify ECG signal。Third, the result of using decision table to classify ECG signal works well, when the training sample is large enough. In this paper, 1000 normal beat, which acquired from the recording 101 from MIT-BIH database, are used as training sample, and the following 500 normal beat as testing sample. The accuracy of the decision table can come to 90.2%.