Research On Heterogeneous Databases And Diagnosis Algorithms Of Electrocardiogram

ECG(Electrocardiogram) has been used in clinical medicine since the 19 th century. It has always played an important role in the process of disease diagnosis. By summarizing and analyzing the previous researches of signal processing, the fact that there are abundant researches of signal processing for EC G can be concluded, but few studies have involved for the problem of heterogeneous database. The heterogeneous database is often applied in reality, so the study of heterogeneous database becomes very necessary and valuable. In addition, neural network algorithm is widely used in EC G recognition field, especially the BP neural network(BPNN). However, when the sample size is too large, the discrimination will become smaller based on classification algorithm of BPNN. It will inevitably lead to inaccurate classification, so it is necessary to solve the defect of BPNN.For the two above problems, this dissertation put forward a set of pretreatment methods to solve the application obstacles about heterogeneous database s and multistage classification algorithm based on BPNN. By a full investigation for EC G diagnosis process, this dissertation proposed a set of pretreatment methods which includes de- noising and sampling frequency conversion for the heterogeneous database. Combining w ith three heterogeneous EC G databases, the research completed a contrast of similarity by entropy value, standard deviation, kurtosis and skewness. And the result of experiment shows that the scheme has a good effect. In addition, an experimental simulation was finished about disease diagnosis for MIT-BIH arrhythmia database and PTB database, and the experiment shows that disease diagnosis can be achieved after heterogeneous database preprocessing. In order to solve the defect of BPNN algorithm, mult iple training networks were obtained by training mult istage network on mult iple false testing sets. Finally, multiple network s were applied for evaluating testing set. An experiment was completed by combining w ith five types of heart beat in MIT- BIH arrhythmia database, and the recognit ion rate reached 96.8%, which was about 5% higher than the common BPNN. What’s more, another experiment was completed that using SVM algorithm combined with the feature of wavelet and waveform, as well as good literature method. These experiments showed that the effection of the multistage algorithm based on BPNN is better than other methods.After the completion of the previous research, this dissertation implemented the EC G-based arrhythmia diagnosis system on the basis of previous work. The diagnosis system uses Android smartphone as a carrier for the client, and ECG data was collected by a portable device. Then smartphone transmits the data to the server by wireless network. The EC G data will be automatic processed, and the result of diagnosis is sent back to the smartphone as reference for customers.