| Among present ECG (Electrocardiogram) algorithms, P wave detection is harder than QRS detection since it is smaller and susceptible to interference. The scare of related database also blocks P wave detection algorithms' development. However P Wave's information is important for cardiovascular diseases diagnosis. Therefore there will be practical value in improving P wave detection algorithm.The basic considerations in P wave recognition research are standard ECG samples' analysis, finding a certain rule for these P wave and detecting it using the rule. To get standard P waves data, the MIT-BIH datasets are used and P wave in these sets are checked by clinical specialist. Each P beats were verified and commented. Then, Support vector machine is adopted in sample training and waves' classification.For each MIT-BIH data set (30 minutes), 20minutes' P wave data should be used in training and others for testing. One dataset will generate about 6000 samples for SVM(Support Vector Machine) training. One Train sample consists of positive and negative parts. Positive parts are derived at PWavePeak point (PWavePeak-Samplelenl2~PWavePeak+Samplelenl2). Negative samples are derived at PWavePeak-3, PWavePeak-6, PWavePeak+3 and PWavePeak+6 points. The PWavePeak meas the peak point of P wave. Samplelen is the length of P waves from 20 to 100 points. To reduce the complexity and algorithm's loading, a preprocess method based on Matlab Cluster is applied for train data sets. After training and regression check, these SVM parameter will be used in actual testing. To get better P waves detection sensitivity and specialty, different core (Poly, RBF and Sigmoid) functions and P wave length are evaluated. The testing results are compared with ECGPUWAVE's report.From the result, some conclusions can be drawn that in complex case original ECG information is not suitable to be used in SVM training and detection. Suitable length of samples is also a very important factor that can improve the performance the algorithm. With suitable filtering and better characters, a sample SVM algorithm can detect P waves as the ECGPUWAVE tool for good quality ECG, with average SE being 89% and SP being 81% respectively.In addition, a graphic-based tool is developed for ECG wave viewing and editing. It can read MIT-BIH's file and help ECG researchers develop new algorithm easily. Furthermore, there are still two issues for P detection and SVM application. One is setup a standard Database for P wave. There is no open database in china so far. The other is testing the SVM during the course of T waves detection. |
