Comparison of different mother wavelets in PVC detection using PNN

Recently, cardiac diagnosis becomes very important to determine cardiac health condition. Since electrocardiogram (ECG) plays important role in the diagnosis, many classification methods are developed by means of analyzing ECG. The motivation of this study is to research on the optimal wavelet that would accurately classify ECG signal into two distinct classes; normal and Premature Ventricle Contraction (PVC) beats when using Probabilistic Neural Network (PNN). About 35 mother wavelets are used to classify 400 R-to-R intervals of normal and PVC beats. The 400 R-to-R intervals are divided into two groups, Gl and G2. The purpose of this is to inspect their consistency. Two features dataset are set up; one with the ECG time information i.e. R-to-R time ratio as well as another two additional features; average power and energy. Meanwhile, the other dataset are without the time information, average power and energy. Both datasets contain the statistical indices for eight wavelet coefficient (approximation and detail) of level seven. The datasets are then fed into PNN. Metric quantifications are computed to examine the optimal mother wavelet. It is observed that "haar", "db3" and "sym3" produce high accuracy, specificity and sensitivity at detail level 3.