چكيده به لاتين
Cardiac arrhythmias is a set of irregular heartbeat states. Cardiac signal analysis to identify and accurate arrhythmias classification plays an important role in the early prevention and diagnosis of cardiovascular diseases. The purpose of this thesis is to present a method for classification of cardiac arrhythmias with high accuracy and with minimum number of features based on impulsive properties of cardiac signal by neural networks. For this purpose, cardiac arrhythmia samples are smoothed and denoised by stationary and discrete wavelet transform and decomposed by using the Symlet wavelet family. In the next level, features based on ECG impulsive nature are extracted. Then, according to step 1, 132 samples, including the 4 most common types of cardiac arrhythmias and normal heart rate are classified into five classes by the feedforward neural network, and in step 2, 122 samples from five other types of cardiac arrhythmias are classified. In both classifications, high accuracy rates of 100% and 98.4% were obtained, respectively. In step 3, to evaluate the efficiency of the extracted features in increasing the number of samples and classes, the data of both groups are merged and reclassified. With an increase in the number of data to 254 samples and a doubling of the number of classes (10 classes of cardiac arrhythmia), the accuracy rate was 92.9%. In step 4 to improve neural network performance, the first and second type of learning vector quantization neural network (LVQ1 and LVQ2.1) is used along with the feedforward neural network and the accuracy is increased to 97.2% compared to step 3. Finally, the most effective features are selected and ranked using six evolutionary and metaheuristic algorithms.
