Design of a System for Diagnosing Cardiac Arrhythmia based on Maximal Overlap Discrete wavelet transform(MODWT) and Long_Short Time Memory (LSTM)

الملخص

In this research, a system was designed and  Implemented that diagnosis 12 cases of cardiac arrhythmia in addition to the normal condition, where this system is consisted of a hardware section and a software section. The hardware section is the ECG electrical electrographic cardiac signal acquisition circuit. The software is the interface that analyses ECG signal (either from the circuit or from an external file) and then disease is diagnosed. The diagnosis was made using one of deep learning methods, which is the neural network with Long_Short term memory (LSTM), after choosing it as a better classifier. This research has been accomplished according to several stages. In the first stage, the signal has been preprocessing using Savitzky_Golay filter .In the second stage the features were extracted in the time and time_frequencey domains. First, in the time domain, the peaks of the ECG signal were determined  by applying Maximal Overlap Discrete Wavelet Transform (MODWT) with 5th level and sym4 wavelet transformation, from which 16 features of ECG signal were obtained using statistical  analysis. Second in the time_frequency domain 8 features of ECG signal were extracted by obtaining the approximation and detail coefficients of ECG signal and then applying statistical  analysis of them. In the third stage, the features in the time domain and (time_frequency) domain were used as inputs of two same classifiers type. The performance of classifier depending on features in time domain was better than the performance of classifier depending on features in (time_frequency) domain, where the accuracy, Sensitivity and specificity of first classifier were respectively: 97.77%, 97.36%, 100%, while the other classifier were 95.55%, 94.73%, 100% respectively. The classifiers were trained and tested with a database MIT_BIH of 118 ECG signals.