الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 161 |  |  |
| | | from | 161 | | |
Abstract
This thesis has been concerned with the analysis and classification of the heart rate variability (HRV) signals. It is an attempt to utilize some parameters derived from the HRV signals for classification purposes. The main objective was to develop an automatic system that processes, analyses and classifies the heart rate variability signals. The HRV signals were derived from records of the American Heart Association (AHA) arrhythmia database. 120 records were selected each of length 5 minutes. They represent four classes of ECG signals: (1) Isolated Uniform Premature Ventricular Contractions (PVCs). (2) Isolated Multiform PVCs. (3) Ventricular Rhythms (Tachycardia). (4) Ventricular Fibrillation. Autoregressive (AR) modeling was utilized and compared to three other methods, with respect to their efficiency as feature extractor of cardiac abnormalities. The three other feature extraction techniques are: Fast Fourier Transform (FFT), Discrete Wavelet Transform (DWT), and Wavelet Packet (WP). For the classification stage, a multi-layer feed forward neural network learned and trained using backpropagation algorithm. The “hold-out” method used to estimate the propability of error of the classifier. The results have shown that the highest classification rate was obtained when using the DWT as a feature extraction technique, it reaches 100% correct classification for all classes. The proposed system which provides non-invasive evaluation of the heart may be useful for the prediction of cardiac abnormalities which might be valuable for clinical use. Key Words: Electrocardiogram - Signal Processing - Autoregressive model - Fourier Transform – Wavelet - Wavelet Packets - Neural Networks.