| Summary: | 博士 === 國立中央大學 === 機械工程研究所 === 96 === Current ECG signal processing methods for feature extraction of QRS complex are mainly focusing on a short-term single heartbeat rather than a long-term multiple heartbeats, in which much useful information is hiding between QRS complexes and only retrievable through long-term monitoring. This dissertation proposes a cepstrum coefficients method to extract the feature from a long term ECG signal. Utilizing this method, one can identify the characteristics hiding inside the ECG signal and classify the signal by its feature vector by DTW (Dynamic Time Warping) and ANN (artificial neural network). First, Normal and Paced-Beat data in the MIT/BIH (Massachusetts Institute of Technology and Beth Israel Hospital) arrhythmia database are used to evaluate the proposed algorithm, and the experiment results have demonstrated successful classification for different cardiac diseases and proved the algorithm to be a fast valid method for long term ECG signal feature extraction. Second, an integrated system for ECG diagnosis that combines cepstrum coefficients method for feature extraction from long-term ECG signals and ANN models for the classification is proposed. Unlike the previous methods using only one single heartbeat for analysis, we analyze a meaningful ECG segment data, usually containing 5-6 heartbeats, to obtain the corresponding cepstrum coefficients and classify the cardiac systems through ANN models. Utilizing the proposed method, one can identify the characteristics hiding inside an ECG signal and then classify the signal as well as diagnose the abnormalities. To evaluate this method, the NOR, LBBB, RBBB and VF types of ECG data from the MIT/BIH database were used for verification. The experiment results showed that the accuracy of diagnosing cardiac disease was above 97.5%. The proposed method successfully extracted the corresponding feature vectors, distinguished the difference and classified long term dynamic signals correctly.
|
|---|
