| Summary: | 碩士 === 國立中正大學 === 電機工程所 === 98 === In this thesis, we propose a disease recognition system based on HRV signals. Three categories of disease are considered in this study, including the Normal Sinus Rhythm (NSR), Atrial Fibrillation (AF), and Congestive Heart Failure (CHF). We also investigate the caparity of the system in differentrating AF from NSR and CHF from NSR.
The proposed system consists of three parts, including feature extraction, feature selection, and classification. Different categories of features are extracted for the system, including the Wavelet Entropy features of HRV, Spectral entropy features of HRV, First Order Difference features of HRV, RR-interval histogram features of HRV,features of HRV, time and frequency features of HRV and nonlinear features from HRV. Our objective is to find features that are robust, independent to individual variability, and representative of specific disease. The analysis of variance Sequential Forward Selection (SFS) method is applied to select features. Two classifier validation schemes are employed in the classification stage. The first scheme uses half samples for training and half for testing three kinds of classifiers, namely KNN, SVM, BAYES classifier to classify the two diseases and one normal into 3 classes. The second scheme uses leave-one-out method to test the accuracy of the KNN, SVM, BAYES classifiers in classifying the two diseases and one normal into 3 classes.
The results show that the accuracy of the first scheme with SVM classifier outperforms the second scheme as high as 90.71% in accuracy is observed when compared to the 88.27% accuracy achieved by using the second scheme.
Categorize AF from NSR use the first scheme with SVM can reach high accuracy of SVM 96.84% and categorize AF from NSR use the first scheme with SVM can reach an accuracy of 94.83%.
|
|---|
