Pattern recognition of physiological signal for the identification of arrhythmia
Loading...
Date
2017-06
Authors
Loo Wei Lung
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Cardiac arrhythmia can be hard to diagnose, especially the types that only causes symptoms every once in a while. This project presents an alternative way to identify cardiac arrhythmia by using physiological signals. The medical devices outputs high rate of false alarms due to low accuracy of the product. The physiological signals involved in this project are ECG, ABP and PPG signals. This study is decomposed into six stages, which are arrhythmia data acquisition, implementation of feature extraction algorithm, exploratory of data analysis, one-way ANOVA statistical analysis, development of ANN arrhythmia classifier and finally performance evaluation for the best input features set and most optimum ANN model. The feature extraction method of ECG signal is Pan-Tompkins algorithm while ABP and PPG signal are using PhysioNet open-source algorithms. The extracted features are divided into two sets and further analysed by using one-way ANOVA to obtain the statistical analysis of features with significant difference for cardiac arrhythmia classifier. This study aims to obtain the best input features set and the most optimum ANN model for cardiac arrhythmia classification. The classifier is consists of two stages, where the first stage classifies the input features to normal or arrhythmia and the second stage classify the input features to four types of arrhythmia. The performances of the developed ANN classifiers is evaluated to determine the best input features and the most optimum classifier. The overall accuracy achieved by the first stage classifier is 81.6% and the second stage is 89.5% with second input features set and training algorithm trainscg.