Abstract
In this paper, we develop and test a system for integrating transformed information of the 12 lead stress ECG signals, at the classifier real-valued output level. A coronary artery disease data set was collected and utilized in this study. Four types of features were extracted using the discrete cosine transform, two levels of the discrete wavelet transform, and dimensionality-reduced data using principle component analysis. For each feature type, 12 neural networks were trained and tested using the backpropagation algorithm. Several experiments have been conducted to test this system. Results have demonstrated superior performance when using a fusion of 12 classifier output values, compared to single lead classifier systems. We observed that a 3-level discrete wavelet transform has computed 95-100% performance success rates, using sensitivity, specificity, or accuracy.
Original language | English |
---|---|
Pages | 183-189 |
Number of pages | 7 |
DOIs | |
State | Published - 2014 |
Keywords
- 12 lead ECG system
- Classifier fusion
- Coronary artery disease
- Discrete cosine transform
- Discrete wavelet transform
- Neural networks
- Pattern recognition
- Principal component analysis
ASJC Scopus subject areas
- Modeling and Simulation