Embedded signal processing system for heart disease type classification using neural networks and genetic algorithms

In this paper, intelligent electrocardiogram (ECG) signal processing system is implemented on the ARM processor embedded board and PDA for ubiquitous healthcare (u-healthcare) applications. For effective noise reduction from ECG signal, environmental context based noise reduction method and reconfigurable filter module are developed. To rind optimal combination of filter blocks, estimated noise context is fed to neural network to select the best filter block among pre-designed six filter blocks. Using genetic algorithm (GA), these six filter blocks are reconfigured if the performance of all of pre-designed filter blocks are not acceptable levels. Five kinds of features are extracted from filtered signals and they are fed to another neural network for disease type classification. Filtered signal and diagnosis results from embedded ECG signal processing system are then sent to PDA for patient and PC for patient's physician. Experimental results showed that proposed neural network and GA based embedded ECG signal processing system can diagnosis live types of heart disease with average classification accuracy of 86%. and it can be used as one of practical solution for u-healthcare services.