Personalized ECG signal classification using block-based neural-network and particle swarm optimization

The purpose of this paper is the classification of ECG heartbeats of a specific patient in five heartbeat types according to AAMI recommendation, using an implementable neural network such as Block-based Neural Network (BBNN). A BBNN is created from 2-D array of blocks that are connected to each other and easily can be expanded. Each block is a neural network. Because of flexibility in structure and internal configurations of BBNN, we can implement that with a reconfigurable digital hardware such as field programmable gate array (FPGA). The internal structure of each block depends on number of incoming and outgoing signals. Therefore, the overall construction of network is determined by the moving of signal through the network blocks. Network structure and the weights are optimized using particle swarm optimization (PSO) algorithm. Input of the BBNN is a vector that the elements of this vector are the features that extracted from ECG signal. In this paper wavelet transform based features and temporal features that extracted from ECG signals create the input vector of BBNN. ECG signals are time varying and also for different people are unique. The BBNN parameters have been optimized by PSO algorithm witch can overcome the possible changes of ECG signals. The performance evaluation using the MIT-BIH arrhythmia database shows a high classification accuracy of 97 %.