Data compression using neural network for digital Holter monitor
Yasunori Nagasaka, Akira Iwata, Nobuo Suzumura · 2003
A data compression algorithm for digital Holter recording using artificial neural networks is proposed. Dual three-layer (one hidden layer) neural networks with only a few units in the hidden layer are used. The networks are tuned with supervised signals that are the same as input signals. Back-propagation is used for the learning process. Network 1 performs data compression, and network 2 is learning with current signals. If the waveform of the electrocardiogram changes, network 2 is copied to network 1. The activation levels of hidden layer units are an encoded representation of the input signal waveforms. The original waveforms can be reproduced from the activation levels using the network between the hidden and output layers.>