Tissue ischemia monitoring using impedance spectroscopy: evaluation of neural networks for ischemia estimation

Abstract

Tissue impedance spectra and pH values, collected during ischemic episodes in human skeletal muscle, were used to train and test Artificial Neural Networks (NN) for ischemia level estimation. The goal was to determine the NN with optimal performance in classifying impedance spectra and their corresponding pH values when varying levels of noise were introduced to the original signal. The performance of two linear associative memory NNs (Hebbian and ADALINE) and the backpropagation (BP) NN were evaluated using impedance spectra in the frequency range from 25 Hz-500 kHz as inputs and the pH values as outputs. Results indicate that a BP NN with a single hidden layer and moderate numbers of neurons is an optimal solution for the authors' research.