The use of a neural network system in the identification of motor unit characteristics from surface detected action potentials: a simulation study

Abstract

The surface detected motor unit action potential (MUAP) morphology depends on many physiological and anatomical characteristics of the contracting muscle that are not directly accessible to measurement. In this paper, a neural network based approach is proposed to estimate the motor unit (MU) parameters from a simulated single surface MUAP. We have developed an estimation system that is composed of the following stages: conduction velocity estimation, signal dimension reduction, MU parameters estimation, and number of MU fibres estimation. The parameter estimation stage employs four multilayer neural networks trained on simulated MUAPs corresponding to various ranges of MU parameters. In the estimation mode, this module produces four MU parameters sets. The selected set of the five muscle characteristics is that which minimises an error criterion on a signal reconstructed from the estimated parameters. The proposed system is tested with several simulated MUAPs signals with additive white noise in order to evaluate its performance. It is shown that the technique performs well when the signal to noise ratio is greater than 20 dB.