An investigation of non-invasive techniques for the estimation of conduction velocity distributions in skeletal muscles and nerve bundles

The mean muscle/nerve conduction velocity is a clinically valuable indicator in the diagnosis and assessment of neuromuscular disorders. The conduction velocity distribution (CVD) has the potential of providing more information to help assess these pathologies. Several techniques for the estimation of nerve conduction velocity distributions have been proposed in the literature. Most of these techniques do not make use of accurate models for the electrical signal picked up at the skin surface. This has led to the implementation of biased estimators. On the other hand, those that describe the evoked nerve signal through physical models have had difficulties in the estimation of the electrical source needed for the CVD estimation techniques proposed.; Two non-invasive CVD estimation techniques are presented in this work, one for nerve bundles and another for skeletal muscles. Both estimators are based on signal models developed with the use of volume conduction theory. The extracellular potential originated by a single active fiber is expressed as the convolution of a source that can be considered independent of velocity with a tissue filter impulse response function that accounts for the velocity dependence. This representation leads to the use of a suitable deconvolution technique to find the CVD estimate. The deconvolution technique consists of solving a minimization problem and does not require availability of the source or term independent of velocity.; The performance of the estimators proposed is evaluated through simulated and experimental data. They are also compared to previous estimators proposed in the literature. The nerve CVD estimator makes use of two somatosensory evoked responses and clearly outperforms its predecessors. On the other hand, the muscle CVD estimator, which uses two correlation functions of voluntary myoelectric signal, is sensitive to errors in the model parameters. Thus, it does not offer a significant improvement over its predecessor.