Identification of firing patterns of neuronal signals

Consideration is given to the decomposition of surface EMG (electromyogram) signals into their constituent single-fiber action potentials (SFAPs) and the identification of their firing patterns. This problem in general is analytically not tractable and computationally very complex. However, it is demonstrated that this difficult problem can be resolved by a specially designed neural network with Gaussian nodes. A modified backpropagation algorithm for Gaussian nodes and a novel method of choosing initial conditions are presented. An extensive computer simulation study show that this decomposition method is feasible for the above problem. Such solutions enable a physician or medical researcher to observe the time behavior of SFAPs in a manner suitable for diagnostic purposes or other medical applications.<<ETX>>