Neural network techniques for a physiological rooted analysis of auditory brain stem average evoked responses (ABSR)

Neural network techniques are proposed to identify the parameters of a mathematical model, rooted on physiological knowledge, which fits an auditory brain stem average evoked responses (ABSR). Fitting should be performed in order to minimize the mean square error between the model and the actual ABSR. Model is implemented by a linear combination of five nonorthogonal functions. Each element k of this `basis´ is defined to formally represent the global postsynaptic activity at the nuclei of the auditory pathway. Fitting is done using an enhanced backpropagation method. The learning set is composed of filtered/synthesized ABSRs. Results shows that the algorithm converges after circa 200 epochs of training for a sum of square error of 0.0005