Practical application of artificial neural networks in the neurosciences

摘要

Abstract: This article presents a practical application ofartificial multi-layer perceptron (MLP) neural networksin neurosciences. The data that are processed arelabeled data from the visual analysis of electricalsignals of human sleep. The objective of this work isto automatically classify into sleep stages theelectrophysiological signals recorded from electrodesplaced on a sleeping patient. Two large data bases weredesigned by experts in order to realize this study. Onedata base was used to train the network and the otherto test its generalization capacity. The classificationresults obtained with the MLP network were compared toa type K nearest neighbor Knn non-parametricclassification method. The MLP network gave a betterresult in terms of classification than the Knn method.Both classification techniques were implemented on atransputer system. With both networks in their finalconfiguration, the MLP network was 160 times fasterthan the Knn model in classifying a sleep period. !20