Application of Kohonen neural network for the elaboration of electrogastrograms

A method is presented for processing and analysis of electrogastrography (EGG) a noninvasive technique by which gastric myolectrical activity is recorded using abdominal surface electrodes. The analysis is based on FFT and on unsupervised artificial neural networks. Three kinds of patterns can be identified on the neurons of a Kohonen output map with 32 x 16 neurons: one relating to noisy spectral profiles, one relating to pre-prandial profiles, one relating post-prandial profiles. It is concluded that the described method is reliable and can be used for objective automated analysis of EGG and for investigation of possible relations of the EGG with gastric pathologies.