Multiparametric analysis of fetal heart rate: comparison of neural and statistical classifiers

Heart Rate Variability analysis has demonstrated as a powerful diagnostic toot in many disease conditions which involve an alteration of the physiological control systems. In this paper we propose to classify Fetal Heart Rate signals through a set of indexes including time domain, frequency domain and other parameters related to signal morphology and regularity. This set is used as the input of an automatic system, whose goal is to detect the risk for the fetus to enter a pathological state. On a database of more than 400 recordings, we tested different classification methods to identify normals from potential pathological fetuses. A neural network approach was compared with classical statistical methods. The multilayer pereeptron, trained with the adaptive backpropagation algorithm, performed better than any tested statistical classifier.