Fetal state assessment from cardiotocogram data using artificial neural networks

Abstract

Cardiotocography is the most widely used method in obstetrics practice for monitoring fetal health status. The main goal of monitoring is early detection of fetal hypoxia. A cardiotocogram is a recording of fetal heart rate and uterine activity signals. The accurate analysis of cardiotocograms is critical for further treatment. Therefore, fetal state assessment using machine learning methods from cardiotocogram data has received significant attention in the literature. In this paper, a comparative study of fetal state assessment is presented by using three artificial neural network models, namely the multilayer perceptron neural network, probabilistic neural network, and generalized regression neural network. The performances of the models are evaluated using publicly available cardiotocogram data by running a tenfold cross-validation procedure. The models' performances are compared in terms of overall classification accuracy. For further analysis, receiver operation characteristic analysis and the cobweb representation technique are used.