Volterra neural analysis of fetal cardiotocographic signals

A novel technique for electronic fetal monitoring is presented. It is based on training a Volterra based neural network to classify the fetal states based on a recorded cardiotocography (CTG) data base. The dataset consists of measurements of fetal heart rate and uterine contraction. The CTG signals have 21 attributes and three fetal states, normal, suspect and Pathologic. The Volterra based neural networks (VNN) employ Volterra series expansion for the input vectors and can produce explicit equations describing any multi-input multi-output (MIMO) system. Moreover, VNN has fast and uniform convergence. Simulations have demonstrated the efficiency of the proposed technique in electronic fetal monitoring. The VNN was able to classify the fetal states in a very low number of iterations with negligible error (practically zero). The conventional neural networks, on the other hand, have failed to achieve a reliable convergence.