Neural networks to estimate the influence of cervix length on the prediction of spontaneous preterm delivery before 37 weeks

Abstract

Neural networks were applied in an effort to predict the risk for early spontaneous preterm delivery using various demographic, clinical, and laboratory inputs. Furthermore, attention has been focused on the influence of cervical length (CL) for the prediction of spontaneous preterm delivery. Data for 59,313 cases of pregnant women were collected and processed. The final data used were those that were considered to offer clear indication on the significance of cervical length on the prediction. The cervical length was measured by sonography in the range of 22-24 weeks of gestation. Preliminary results showed a prediction rate of approximately 65% was attained through the application of a variety of neural network topologies. It has been found that if the cervical length is excluded from the input data, this results in an approximately 10% decrease in the prediction yield, as obtained from the neural network predictor, thus the sensitivity to cervical length is quite significant.