Pilot study of the capabilities of neural network data analysis in predicting placental disorders: A prospective study

Background. Placental disorders underlie the development of a large number of pregnancy complications, such as growth retardation, fetal hypoxia and distress, preeclampsia, etc. Fetal hypoxia occurs in 10% of all pregnancies and is the cause of perinatal losses in 40% of cases. Uteroplacental hypoxia is associated with impaired placental formation in early pregnancy and its angiogenesis in later stages. Meanwhile, there are currently no technologies that can predict the development of placental disorders with a high degree of probability.

Aim. To evaluate the capabilities of neural network data analysis in predicting placental disorders.

Materials and methods. The prospective analysis of the features of the course of 99 pregnancies was conducted. Based on the results of the study, 2 groups were formed: the control group included 51 patients whose pregnancy was not complicated by the development of placental disorders, the main group included 48 patients whose pregnancy proceeded against the background of placental disorders.

Results. The technology for predicting placental disorders is implemented on the basis of the multilayer perceptron, the percentage of incorrect predictions during the training of which was 7.1%. The structure of the trained neural network included 8 input neurons, which were the parameters included in the Astraia protocol (height of the pregnant woman, coccygeal-parietal size, thickness of the collar space and heart rate of the fetus, pulsation index in the uterine arteries, the content of β-hCG and PAPP-A in the blood of the pregnant woman), as well as the volume of amniotic fluid.

Conclusion. An integrated approach based on neural network analysis of study parameters available for widespread clinical practice (Astraia protocol), as well as amniotic fluid volume, should be considered promising for predicting placental disorders due to its high information content (Se=0.87, Sp=0.98, ROC-AUC 0.921±0.04 [95% CI 0.843–0.998]; p<0.001). In our opinion, the use of this technology will be useful for identifying patients at risk in order to prevent the development of placental disorders and will reduce the incidence of adverse perinatal outcomes.