Prediction of Fetal Blood Pressure during Labour with Deep Learning Techniques

Abstract

Our objective is to develop a model for the prediction of minimum fetal blood pressure (FBP) during fetal heart rate (FHR) decelerations. Experimental data from umbilical occlusions in near-term fetal sheep (2698 occlusions from 57 near-term lambs) were used to train a convolutional neural network. This model was then used to estimate FBP for decelerations extracted from the final 90 min of 53,445 human FHR signals collected using cardiotocography. Minimum sheep FBP was predicted with a mean absolute error of 6.7 mmHg (25th, 50th, 75th percentiles of 2.3, 5.2, 9.7 mmHg), mean absolute percentage errors of 17.3% (5.5%, 12.5%, 23.9%) and a coefficient of determination R2=0.36. While the model was unable to clearly predict severe compromise at birth in humans, there is positive evidence that such a model could predict human FBP with further development. The neural network is capable of predicting FBP for many of the sheep decelerations accurately but performed far from satisfactory at identifying FHR segments that correspond to the highest or lowest minimum FBP. These results indicate that with further work and a larger, more variable training dataset, the model could achieve higher accuracy.