Automatic Placental Distal Villous Hypoplasia Scoring using a Deep Convolutional Neural Network Regression Model

Distal villous hypoplasia (DVH) is a placental abnormality associated with fetal growth restriction, which can be assessed in histopathological images of the placenta. In this study, we use a convolutional neural network (CNN) regression model to score DVH in 724 placental images (grade 0 = no DVH; grade 1 = mild to moderate DVH; grade 2 = severe DVH). We applied a patch-wise approach to train the network and scored the whole image by averaging the patch scores. Image scores were compared to a ground truth score obtained by manual scoring by a pediatric pathologist. We obtained a mean squared error of 0.19 and a strong Pearson’s correlation (r = 0.82) between the image score by the CNN and ground truth. Results suggest that this deep learning approach is a promising approach for automated analysis of placental histopathology images.