Disease Identification in Histopathological Imaging Using Multi-Resolution Hierarchical Convolutional Neural Networks
After a child is born, the examination of the placenta by a pathologist for abnormalities such as infection or signs of oxygen deprivation can provide important information about the immediate and long-term health of the infant. Moreover, pathological detection of a placental blood vessel lesion called decidual vasculopathy (DV) has been shown to predict pre-eclampsia adverse outcomes in subsequent pregnancies. However, due to the high volume of deliveries at large hospitals, current pathological workflow only performs this inspection on a small percentage of delivered placentas, while the rest are discarded without inspection. We introduce a hierarchical machine learning (ML) approach for the automated detection and classification of DV lesions, along with a method of pooling learned image features from many blood vessels along with patient metadata in order to predict the presence of disease at a global level using a hierarchical CNN deep learning algorithm.
Ultimately, the proposed framework will allow many more placentas to receive inspection in a more standardized manner, providing feedback as to which infants would benefit most from more in-depth pathological inspection. This examination will also allow for real-time adjustment to infant care and possible chemoprevention (e.g. aspirin therapy) to prevent pre-eclampsia in women identified to be at-risk with future pregnancies.