The placenta is a window into the events of pregnancy and the health of the mother and baby, yet only about 20% of placentas in the US are assessed by pathology exams and placental data is often neglected in pregnancy research. Since both the mother and fetus contribute to and modulate placental development and function, data from placental examination may inform short- and long-term clinical care of both mother and child. Placental pathology remains under-used due to the time, cost, expertise, and facilities needed, even in high-resource settings. Placental assessment can and should be more accessible to pathologists, clinicians, and researchers, and assessment at birth can more readily aid clinical decisions and relate findings to patients.
Prior work has used photographic images to measure characteristics such as shape and cord coiling and related these characteristics to placental diagnoses and outcomes of clinical importance. This project aims to leverage the simplicity and low cost of digital photographs and the computational and decision power of recent advances in artificial intelligence (AI) to create software for comprehensive placental assessment from images of gross placentas. The software could address the need for widespread, simple placenta assessment, particularly when information is needed urgently, pathologists are not highly trained for placental pathology, or where resources only allow a small fraction of placentas to be reviewed.
The investigative team, with extensive expertise in placental pathology and research, clinical care, medical informatics/AI, and image understanding, has developed an initial prototype with promising results for predicting several clinically impactful diagnoses. Our preliminary data demonstrates that a wide range of data can be collected from placental photos and that computational techniques allow the connection of abstracted data to predict placental disease. The goal of this proposal is to develop and validate software to assess placentas from digital photographs in any delivery setting.
An extensive, first-of-its-kind dataset will be created from three large hospitals including images and expert pathology reports from pregnancies with abnormal and healthy outcomes (n>50,000). Combined, these sites include a range of characteristics across income, race/ethnicity, health risks, and hospital resources. The resulting software will glean visual characteristics from the disc, cord, and membranes and accurately identify specific parameters (e.g., abnormal shape) and diagnoses (e.g., chorioamnionitis). This software has the ability to strengthen pathology exams by standardizing and enhancing the data collected at the gross level, providing better information to pathologists for diagnoses. The immediate information could impact clinical care before hospital discharge, and ease-of-use will allow inclusion in pregnancy research.
With such huge advances in technology, placental assessment at birth can no longer be viewed as nonessential or too difficult. When fully developed and validated clinically in a range of birth settings, this software could have the power to impact the care of millions of mothers and children around the world.