Improving Maternal and Infant Health Conditions through Placenta Image Analysis

2019-12-03

Placenta can provide important information about the health of both the mother and the baby, but in the United States, only 20% of placentas undergo pathological examination after delivery. Now, researchers have developed a new solution that can generate accurate and automated placental diagnostic reports through computerized image analysis. Their research allows for the examination of all placentas, reducing the number of comprehensive pathological examinations for normal placentas, and creating a less resource-intensive research analysis pathway that is expected to have a positive impact on maternal and infant health. Alison Gernand, an assistant professor of nutrition science at the College of Health and Human Development at Pennsylvania State University, said, "The placenta drives everything for the mother and the baby, but we lack 95% of global birth placenta data. If we can create a more efficient testing process, it will allow us to collect more comprehensive data to examine the relationship between the placenta and maternal and fetal health outcomes, helping us examine the placenta in minutes instead of days." This new technology uses artificial intelligence to analyze images of each side of the placenta and generate a report that contains key information that may affect maternal and infant clinical care, such as whether the fetus is receiving sufficient oxygen in the uterus or if there is a risk of infection or bleeding. Currently, there is no clear standard for placental testing, and low-income countries and areas where childbirth is more common often lack resources for basic placental analysis. This digital tool provides a solution that allows individuals to perform simple placental testing with just a smartphone or tablet equipped with appropriate software. Gernand explained, "Even in resource-limited areas, people usually have a smartphone. Our goal is for midwives to take photos, analyze them with authorized software, and immediately provide information to care for the mother and baby." To establish this system, researchers analyzed 13,000 high-quality placental images and corresponding pathological reports from Northwestern Memorial Hospital. Then, they labeled a set of training images with data points crucial for understanding the placenta, such as incomplete areas and umbilical cord insertion points. These images were used to train a neural network using CPU and GPU servers, which can automatically analyze new placental images to detect features associated with abnormalities and potential health risks. Their system made effective predictions on unlabeled images, and comparison with the original pathological reports showed that the system has high accuracy and clinical potential.

RECENT AI NEWS

RECENT AI TOOLS