Researchers from Carnegie Mellon University and the University of Pittsburgh Medical Center have created an artificial intelligence (AI) algorithm that can examine placenta slides and detect health problems. The algorithm helps pathologists understand which images to focus on. The algorithm scans an image, locates blood vessels and finds patterns of blood vessels that identify disease.
Placentas are examined after birth to look for health risks that could indicate the mother may have unhealthy pregnancies later. In the placenta, pathologists look for a type of blood vessel lesion called decidual vasculopathy (DV). DV indicates that a mother is at risk for preeclampsia in future pregnancies. But there are hundreds of blood vessels on a single slide of the placenta, and only one diseased vessel is needed to indicate risk. This is a time consuming and difficult task that takes pathologists years of training to detect.
![A digitized whole slide image of a placental membrane roll [low magnification; haemotoxylin and eosin (H&E) stain]. To illustrate relative scale, the blue square indicates a single blood vessel. Source: College of Engineering, Carnegie Mellon University](/images/assets/641/15641/241921_web.jpg)
A digitized whole slide image of a placental membrane roll [low magnification; haemotoxylin and eosin (H&E) stain]. To illustrate relative scale, the blue square indicates a single blood vessel. Source: College of Engineering, Carnegie Mellon University
The algorithm was trained on images of a thin slice of a placenta sample. These images were labeled diseased or healthy.
A novel approach was used to enable the algorithm to follow a series of steps to make a difficult task more manageable. In the first step, the algorithm detects all blood vessels in the image. Then, each blood vessel is analyzed individually to create smaller data packets for the algorithm. Finally, the algorithm analyzes each blood vessel and determines if it is diseased or healthy. If the algorithm finds a diseased vessel, then the picture is marked and a pathologist will check the image.
The algorithm was not designed to replace pathologists. Rather, it could be used as a pathologist’s assistant to speed up the diagnosis process.
A paper on this technology was published in the American Journal of Pathology.
