Carnegie Mellon University (CMU) researchers have developed a machine learning technique that scans placenta samples for signs of health risks in future pregnancies.
The system aims to assist the work done by doctors, who sometimes scan the placenta to look for signs of complications in the subsequent pregnancy.
Among the biggest warning signs are blood vessels with called lesions decidual vasculopathies. Their presence suggests that a mother may be suffering from pre-eclampsia, a condition that creates complications in 2-8% of pregnancies and which can be fatal for both mother and baby.
If these lesions are detected early, the condition can be treated before symptoms appear. However, because the exam requires a lot of time and highly specialized skills, it is usually conducted rarely.
Complications in pregnancy: a new approach based on AI
Carnegie Mellon's approach aims to make the evaluation more accessible by automatically obtaining data on possible pregnancy complications from the placenta with AI.
“Pathologists train for years to be able to find disease in these images, but there are so many pregnancies going through the hospital system that they don't have time to inspect every placenta,” the researcher said Daniel Clymer.
Our algorithm helps pathologists know which images they should focus on by scanning an image, localizing blood vessels, and finding blood vessel patterns that identify decidual vasculopathy.
Daniel Clymer, Carnegie Mellon University
How the system works
The team has trained their own system artificial intelligence to locate lesions by providing images of placenta samples.
The image A shows a placenta slide with a blue square indicating a single blood vessel. B shows a healthy blood vessel. C shows the effect of decidual vasculopathy: hypertrophic smooth muscle around the lumen of the blood vessel. This can cause complications in pregnancy.
The system first detects all the blood vessels in the image, then determines whether each individual vessel is healthy or not.
The algorithm also evaluates and correlates various characteristics of the pregnancy, such as gestational age and any conditions of the mother. If the system detects any abnormalities, it marks the placenta as diseased and warns of possible complications in pregnancy.
During testing, the algorithm classified lesions more accurately than professional pathologists.
Future applications
Researchers don't expect the system to completely replace medical professionals. Instead, they want it to mark areas that pathologists should take a closer look at.
Ultimately, they hope it will reduce the cost of the test, opening up access to more mothers and mothers-to-be, ensuring they have a complication-free pregnancy.
you can read the complete research paper free in The American Journal of Pathology.