UC San Francisco researchers have found a way to double doctors’ accuracy in detecting the vast majority of complex fetal heart defects in utero – when interventions could correct them or dramatically improve a child’s chances of survival. – by combining routine ultrasound imaging with machine learning. IT tools.
The team, led by UCSF cardiologist Rima Arnaout, MD, has trained a group of machine learning models to mimic the tasks clinicians follow to diagnose complex congenital heart disease (CHD). Worldwide, humans detect as little as 30 to 50 percent of these conditions before birth. However, the combination of human ultrasound and machine analysis enabled researchers to detect 95% of coronary heart disease in their test data set.
The results appear in the May issue of Natural medicine.
Fetal ultrasound screening is universally recommended during the second trimester of pregnancy in the United States and by the World Health Organization. Diagnosing fetal heart defects, in particular, may improve newborn outcomes and allow for further research into in utero therapies, the researchers said.
Second trimester screening is a rite of passage during pregnancy to find out if the fetus is a boy or a girl, but it is also used to screen for birth defects.. “
Rima Arnaout, study lead author and assistant professor, University of California-San Francisco
Typically, imaging includes five views of the heart that could allow clinicians to diagnose up to 90% of congenital heart disease, but in practice only about half of these are detected in non-specialist centers.
“On the one hand, heart defects are the most common type of birth defect and it is very important to diagnose them before birth,” said Arnaout. “On the other hand, they are still rare enough that their detection is difficult even for trained clinicians, unless they are highly sub-specialized. And too often, in clinics and hospitals around the world, the sensitivity and specificity can be quite low. “
The UCSF team, which included fetal cardiologist and lead author Anita Moon-Grady, MD, trained the machine tools to mimic the work of clinicians in three steps. First, they used neural networks to find five views of the heart that are important for diagnosis. Then they again used neural networks to decide whether each of those views was normal or not. Then a third algorithm combined the results of the first two steps to give a final result indicating whether the fetal heart was normal or abnormal.
“We hope this work will revolutionize screening for these birth defects,” said Arnaout, UCSF Bakar Computational Health Sciences Institute Fellow, UCSF Intelligent Imaging Center and Chan Zuckerberg Biohub Intercampus Research Award Fellow. . “Our goal is to help pave the way for the use of machine learning to solve diagnostic problems for the many diseases for which ultrasound is used for screening and diagnosis. “