Ultrasound imaging and AI combine to ‘revolutionize’ fetal heart defect diagnosis

Scientists have developed a new approach to detecting fetal heart flaws in the womb utilizing artificial intelligence and imaging to potentially double a doctor’s accuracy at spotting this concern.

University of California, San Francisco, experts trained a series of machine learning models to mimic a physician’s process of diagnosing defects on ultrasound scans. While human providers typically spot as few as 30% of these conditions before birth, the AI model bumped the rate up to 95%, researchers wrote recently in Nature Medicine.

“We hope this work will revolutionize screening for these birth defects,” lead author Rima Arnaout, MD, a cardiologist, professor, and member of the UCSF Center for Intelligent Imaging, said in a statement. “Our goal is to help forge a path toward using machine learning to solve diagnostic challenges for the many diseases where ultrasound is used in screening and diagnosis.”

Fetal imaging is universally recommended by the World Health Organization in the second trimester of pregnancy to both determine the sex and screen for any clinical concerns. Catching cardiovascular concerns early on can help to bolster birth outcomes while further research on possible in utero therapies, those involved noted.

Arnaout and colleagues utilized a three-step model to reach their conclusions. They first deployed an algorithm to find five views of the heart for diagnosis, while a second decided whether the images were normal or not. A third algorithm combined results from the first two to reach a final diagnosis. Training altogether incorporated nearly 108,000 images from more than 1,300 echocardiograms and screening ultrasounds, gathered from 18- to 24-week-old fetuses.

The team tested the AI model on a set of 4,100 fetal surveys. It achieved 95% sensitivity, 96% specificity and 100% negative predictive value in distinguishing normal from abnormal hearts.

“Model sensitivity was comparable to that of clinicians and remained robust on outside-hospital and lower-quality images,” the authors concluded. “Applied to guideline-recommended imaging, ensemble learning models could significantly improve detection of fetal [congenital heart disease], a critical and global diagnostic challenge,” they added later.

Trimed Popup
Trimed Popup