AI software improves ultrasound detection of congenital heart defects in pregnancy

Key takeaways:

• An AI software system improved ultrasound detection of fetal congenital heart defects compared with unaided interpretation.
• The AI system was also associated with shorter mean reading time of images.

An AI software system trained to detect findings suspicious for fetal congenital heart defects significantly improved sensitivity and specificity compared with unaided reviews of 2D scans, researchers reported at The Pregnancy Meeting.

Congenital heart defects are a leading cause of infant morbidity and mortality partly due to low prenatal detection rates; however, AI systems may help to improve detection on fetal ultrasound exams while also reducing clinician workload, according to Jennifer Lam-Rachlin, MD, assistant clinical professor in the Raquel and Jaime Gilinski department of obstetrics, gynecology and reproductive science at the Icahn School of Medicine at Mount Sinai West and director of fetal echocardiography at Carnegie Imaging for Women.

“About one-quarter of congenital heart defects will warrant some immediate treatment, monitoring or even surgery to improve outcomes,” Lam-Rachlin told Healio. “Congenital heart defects are also the most common birth defect, accounting for about 1% to 2% of pregnancies. The ability to identify that there is a heart defect — that this child will need immediate evaluation or surgery postdelivery — is a big factor in how well that child is going to do. We tried to tackle how to improve prenatal detection and identify babies that should be delivered at a hospital that has ability to take care of an infant with heart defects.”

Researchers utilized a dataset of 200 ultrasound cases from 11 centers in the U.S. and France from singleton pregnancies at 18 to 24 weeks’ gestation, including 100 exams having at least one suspicious finding for a congenital heart defect. The ground truth for presence or absence of each finding was determined by a panel of expert fetal cardiologists; the exams were not used for training the AI system (BrightHeart).

“The AI software evaluates 2D noncolored ultrasound clips of the fetal heart to detect the presence or absence of eight findings that are suspicious for congenital heart defects,” Lam-Rachlin told Healio.

A team of 14 OB/GYNs and maternal-fetal medicine specialists with a wide range of experience (1 to 30 years) reviewed each exam aided and unaided by the AI system, in randomized order, and annotated them for the presence or absence of any abnormal findings. Researchers calculated receiver operator characteristics (ROC) area under the curve, sensitivity and specificity compared with the ground truth analyses.
higher for aided vs. unaided reviews (0.97 vs. 0.83; P = .002).

Findings were similar when assessing sensitivity for aided vs. unaided (0.94 vs. 0.78) and specificity (0.97 vs. 0.76). Results did not change when stratified by patient BMI or by image quality.

“The AI findings significantly improved the readers’ performance,” Lam-Rachlin told Healio. “We also looked at whether it mattered which finding was analyzed. Did one finding overperform? We did not see that with any of the findings. This can help OB/GYNs and [maternal-fetal medicine specialists] improve their detection rate of hearts suspicious for a major congenital heart defect.”

Additionally, researchers found that mean reading time was shorter for aided vs. unaided reviews (226 seconds vs. 274 seconds; P < 001).

“We need to implement this in real-world practice,” Lam-Rachlin told Healio. “In our study, 50% of the hearts had an abnormality, which is not the general incidence in the real population. We want to see if this translates well into clinical practice in a prospective setting.

“As a field, we tend to be a little behind because we are hesitant to implement new technology,” Lam-Rachlin said. “The important takeaway is this does not replace us — it helps us in our day-to-day clinical practice.”

In November, BrightHeart announced it received FDA 510(k) clearance for its AI software. In a press release, the company stated that with the clearance it is “uniquely positioned” to address the challenges of resource constraints and physician shortages in prenatal care.

“In this next phase, we aim to deliver our transformative technology to clinicians and expectant families, making a measurable impact on prenatal care outcomes,” Michael Butchko, chairman of BrightHeart, said in the release.

Reference:

• BrightHeart secures FDA clearance for first AI software revolutionizing prenatal fetal heart ultrasound evaluations. https://www.businesswire.com/news/home/20241115006631/en/BrightHeart-Secures-FDA-Clearance-for-First-AI-Software-Revolutionizing-Prenatal-Fetal-Heart-Ultrasound-Evaluations. Published Nov. 18, 2024. Accessed Feb. 14, 2025.

For more information:

Jennifer Lam-Rachlin, MD, can be reached at jlam@mfmnyc.com; Instagram: @jlamrachlin.