Jennifer Lam-Rachlin, MD, discusses AI software to screen for major congenital heart defects

While congenital heart defects (CHDs) are among the most common birth defects, they are also frequently missed during routine prenatal ultrasounds, according to Jennifer Lam-Rachlin, MD, MFM, Carnegie Imaging for Women PLLC; Assistant professor, Maternal Fetal Medicine Division, Mount Sinai West. Lam-Rachlin presented a real-world validation study of an AI-based screening tool at the 2026 Society for Maternal-Fetal Medicine (SMFM) Pregnancy Meeting. The study moved beyond controlled "reader studies" to evaluate how an FDA-cleared AI assistant performs in a high-volume, specialized diagnostic center.

Implementing "AI Assistance" in the clinical workflow

The study utilized a cloud-based software developed by BrightHeart, accessible via a cart-side tablet. Lam-Rachlin emphasized that the technology is designed to assist rather than replace the clinician.

"There is a fear of replacement, but I would always use the word 'AI assistance' because it really is just assisting us," she explained.

The workflow involves sonographers sending anonymized clips to the cloud while scanning. The AI requires at least 2 seconds of the four-chamber (4CH) view, as well as the left (LVOT) and right (RVOT) ventricular outflow tracts. Lam-Rachlin noted that the analysis is typically completed before the sonographer finishes the anatomy set, allowing them to identify if there is a "suspicious finding" that warrants a second look before the patient leaves the exam room.

High sensitivity and the learning curve of implementation

Over a 7-month period, the team analyzed 1,253 anatomy ultrasounds. The AI successfully detected 100% of the severe CHD cases (n = 4), which included transposition of the great arteries and pulmonary stenosis, as well as 2 significant extracardiac abnormalities affecting the heart.

Among the 1,236 cases without cardiac abnormalities, the AI achieved a specificity of 96.7% (95% CI: 95.5%- 97.6%), resulting in a manageable findings alert rate of 3.3%.

Lam-Rachlin highlighted a learning curve regarding view completeness. While the initial completion rate was 77%, it rose to over 90% by the fourth month of use. "If you take a snapshot of the period where completion rates were above 90%, the false alarm rate was only 1.5%, which is phenomenal," she observed.

Improving efficiency without adding time

For specialized centers that are already "scan-heavy," the primary benefit of AI may be efficiency. Lam-Rachlin reported that the software did not add time to the examinations and, in some instances, appeared to speed up the process by providing a rapid preliminary analysis.

“More than 1% of the time, we are probably second-guessing ourselves on the heart anyway, so this is pretty good,” Lam-Rachlin concluded.

The study data “support the clinical usability and generalizability of AI in prenatal cardiac screening and highlight its potential role in improving early CHD detection,” Lam-Rachlin and colleagues concluded.

Reference:

Lam-Rachlin J, Fox NS, Roth C, et al. Validation of an AI Software for Major CHD Screening in a Prenatal Diagnostic Center. Abstract. Presented at: SMFM 2026. February 8-13, 2026. Las Vegas, Nevada.