Poster Session 2
Category: Ultrasound/Imaging
Poster Session 2
Noah AB Saphier
High School Student
Dwight-Englewood School
Englewood, New Jersey, United States
Carl J. Saphier, MD (he/him/his)
Medical Director
Women's Ultrasound, LLC
Englewood, New Jersey, United States
We retrospectively analyzed 3925 ultrasound exams of singleton pregnancies at 18-24 weeks performed and interpreted solely by a single MFM physician-sonographer at a single center in the United States according to AIUM practice parameters from 2014 to 2025. An AI software developed by BrightHeart was applied to first assess view completeness (≥ 2 seconds of 4CH, LVOT and RVOT views in grayscale ultrasound cines). For the 1858 scans meeting view completeness criteria, the AI software analyzed all cines and assessed the presence of 8 morphological findings suspicious for CHD. The primary outcome was the alarm rate (presence of an AI-identified flag) in fetuses determined to be morphologically normal during the ultrasound exam and subsequent clinical follow-up.
Results:
Of the 1858 ultrasound exams, 1753 had no cardiac abnormality, 100 had minor suspected abnormality and 5 had severe CHD identified. The AI alarm rate in exams without cardiac abnormality was 5.5% (95% CI: 4.6-6.7) and remained low across subgroups including BMI, fetal presentation, and 3 generations of ultrasound devices (see Figure). The AI software identified at least 1 flag in all 5 cases with severe CHD.
Conclusion:
Retrospective application of an AI tool demonstrated accurate detection of severe CHD, while maintaining a low alarm rate among exams without cardiac abnormality across a diversity of BMI, range of fetal presentations, and several generations of ultrasound devices. The findings suggest the robustness of AI to provide a low alarm rate while also providing a high detection of severe CHD, thereby avoiding an overwhelming burden for a second tier evaluation.