(1041) Prenatal aortic coarctation sign screening using automatic AI-assisted vessel diameter measurement

Objective: Aortic coarctation (AoC) is a congenital heart defect characterized by hypoplasia of the aortic arch visible on the 3-vessel view (3VV). This study aims to (i) develop a pipeline using artificial intelligence (AI) for automatic vessel diameter measurement, (ii) assess whether vessel diameter is a reliable predictor of AoC prenatally.

Study Design: An AI software (under development, not yet FDA approved) trained on thousands of annotated images from US and European sites is used for the automatic detection and segmentation of the pulmonary artery (PA) and aorta (Ao) on 2D 3VV images. Post-processing algorithms are applied to achieve automatic and standardized diameter measurements. The model's performance is assessed on a test set of images on which the models were not trained, consisting of 2548 non-pathological images and 32 images from studies where AoC was diagnosed prenatally by a multidisciplinary team (Hôpital Necker, Paris).

Results: The models demonstrated high accuracy in vessel diameter measurement. The sign used for AoC screening is the diameter ratio of the PA to that of the Ao. With a ratio greater than 1.6, an AoC should be suspected. This ratio is calculated over the entire test set for the two populations, and a shift is observed (Figure 1). For healthy data, the mean ratio is 1.2±0.3 mm, while for the pathological data, the mean ratio is 1.8±0.3 mm. Our method achieves a sensitivity of 71.9% and a specificity of 87.8% for AoC sign screening.

Conclusion: This study validates AI’s potential for automatic AoC sign screening. It offers a fast and reliable alternative to manual measurement methods. However, the AP/Ao ratio alone has limitations for screening CoA prenatally. Other indicators should be investigated. This study used cases detected prenatally, introducing a bias. Future work will look at cases that were diagnosed postnatally.