EP01.26: Diagnostic performance of antenatal diagnosis of congenital heart disease with artificial intelligence versus trained specialists

To determine the diagnostic performance of antenatal diagnosis of congenital heart disease with artificial intelligence versus trained specialists. Cross-sectional study with obstetric ultrasound images from centres involved in the program for the diagnosis, training, logistics, education, organisation, and follow-up of patients with congenital heart disease (DELFOS). A group of trained specialists was exposed to a set of 50 training images simulating the training process of the Artificial Intelligence (AI) model. Subsequently, the participants had to determine in 56 images corresponding to 30 patients whether there was or not a presence of congenital heart disease. The final results were grouped at the patient level for both the specialist study and the AI model. Study participants did not have access to the patient's medical history or additional data, so they had to perform the analysis with the same information as the AI model. 48 trained specialists from the basic screening and anomaly detection course of DELFOS were included. Sensitivity, specificity, false negative rate, and false positive rate were calculated for the group of trained specialists and AI. Additionally, comparison was made with the developed AI method and with random performance. It can be observed that trained specialists have higher sensitivity levels than the AI model but have lower specificity. This is due to the false positive rate calculated for trained specialists being significantly higher than that indicated by the AI model. In our experience, the diagnostic performance of AI (DELFOS) demonstrated higher specificity and a lower false positive rate compared to trained specialists. However, specialists had higher sensitivity. This balance between sensitivity and specificity is crucial for the detection of congenital heart disease, so the incorporation of an AI model complements and enhances detection capability.