Atrioventricular Septal Defects Detection

Interaction between clinicians and AI to detect of fetal atrioventricular septal defects on ultrasound. The study demonstrated that when the AI model was incorrect, clinicians’ performance declined. Moreover, when clinicians were given information on the workings of the AI model, e.g., model confidence or grad-CAM, clinicians were more likely to trust the AI model inappropriately, i.e., when the model was wrong. The AI model still improved accuracy when evaluated the image correctly, and researchers concluded that an AI model may become reasonably accurate enough to assist clinicians in screening atrioventricular septal defect, given future research for a solution to mitigate mistakes in both the AI model and in trust calibration between clinicians and the AI model.