Foundation of Prenatal Risk

Most pregnant women undertake at least two ultrasound exams as well as physical and biochemical examinations. Nevertheless, many adverse pregnancy outcomes are detected timely with only very low sensitivity. Preterm birth 40%, Low weight for gestational age 25%, congenital heart disease 40%. Maternal health history as well as ultrasound exams are obvious objects for deep learning-based early detection of adverse outcomes. We develop foundation models for mother's health history and for ultrasound exams and shows fine-tunings with substantial increase in sensitivity of several adverse outcomes within the time-window of intervention. Data are based on national Danish data from more than 700.000 pregnancies.