U-net Based Autonomous Fetal Segmentation From 2D and 3D Ultrasound Images

Abstract

There are various biometric parameters of the fetus that need to be evaluated to monitor prenatal diagnosis during pregnancy. Biometric parameters such as head circumference, abdominal circumference, cortical volume, the volume of the brain, crown-rump length, femur length, etc. play a very important part in the character ization and detection of the development of the fetus. Gestational age is one of the most effective parameters for monitoring fetal growth and development, as well as diagnosing any abnormalities, among several quantitative indices. To estimate the gestational age, birth size, weight, and to monitor prenatal abnormalities, many bio metric parameters such as head circumference (HC), abdomen circumference (AC), and femur length (FL) must be measured. We can extract these parameters from the segmentation of an MRI scan. However, performing full manual segmentation is exhaustive and time-consuming. Ultrasound imaging has been shown to be more efficient than MRI for measuring such biometric characteristics.. Also, in this case, manual segmentation requires experts’ experience and skills, clinical experience of the staff which is time-consuming. As a result, we propose a fully autonomous segmentation method based on U-Net architecture for fetal biometric parameters such as head circumference (HC), abdomen circumference (AC), and femur length (FL), which eliminates the need for manual intervention, reduces computational complexity, and greatly speeds up the segmentation process. U-Net is a convolu tional neural network that was created for performing segmentation on biomedical images. Our goal is to train the network such that it can create high-resolution 2D and 3D ultrasound images of each segmented fetal area.