Automatic segmentation in fetal ultrasound images based on improved U-net

Abstract

As an effective way of routine prenatal diagnosis, ultrasound (US) imaging has been widely used in clinical practice. Biosignatures obtained from fetal segmentation contribute to fetal development and health monitoring. However, the artifacts, speckle noises, quality of imaging equipment and other factors make the segmentation of fetal US images extremely challenging. In this paper, aiming to improve the depth of the model, as well as to avoid the vanishing gradient problem and exploding gradient problem, we propose Residual U-net and ASPP U-net based on U-net, which further improves the accuracy of segmentation without increasing the depth of the model. The results of our experiments show that the network proposed in the paper can effectively improve the segmentation accuracy in fetal US images.