Segmentation of the fetal envelope on ante-natal MRI

Recent improvements in MRI scanners have enabled the acquisition of spatially consistent 3D images of the fetus. While recent works focused on the fetal brain segmentation, important outcome could be obtained from the segmentation of the whole fetal body envelope, such as accurate fetal weight estimation. In this paper, we propose to segment this envelope using a semi-automatic approach. MRI images were acquired using the Steady State Free Precession sequence. Taking advantage of the T2-weighting of this sequence, a set of fetal structures is identified and a simplified model of the fetal skeleton is instantiated in the images. An articulated model of a generic fetus is then registered in the images to initialize the fetal envelope segmentation and optimized using graph-cuts to refine the segmentation. Promising and robust results were obtained on a set of nine volumes.