MGB-NET: Orbital Bone Segmentation from Head and Neck CT Images Using Multi-Graylevel-Bone Convolutional Networks

Abstract

For the reconstruction of the orbital wall of the cranio-maxillofacial surgery, the segmentation of the orbital bone is necessary to support the eye globe position and restore the volume and shape of the orbit. However, due to the wide range of intensities of the orbital bones, conventional U-Net-based segmentation shows under-segmentation in the low-intensity thin bones of the orbital medial wall and orbital floor. In this paper, we propose a multi-gray-bone-Net (MGB-Net) for orbital bone segmentation that improves segmentation accuracy of high-intensity cortical bone as well as low-intensity thin bone in head-and-neck CT images. To prevent under-segmentation of the thin bones of the orbital medial wall and orbital floor, a single orbital bone mask is convert into two masks for cortical bone and thin bone. Two SGB-Nets separately are trained on these masks and each cortical and thin bone segmentation result is integrated to obtain the whole orbital bone segmentation result. Experiments show that our MGB-Net achieves improved performance for whole orbital bone segmentation as well as segmentation of thin bone of orbital medial wall and orbital floor.