A novel framework to segment out cervical vertebrae

Abstract

Cervical (neck) pain is becoming the leading issue worldwide. Accurate segmentation of cervical spine can assist radiologists and doctors in early and batter clinical diagnosis of cervical abnormalities. This paper addresses the challenging issue of cervical image segmentation. Various techniques have been proposed earlier, but still work is needed for accurate identification and segmentation of vertebral bodies. The technique required to segment out vertebra should be robust and invariant to variation in shape, noise, rotation, scale and occlusions. In this paper, we have presented a novel technique for cervical vertebrae segmentation, which is based on deep learning, i.e., U-Net. This technique would have an ability to capture information embedded in x-ray data images for segmentation purpose and have an ability to overcome noise and occlusions in batter and efficient way. Keywords: Image segmentation, X-ray imaging, Biomedical imaging, Deep learning, Object segmentation, Bones, Convolution