Cone Beam CT Series Images Rigid Registration for Temporomandibular Joint via Self-supervised Learning Network

Abstract

Registration of the temporomandibular joint (TMJ) cone beam CT (CBCT) images plays an important role in the medical treatment of temporomandibular joint disorders (TMD) and related diseases. To highlight changes in the condyle bone of TMJ, accurate CBCT images registration is still a challenging work. In this paper, we proposed a self-supervised learning network to realize rigid registration for the TMJ CBCT series images. Without adopting the method of optimization iteration and similarity measurement, the transformation parameters of the rigid registration are directly regressed through our network. Then the warped image is obtained through spatial transformer network. The experimental results also proved the feasibility of this method, and it can greatly improve the accuracy and processing speed of rigid registration.