Registration Of Three-dimensional Ultrasound And Magnetic Resonance Images

Registration of ultrasound and magnetic resonance images is one of the key techniques in surgical navigation. In recent years, with the improvement of ultrasound imaging quality, the navigation system with ultrasound guidance has been widely applied. However, since the signal-to-noise ratio of ultrasound image is lower than that of magnetic resonance image, the imaging quality of ultrasound image is still not satisfactied. Registration of intraoperative ultrasound images with preoperative magnetic resonance images can locate the trajectory of intraoperative instruments more accurately and implement interventional operation better.The initial registration of ultrasound and MRI images with the aid of optical tracking system is investigated in this paper, and a novel precise registration method based on pseudo ultrasound and mutual information is then proposed by combining the PV(Partial Volume) interpolation technique and Powell optimization algorithm. The main work is shown as follows:(1) This paper studied the interpolation method, the transformation space, search algorithm and similarity measure technologies related to image registration.(2) The initial registration of ultrasound and MRI images has been realized with the aid of optical tracking system. The registration process mainly includes: calibrating ultrasound probe using N linear templates, designing and making phantom, reconstructing 3D ultrasound image from 2D ultrasound images and calculating the transformation matrix between ultrasound and MRI images.(3) The precise registration method of ultrasound and MRI images is put forward. Firstly, MRI images are segmented based on gaussian mixture model; then, the pseudo ultrasound images are achieved by mapping gray levels of ultrasound images to segmented MRI images. The precise registration of two model images is accomplished through the mutual information of the pseudo ultrasound images and ultrasound images. In the process of registration, PV interpolation and improved Powell optimization algorithm have been also used in searching maximum mutual information to reduce the possibility of falling into the local extremum.The experiments of phantom and clinical brain data show the effectiveness of the registration method proposed in this paper.