2D/3D Image Registration In Image-guided Surgery

Image-guided surgery(IGS) is one of the most commonly used and minimally invasive surgery which can precisely locate the lesions of patients with a high success rate for treatment. Before the implementation of surgeries, we need obtain high-resolution three dimension(3D) images which are used to assist clinicians to estimate the severity of patients’ lesions and plan the treatments; meanwhile we also need two dimension(2D) images which are used to real-timely correct surgical instruments relative to patients’ lesions so as to accurately cut patients’ lesions and inject the medicines during the surgeries. Therefore, The establishment of the spatial relationship between pre-operative 3D images and intra-operative 2D images, i.e., 2D/3D image registration, is crucial for the successful implementation of image-guided surgery. In addition, image registration is necessary not only for precise surgical navigation, treatment estimation and radiotherapy planning, but also for the preprocess of image fusion. In this dissertation, we first analyze the influence of different image information used for 2D/3D registration on the performances of image registration in terms of accuracy and robustness, registration time as well as registration success rate; and then mainly focus on proposing a novel image information for image registration(i.e., intensity distance information), a novel kind of mutual information integrating image intensity with intensity distance information, and morphological dilation driven 3D region growing algorithm for lung airway tree segmentation in feature-based image registration. The specific contributions of this dissertation are as follows.A novel image information, i.e., intensity distance(ID) has been proposed to improve the performances of intensity-based image registration methods which are implemented traditionally using the single intensity information, and then ID is used to construct a new similarity measure, called intensity distance difference(IDD) for 2D/3D image rigid registration. This distance information is referred as to the sum of Euclidean distances from the coordinates of pixels with the same intensity, which means intensity distance is totally different from image intensity, but there is a one-to-one relationship between them. The use of intensity distance in image registration showed the better performances of registrations than those by using the single image intensity information in terms of accuracy and robustness as well as registration time. Because intensity distance involved three different kinds of image information, i.e., image intensity, the number of pixels corresponding to image intensity and the coordinates of these pixels.A kind of novel mutual information has been proposed by introducing intensity distance information in different ways(i.e., addition and multiplication) to the most commonly used and traditional mutual information(respectively called distance coefficient MI and distance weighted MI) so as to improve the accuracy of registration and reduce registration time. These novel measures are constructed by using not only image intensities but also intensity distances, which means they can overcome the disadvantages of traditional mutual information from single image intensity information, and thus are more accurate and more robust than traditional measures.Morphological dilation driven 3D region growing algorithm for the extraction of lung airway tree has been proposed by incorporating morphology dilation with 3D region growing algorithm for feature-based image registration in order to reduce segmentation time and improve registration performances. Because the extraction of image features is complex and usually needs manual intervention, which means feature-based image registrations are time-consuming and need to be improved in terms of accuracy and robustness. The algorithm for lung airway tree first uses a variant of region growing to segment the main bronchioles without segmentation leak, and then drives region growing by morphology dilation, where centerlines of airway tree are extracted so as to refine the main bronchioles. Using these segmented airway trees as image features in feature-based registrations, their registration results are better than those of original region growing.