Registration of images containing rigid structures using deformable models and its medical application

Abstract

This thesis presents research and development in registration of images containing rigid structures using deformable models. In the existing methods of deformable image registration, the feature based methods using purely interpolation are often used under the assumption that all parts of the image can be deformed non-linearly. However, it is not applicable for some images, such as medical images which contain some bony structures. Therefore, it is more reasonable in image registration to treat the parts representing rigid structures differently from the remaining parts.
A novel method is proposed in this thesis that uses a flexible spring-mass system to provide a good representation of image information. Anatomical structures are represented by spring masses and their interconnecting springs located along the structure boundaries. Hence, the system can treat some pre-selected parts of an image as rigid structures, whilst it is possible to elastically deform the rest of the image. The deformation caused by the displacement of the features is modelled by the spring-mass system via the motion equation. To ensure high quality registration, the image content is introduced as external forces acting on the masses and is iteratively employed into the motion equation to achieve a better and better registration result (It should be noted that the improvement in registration quality is limited by the quality and content of original images). The performance of the proposed method is shown using simulated data, and is validated using real Computed Tomography (CT) images. Objective measures and images that enable subjective evaluation of the new technique are provided.
The proposed method is fast, robust and offers high registration accuracy. This is due to the following developments:
• Innovatory approach based on the spring-mass system for image registration.
• A new spring force formula to support the new method.
• External forces based on image content are employed and applied via an iterative approach to ensure the high quality registration.
• A fast initialisation method for the spring-mass system generation.
• A fast interpolation technique to reduce the computation time in the iterative registration stage.