Efficient dense non-rigid registration using the free-form deformation framework

• Main Author: Modat, M.
• Published: University College London (University of London) 2012
• Subjects: 610
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.565833

Medical image registration consists of finding spatial correspondences between two images or more. It is a powerful tool which is commonly used in various medical image processing tasks. Even though medical image registration has been an active topic of research for the last two decades, significant challenges in the field remain to be solved. This thesis addresses some of these challenges through extensions to the Free-Form Deformation (FFD) registration framework, which is one of the most widely used and well-established non-rigid registration algorithm. Medical image registration is a computationally expensive task because of the high degrees of freedom of the non-rigid transformations. In this work, the FFD algorithm has been re-factored to enable fast processing, while maintaining the accuracy of the results. In addition, parallel computing paradigms have been employed to provide near real-time image registration capabilities. Further modifications have been performed to improve the registration robustness to artifacts such as tissues non-uniformity. The plausibility of the generated deformation field has been improved through the use of bio-mechanical models based regularization. Additionally, diffeomorphic extensions to the algorithm were also developed. The work presented in this thesis has been extensively validated using brain magnetic resonance imaging of patients diagnosed with dementia or patients undergoing brain resection. It has also been applied to lung X-ray computed tomography and imaging of small animals. Alongside with this thesis an open-source package, NiftyReg, has been developed to release the presented work to the medical imaging community.