Deformable Image Registration For Neurosurgical Procedures

Abstract

Image-guided neurosurgery based on navigation systems has been developed to assist intracranial tumor resection surgery. The success of the surgery relies heavily on the precision of the navigation system, which decreases due to a phenomenon called "brain shift". During craniotomies, the soft tissues of the brain deform due to gravitational force, cerebro spinal fluid leakage, intracranial pressure change or surgical interventions. This significantly deteriorates the assumption of linear geometrical differences between pre-operative and intra-operative images. To compensate for such non-linear deformations, non-rigid registration techniques are employed. The first contribution of this PhD work is the development of a new approach for inter-subject non-rigid registration of 3D magnetic resonance (MR) brain images. It is motivated by the ideas derived from elastodynamics which is the subclass of linear elastic theory. We proposed to model the non-rigid deformations as elastic waves which are characterized by elastodynamics wave equation. The registration process ensues in a hierarchical fashion, thus reducing the risk of obtaining a local optimal transformation. Experimental results demonstrated that the proposed deformable registration method leads to very promising results when applied to the problem of inter-subject registration and that favorably compared against classical registration approaches. The second contribution of this work is the incorporation of topology preservation property into our proposed inter-subject non-rigid registration method. We proposed to impose the topology preserving penalty on the deformation by constraining the Jacobian determinant of the transformation to be positive over the entire image domain. This property ensured that the recovered transformations do not exhibit tearing or foldiiing effects. The results of the proposed registration approach were compared in terms of Kappa index and relative overlap over segmented anatomical structures to that obtained with existing topology preserving non-rigid image registration methods and non topology preserving variant of our proposed registration scheme. The Jacobian determinant maps obtained with our proposed registration method were qualitatively and quantitatively analyzed. The results demonstrated that the proposed scheme provides good registration accuracy and results into smooth transformation with a guarantee to preserve topology. The third contribution of this PhD work is that we developed a new inverse consistent non-rigid image registration method based on elastodynamics. Inverse consistency property renders the registration procedure unbiased towards the order of input images. This assures that the forward and reverse transformations are inverses of each other which do not change by switching the input images. We introduced the inverse consistency constraint into the inertial force that is part of the elastodynamics wave equation which governs the underlying non-rigid deformations. We conducted image registration experiments, with and without inverse consistency constraint, on three different datasets comprising of 3D MR brain scans. The extent to which the proposed registration scheme enforced inverse consistency was analyzed through inverse consistency error. The results revealed that the inverse consistency error reduced by 99% with our inverse consistent registration method as compared to the non-inverse consistent counterpart. Thus, the proposed inverse consistent registration method seems very promising both in terms of registration accuracy and inverse consistency error