Study Of Non-rigid Registration Technologies For Medical Images

Image registration is a process of estimating an optimal transformation or mapping to obtain spatial correspondence among different images. It is one of the most fundamental research areas for various medical applications, such as radiotherapy plan, image-guided surgery, motion estimation, image segmentation and tracking changes over treatment time. Due to the non-rigid characteristics of human tissues, nonrigid registration of medical images has been extensively studied and various nonrigid image registration methods have been proposed. However, it remains an open problem in complicated deformation and large variation.In order to improve the registration accuracy based on the elastic body model, an adaptive strategy is proposed to construct the FEM mesh pyramid. In this strategy, the region intensity variance and region deformation are used as two indicators for determining whether the mesh in the sub-region should be refined. With these indicators, the mesh structure is adaptively adjusted by means of region triangulation refinement. Meanwhile, an adaptive mesh pyramid is constructed in the multi-resolution strategy. We compared our approach to other three models: 1) the conventional multi-resolution FEM registration algorithm; 2) the FEM elastic method that uses variation information for mesh refinement; and 3) the robust block matching(RBM) based registration. Comparisons among different methods in a dataset with 20 CT image pairs upon artificial deformation demonstrate that our registration method achieved significant improvement in accuracies. Experimental results in another dataset of 40 real medical image pairs for both mono-modal and multi-modal registrations also show that our model outperforms other three models in accuracy.In order to tackle the problem of local under-or-over registration for methods implemented by optimizing the global similarity, a novel strategy is proposed to incorporating the landmark information in the energy function to constrain the local deformation in the vicinity of the landmark. We obtain the landmark information by means of automatically extracting the corresponding scale-invariant features from the reference image and float image. Meanwhile the gradient field of the image similarity metric is derived for the image external force. The qualitative and quantitative evaluations of our proposed method on both MR image pairs and mammographic images show that the aforementioned strategy improves the alignment between the reference and float image.Due to being derived from linear assumption, most elastic body based non-rigid image registration algorithms are facing challenges for soft tissues with complex nonlinear behavior and with large deformations. To take into account the geometric nonlinearity of soft tissues, we propose a registration algorithm on the basis of Newtonian differential equation. The material behavior of soft tissues is modeled as St.Venant-Kirchhoff elasticity, and the nonlinearity of the continuum represents the quadratic term of the deformation gradient under the Green- St.Venant strain. In our algorithm, the elastic force is formulated as the derivative of the deformation energy with respect to the nodal displacement vectors of the finite element; the external force is determined by the registration similarity gradient flow which drives the floating image deforming to the equilibrium condition. Experimental results show that our model outperforms the conventional model in estimating large deformations.Nonrigid registration of multimodal medical images remains a challenge in image guided interventions. A common approach is to use mutual information(MI), which is robust to those intensity variations across modalities. However, primarily based on intensity distribution, MI does not take into account of underlying spatial and structural information of the images, which might lead to local optimization. To address such a challenge, this paper proposes a two-stage multimodal nonrigid registration scheme with joint structure information and local entropy. In our two-stage multimodal nonrigid registration scheme, both the reference image and floating image are first converted to a common space. An unified representation in the common space for multimodal images is construct by fusing the structure tensor trace with the local entropy. Through the representation that reflects its geometry uniformly across modalities, the complicated deformation field is estimated base on the local affine model based registration using the intensity distance in the common space. The two-stage multimodal registration with joint structural tensor and local entropy outperforms the conventional MI based method. Both the structural tensor and the local entropy play the positive role.Finally, to preserve the image local details and edges, an adaptive nonlinear regularization term is incorporated in the energy function, which takes into account the regional statistical intensity information and the magnitude characteristic of the Chebyshev low pass filter. Experimental results on a large number of real medical image pairs demonstrate the effectiveness of our proposed method for non-rigid medical image registration.