Wufan Chen

Multiresolution Elastic Registration of X-Ray Angiography Images Using Thin-Plate Spline

X-ray angiography, a powerful technique for the visualization of blood vessels, has been widely used in clinical practice. However, due to unavoidable motion of patient, the subtraction images often suffer from misregistration artifacts. In order to improve the quality of subtraction images, registration algorithms are often employed before direct subtraction of mask and live images. A novel multiresolution elastic registration algorithm is proposed for the registration of the digital angiographic images using thin-plate spline (TPS). Our main contribution is a multiresolution search strategy specifically designed for the template matching method. In this strategy, the mask image is decomposed to coarse and fine sub-image blocks iteratively using the pyramid approach. Experimental results show that the multiresolution refinement strategy is well adapted to the template matching method, and can achieve better performance than comparable single step algorithms, because local minima can be overcome by the gradual coarse-to-fine approach that also ensures convergence. Registration results of four typical similarity measures, namely energy of histogram of differences (EHD), mutual information (MI), correlation and sum of squared differences (SSD), are compared. Three different interpolation methods, including nearest-neighbor, bilinear and bicubic, are also tested and compared. The overall conclusion is that the multiresolution refinement algorithm based on EHD combined with the bicubic interpolation method is very robust and effective for the registration of X-ray angiography images, which can obtain sub-pixel registration accuracy and is fully automatic. In addition, the objective measurement method developed in this paper on simulated data makes it possible to quantitatively evaluate the quality of the elastic registration results

Automatic Segmentation of Coronary Angiograms Based on Probabilistic Tracking

This paper presents a novel tracking method for automatic segmentation of coronary artery tree in the X-ray angiographic images, based on probabilitistic vessel tracking and structure pattern inferring. The method is composed of two main steps, namely preprocessing, and tracking. In the preprocessing step, multiscale Gabor filtering and Hessian matrix analysis are used to enhance and extract vessels from the original angiographic image, leading to a vessel feature map as well as a vessel direction map. In the tracking step, a probabilistic tracking operator is proposed to extract vessel segments or branches, together with a detector to identify vessel structure. The identified structure pattern is used to control the tracking process. By appropriate integration of these advanced preprocessing and tracking steps, the algorithm is able to extract both vessel axis-lines and edge points, and to measure the arterial diameters in various complicated cases. The experimental results were satisfying.

A new method for robust contour tracking in cardiac image sequences

For the segmentation and robust tracking of the cardiac image sequences (CIS) of magnetic resonance (MR), an optimized algorithm is presented in this paper, which is based on the active contour framework. To use the active contours model (ACM) estimating the cardiac motion, a new concept of generalized fuzzy gradient vector flow (GFGVF) is presented and compared with the classical gradient vector flow (GVF). Then a modified ACM is proposed for motion tracking, which is based on two new external forces: one is the GFGVF field; the other is the relativity of optical flow field (OFF) on predictive contour. For robust tracking the outline of interest, a set of motion equations is presented to describe two correlative updating steps. Another, given some prior terms and likelihood one, the motion state of each point can be found by the maximum a posteriori probability (MAP).

Hessian based image structure adaptive gradient vector flow for parametric active contours

Active contours have been one of the most successful methods for image segmentation during the last two decades, but one of the shortcomings of being unable to converge to concavity is a handicap to its effectiveness. In order to address this issue, the gradient vector flow (GVF) was put forth. Although there have been a great number of works on GVF, the image structure has seldom been incorporated into GVF algorithm. In this work, the image structure characterized by the Hessian matrix is incorporated into the GVF algorithm by reformulating the smoothness constraint of GVF into matrix form. In this way, the associated diffusion PDEs are anisotropic and the modified GVF snake can converge to very long concavity and preserve weak edge simultaneously. Experiments and comparisons are presented to demonstrate the properties of the proposed strategies.

Segmentation of Brain Magnetic Resonance Angiography Images Based on MAP-MRF with Multi-pattern Neighborhood System

Existing maximum a posteriori probability and Markov random field (MRF) models have limitations associated with that the ordinary neighborhood system being unable to differentiate subtle changes due to several-to-one correspondence within the neighborhood. Aiming at overcoming the limitations and applications to segmentation of cerebral vessels from magnetic resonance angiography images, we proposed a multi-pattern neighborhood system and corresponding energy equation to enable the MRF model for segmenting fine cerebral vessels with complicated context. In the implementation, a candidate space of cerebral vessels was employed to reduce the time-consumption, which was based on a threshold of the response to multi-scale filtering. A set of phantoms simulating segmentation challenges of vessels have been devised to quantitatively validate the algorithm. In addition, ten three-dimensional clinical datasets have been used to validate the algorithm qualitatively. It has been shown that the proposed method could yield smaller error and improve the spatial resolution of MRF model.

Robust medical image registration based on phase congruency and regional mutual information

Abstract In this paper, a new approach of multi-modality image registration is represented with not only image intensity, but also features describing image structure. There are two novelties in the proposed method. First, instead of standard mutual information based on joint intensity histogram, a graph-based implementation of multi-dimensional regional mutual information is employed, which allows neighbourhood information to be taken into account. Second, a new feature image is obtained by means of phase congruency, which is invariant to brightness or contrast changes. By incorporating these features and intensity into regional mutual information, we can combine aspects of both structural and neighbourhood information together, which offers a more robust and a high level of registration accuracy that is essential in application to the medical domain.

Hybrid rigid and non-rigid registration algorithm for alignment of intra-subject thoracic and abdominal images

Abstract A hybrid rigid and non-rigid registration algorithm has been presented to register thoracic and abdominal CT images of the same subject scanned at different times. The bony structures are first segmented from two different time CT images, respectively. Then, the segmented bony structures in the two respective images are registered based on their boundary points using a soft correspondence matching algorithm, with a rigid transformation constraint on each bony structure. With estimated correspondences in bony structures, the dense deformations in the entire images are interpolated by a thin plate spline (TPS) interpolation technique. To improve the alignment of soft tissues in the images as well, a normalised mutual information based B-spline registration algorithm is used to iteratively refine the registration of soft tissues, and at the same time keep the rigid transformation for each bony structure. This registration refinement procedure is repeated until the algorithm converges. The proposed hybrid registration algorithm has been applied to the clinical data with very encouraging results as measured by two clinical radiologists.

Medical image elastic registration based on discontinuity adaptive Markov random field model

Abstract We present a novel elastic registration algorithm using discontinuity adaptive Markov random field (DA-MRF) model. We use B-spline to model the deformation field, and then the coefficients of B-spline are the parameters to be evaluated. We model the coefficients fields as MRFs. The optimal deformation is sought as maximum a posteriori (MAP) configurations through a minimisation problem that includes two terms: the pixel-wise mean-square distance measure between the reference and the warped test image (data term), and continuity constraints imposed on pairs of neighbouring coefficients that promotes a smooth deformation (regularisation term). We use two- and three-dimensional medical images to test the proposed algorithm. Experimental results show that the proposed method is robust and accurate, suitable for clinical application.

Contour tracking of left ventricle based on generalized fuzzy particle filter

Summary form only given. For medical image sequences, the method of contour-based tracking proved to be a powerful tool for boundary delineation. During contour evolution, the particle filter (PF) can be used to track the feature points by enforcing spatio-temporal local constraints to handle the observation noise. To optimize the importance ratios (IR) of PF and improve its capability, a new approach of generalized fuzzy particle filter (GFPF) is presented. Compared with the unscented particle filter (UPF) that is currently a good method for object tracking, GFPF shows more advantages, including lower particle degeneracy, higher precision and so on. In addition, a likelihood estimation model is constructed to provide the observation data for GFPF. By theoretic analysis and contrast experiments, it is clear that GFPF is a good method for left ventricle tracking.