Search

Point-based registration with known correspondence is often used as either a stand-alone method or a part of a more complex algorithm. The goal of this type of registration is to align two sets of points with the same number of corresponding points using a selected transformation type. Presented are closed form solutions for the transformation parameters that optimally align two point sets in the least squares sense for the following transformation types: rigid, similarity, rigid with nonuniform scales, and a linear combination of basis functions. It is shown that those registration methods whose underlying transformations form a group satisfy the identity, symmetry, transitivity, and distortion properties. [ABSTRACT FROM AUTHOR]

The aims of this paper are to devise robust methods for the description of the variability in shapes of long bones using 3D statistical shape models (SSMs), and to test these on a dataset of humeri that demonstrate significant variability in shape. 30 primate humeri were CT scanned and manually segmented. SSMs were constructed from a training set of landmarks. The landmarks of the 3D shapes are extracted automatically using marching cubes and point correspondences are automatically obtained via a volumetric non-rigid registration technique using multiresolution B-Spline deformations. The surface registration resulted in no discernible differences between bone shapes, demonstrating the high accuracy of the registration method. An analysis of variations is applied on the shapes based on the model we built. The first mode of variation accounted for 42% of the variation in bone shape. This single component discriminated directly between great apes (including humans) and monkeys. [ABSTRACT FROM AUTHOR]

External beam radiotherapy treats cancer lesions with ionizing radiation. Successful treatment requires a correct definition of the target volume. This is achieved using pre-treatment MR and CT images. However, due to changes in patient position, tumor size and organ location, adaptation of the treatment plan over the different treatment sessions might be wanted. This can be achieved with extra MR and CT images obtained during treatment. Bringing all images into a common reference frame, the initial segmentations can be propagated over time and the integrated dose can be correctly calculated. In this article, we show in two patients with rectum cancer and one with neck cancer that a significant change in tumor position and shape occurs. Our results show that nonrigid registration can correctly detect these shape and position changes in MR images. Validation was performed using manual delineations. For delineations of the mandible, parotid and submandibular gland in the head-and-neck patient, the maximal centroid error decreases from 6 mm to 2 mm, while the minimal Dice similarity criterium (DSC) overlap measure increases from 0.70 to 0.84. In the rectal cancer patients, the maximal centroid error drops from 15 mm to 5 mm, while the minimal DSC rises from 0.22 to 0.57. Similar experiments were performed on CT images. The validation here was infeasible due to significant inaccuracies in the manual delineations. [ABSTRACT FROM AUTHOR]

MR – CT image registration has been used in the liver cancer treatment with an open MR Scanner to guide percutaneous puncture for ablation of tumors. Due to low magnetic field and limited acquisition time, MR images do not always show the target clearly. Sometimes, assistance of CT images is helpful for the navigation to the target. The shape of the liver within the surgical procedure is different from that of preoperative CT images due to the patient position for the convenience of surgery. It is quite difficult to match the images accuracy during surgery. In this paper, we have proposed a method to improve the registration accuracy of images with an open MR scanner and preoperative CT images of the liver. The method includes three parts. Firstly a semiautomatic method is used to extract the liver from MR and CT images as region of interest (ROI). Then, an affine registration is used to match the images roughly. Finally, BSpline-based nonrigid registration is applied. The results are found to be satisfactory with visual inspection by experts and with evaluation by the distance of two liver surfaces, while comparing with other methods. [ABSTRACT FROM AUTHOR]

This paper exploits the different properties between the local neighborhood of global optimum and those of local optima in image registration optimization. Namely, a global optimum has a larger capture neighborhood, in which from any location a monotonic path exists to reach this optimum, than any other local optima. With these properties, we propose a simple and computationally efficient technique using transformation disturbance to assist an optimization algorithm to avoid local optima, and hence to achieve a robust optimization. We demonstrate our method on 3D rigid registrations by using mutual information as similarity measure, and we adopt quaternions to represent rotations for the purpose of the unique and order-independent expression. Randomized registration experiments on four clinical CT and MR-T1 datasets show that the proposed method consistently gives much higher success rates than the conventional multi-resolution mutual information based method. The accuracy of our method is also high. [ABSTRACT FROM AUTHOR]

Recently, it is shown the morphology of the white matter, obtained by ultrasound (US) imaging, has prognostic implications for certain brain pathologies. Periventricular Leukomalacia (PVL) is such a pathology diagnosed on infants with a very low birth weight (<1500g). Next to the US images Magnetic Resonance Image (MRI) volumes are commonly used for the inspection of this pathology. Since on both modalities, up to now, we still lack a golden standard for the quantification of the syndrome cross-validation through a multi-modal registration is highly beneficial to the clinical diagnosis. In this article we present a 2D US - 3D MRI registration scheme combining an interactive initialization step, B-spline image interpolation, a mutual information based metric and an evolutionary algorithm optimization scheme. [ABSTRACT FROM AUTHOR]

Mutual information has been successfully used as an effective similarity measure for multimodal image registration. However, a drawback of the standard mutual information-based computation is that the joint histogram is only calculated from the correspondence between individual voxels in the two images. In this paper, the normalized mutual information measure is extended to consider the correspondence between voxel blocks in multimodal rigid registration. The ambiguity and high-dimensionality that appears when dealing with the voxel neighborhood is solved using uniformly distributed random lines and reducing the number of bins of the images. Experimental results show a significant improvement with respect to the standard normalized mutual information. [ABSTRACT FROM AUTHOR]

We present a novel approach for a combined homogenization and registration technique. Medical image data is often disturbed by inhomogeneities from staining, illumination or attenuation. State-of-the-art approaches tackle homogenization and registration separately. Our new method attacks both troublemakers simultaneously. It is modeled as a minimization problem of a functional consisting of a distance measure and regularizers for the displacement field and the grayscale correction term. The simultaneous homogenization and registration enables an automatic correction of gray values and improves the local contrast. The combined approach takes slightly more computing time for an optimization step as compared to the non-combined scheme and so is much faster than sequential methods. We tested the performance both on academic and real life data. It turned out, that the combined approach enhances image quality, especially the visibility of slightly differentiable structures. [ABSTRACT FROM AUTHOR]

In this paper, we propose a computationally efficient method for medical image registration. The centerpiece of the approach is to reduce the dimensions of each image via a projection operation. The two sequences of projection images corresponding to each image are used for estimating the registration parameters. Depending upon how the projection geometry is set up, the lower dimension registration problem can be parameterized and solved for a subset of parameters from the original problem. Computation of similarity metrics on the lower dimension projection images is significantly less complex than on the original volumetric images. Furthermore, depending on the type of projection operator used, one can achieve a better signal to noise ratio for the projection images than the original images. In order to further accelerate the process, we use Graphic Processing Units (GPUs) for generating projections of the volumetric data. We also perform the similarity computation on the graphics board, using a GPU with a programmable rendering pipeline. By doing that, we avoid transferring a large amount of data from graphics memory to system memory for computation. Furthermore, the performance of the more complex algorithms exploiting the graphics processor’s capabilities is greatly improved. We evaluate the performance and the speed of the proposed projection based registration approach using various similarity measures and benchmark them against an SSE-accelerated CPU based implementation. [ABSTRACT FROM AUTHOR]

Since the 1970’s B-splinesContrary to popular usage in the CAGD community, we follow de Boor’s distinction discussed in [1] of using the term ‘B-spline’ to denote the shape function of limited support and not the B-spline object (e.g curve, surface, volume). have evolved to become the de facto standard for use in curve and surface representation. This resulted in a relatively recent proliferation of nonrigid image registration techniques based on B-splines. These techniques fall under the general Free-Form Deformation (FFD) approach in which the object to be registered is embedded within a B-spline object. The deformation of the B-spline object represents the transformation space of the registered object. In this paper, we describe the implementation of our finite element methodological (FEM) approach using B-splines. This registration framework subsumes essential components of currently popular FFD image registration algorithms, while providing a more principled and generalized control mechanism for nonrigid deformation. Our implementation constitutes an extension of the existing FEM library of the Insight Toolkit (ITK). We discuss the theoretical implications and provide experimental results of our proposed methodology. [ABSTRACT FROM AUTHOR]

The Allen Brain Atlas project aims to bridge the divide between genomics and neuroanatomy by mapping the expression of the entire C57BL/6J mouse genome onto a high-resolution 3D anatomical reference atlas of the mouse brain. We present the image registration approach used to generate this anatomical reference from histological sections. Due to the large number of sections (525) and the presence of debris and distortions, a straightforward alignment of each slice to its neighbors fails to accurately recover the 3D shape of the brain. On the other hand, multimodality registration of histology slices to an MRI reference compromises correspondences between neighboring slices. Our approach combines the high-frequency component of slice-to-slice histology registration with the low-frequency component of the histology-to-MRI registration to produce a coarse-to-fine reconstruction that is accurate both in its global shape and in the alignment of local features. [ABSTRACT FROM AUTHOR]

Therapeutic outcomes from radiation treatment planning (RTP) may be improved by combining tumour metabolism data with anatomic location via PET/CT image fusion. The aim of this work is to reveal the spatial accuracy, reproducibility and speed of registration techniques used in Gemini PET/CT system as a part of quality assurance. A multi-layer alignment device and a Rando Man phantom were used to validate PET/CT intrinsic alignment. Known transformations were performed to simulate different misalignment between PET and CT phantom images. Point-based and image-based registration techniques used to correct misalignment were assessed and compared quantitatively by measuring absolute distances between the centroids of corresponding fiducial markers in the registered volumes. In comparison with phantom studies using image-based registration, point-based registration shows better spatial accuracy and faster correlation, but longer time and more effort in human intervention. The experimental findings confirm that Gemini can produce robust intrinsic image alignment, as well as the accurate PET/CT image registration with careful user interaction. [ABSTRACT FROM AUTHOR]

An elastic registration algorithm based on fuzzy set is proposed in the paper. The fuzziness of medical images is shown in two aspects: (1) the intensity of the pixels in medical images is fuzzy. The same kind of tissue may has different intensity and the same intensity may correspond to different tissues in one image; (2) the space position of image pixel is fuzzy. In the paper, we applied the fuzzy theory to the first aspect and presented the concept of fuzzy mutual information and its optimization method. For the second aspect, a multiresolution registration method based on fuzzy set and its optimization method is presented. 16 groups of experiments have been done and the results showed that the elastic registration algorithm based on fuzzy set can improve the accuracy and robustness of registration algorithm greatly. [ABSTRACT FROM AUTHOR]

Because of its robustness and accuracy for a variety of applications, either monomodal or multimodal, mutual information (MI) is a very popular similarity measure for (medical) image registration. Calculation of MI is based on the joint histogram of the two images to be registered, expressing the statistical relationship between image intensities at corresponding positions. However, the calculation of the joint histogram is not straightforward. The discrete nature of digital images, sampled as well in the intensity as in the spatial domain, impedes the exact calculation of the joint histogram. Moreover, during registration often an intensity will be sought at a non grid position of the floating image. This article compares the robustness and accuracy of two common histogram estimators in the context of nonrigid multiresolution medical image registration: a Parzen window intensity interpolator (IIP) and generalised partial volume histogram estimation (GPV). Starting from the BrainWeb data and realistic deformation fields obtained from patient images, the experiments show that GPV is more robust, while IIP is more accurate. Using a combined approach, an average registration error of 0.12 mm for intramodal and 0.30 mm for intermodal registration is achieved. [ABSTRACT FROM AUTHOR]

The purpose of this paper is to demonstrate the clinical advantages of using semiautomatic volume registration where automatic registration is problematic due to large deformations, small bone anatomy, or extraneous structures. Examples are drawn from clinical cases of MRI/PET breast studies, CT angiography/SPECT cardiac studies, and total wrist arthroplasty. These types of studies should be contrasted with those involving the head, thorax, and pelvis where there is much less deformation and the existence of (some) large bones facilitates automatic matching. [ABSTRACT FROM AUTHOR]

One of the main factors that affect the accuracy of intensity-based registration of two-dimensional (2D) X-ray fluoroscopy to three-dimensional (3D) CT data is the similarity measure, which is a criterion function that is used in the registration procedure for measuring the quality of image match. This paper presents a unifying framework for rationally deriving point similarity measures based on Markov random field (MRF) modeling of difference images which are obtained by comparing the reference fluoroscopic images with their associated digitally reconstructed radiographs (DRR’s). The optimal solution is defined as the maximum a posterior (MAP) estimate of the MRF. Three novel point similarity measures derived from this framework are presented. They are evaluated using a phantom and a human cadaveric specimen. Combining any one of the newly proposed similarity measures with a previously introduced spline-based registration scheme, we develop a fast and accurate registration algorithm. We report their capture ranges, converging speeds, and registration accuracies. [ABSTRACT FROM AUTHOR]