Image registrations that are based on similarity measures simply adjust the parameters of an appropriate spatial transformation model until the similarity measure reaches an optimum. The numerous similarity measures that have been proposed in the past are differently sensitive to imaging modality, image content and differences in the image content, selection of the floating and target image, partial image overlap, etc. In this paper, we evaluate and compare 12 similarity measures for the rigid registration. To study the impact of different imaging modalities on the behavior of similarity measures, we have used 16 CT/MR and 6 PET/MR image pairs with known 'gold standard' registrations. The results for the PET/MR registration and for the registration of CT to both rectified and unrectified MR images indicate that mutual information, normalized mutual information and the entropy correlation coefficient are the most accurate similarity measures and have the smallest risk of being trapped in a local optimum. The results of an experiment on the impact of exchanging the floating and target image indicate that, especially in MR/PET registrations, the behavior of some similarity measures, such as mutual information, significantly depends on which image is the floating and which is the target.

Purpose: A promising patient positioning technique is based on registering computed tomographic (CT) or magnetic resonance (MR) images to cone-beam CT images (CBCT). The extra radiation dose delivered to the patient can be substantially reduced by using fewer projections. This approach results in lower quality CBCT images. The purpose of this study is to evaluate a number of similarity measures (SMs) suitable for registration of CT or MR images to low-quality CBCTs. Methods and Materials: Using the recently proposed evaluation protocol, we evaluated nine SMs with respect to pretreatment imaging modalities, number of two-dimensional (2D) images used for reconstruction, and number of reconstruction iterations. The image database consisted of 100 X-ray and corresponding CT and MR images of two vertebral columns. Results: Using a higher number of 2D projections or reconstruction iterations results in higher accuracy and slightly lower robustness. The similarity measures that behaved the best also yielded the best registration results. The most appropriate similarity measure was the asymmetric multi-feature mutual information (AMMI). Conclusions: The evaluation protocol proved to be a valuable tool for selecting the best similarity measure for the reconstruction-based registration. The results indicate that accurate and robust CT/CBCT or even MR/CBCT registrations are possible if the AMMI similarity measure is used.

We propose a completely automated algorithm for the detection of the spinal centreline and the centres of vertebral bodies and intervertebral discs in images acquired by computed tomography (CT) and magnetic resonance (MR) imaging. The developed methods are based on the analysis of the geometry of spinal structures and the characteristics of CT and MR images and were evaluated on 29 CT and 13 MR images of lumbar spine. The overall mean distance between the obtained and the ground truth spinal centrelines and centres of vertebral bodies and intervertebral discs were 1.8 +/- 1.1 mm and 2.8 +/- 1.9 mm, respectively, and no considerable differences were detected among the results for CT, T(1)-weighted MR and T(2)-weighted MR images. The knowledge of the location of the spinal centreline and the centres of vertebral bodies and intervertebral discs is valuable for the analysis of the spine. The proposed method may therefore be used to initialize the techniques for labelling and segmentation of vertebrae.

In the past, a number of methods were proposed for quantitative assessment of vertebral rotation from three-dimensional (3D) images. However, these methods were based on manual identification of distinctive anatomical landmarks, required manual determination of cross-sections from 3D images, and measured only axial vertebral rotation instead of the rotation in 3D. In this paper, we propose an automated method for quantitative assessment of vertebral rotation in 3D that is based on finding the planes of vertebral symmetry by matching image intensity gradients on both sides of each plane. The method was evaluated on 28 images of normal and pathological vertebrae, obtained by computed tomography (CT) and magnetic resonance (MR). For each vertebra, final angle displacements of 200 initial angle displacements, uniformly distributed within 30 degrees from manually obtained reference angles, were obtained. The results show that by the proposed method, vertebral rotation can be successfully estimated in 3D with an average accuracy of 1.0 degrees and precision of 0.5 degrees.

In this paper, a novel method for EEG to MRI registration is proposed. Initial registration is achieved by extracting and matching symmetry planes of MRI and EEG data, followed by iterative registration based on minimizing a cost function. Comparison of the intensity distributions of the whole MR image and MRI voxels around a head surface point yields global similarities, while the comparison of intensity distributions of MRI voxels around corresponding EEG points, which reflects the head's sagittal symmetry, yields local similarities. Therefore, when the EEG points are registered to the MR image, maximal global and local similarities should be obtained. The cost function, incorporating global and local similarities, was the sum of Kullback-Leibler divergences between corresponding intensity distributions. The proposed method was evaluated on clinical MRI data with simulated EEG data, yielding mean registration error of 0.48 +/- 0.33 mm, while with real EEG data an average root-mean-square point-to-surface error of 2.27 +/- 0.02 mm was obtained.

The purpose of this study is to present a framework for quantitative analysis of spinal curvature in 3D. In order to study the properties of such complex 3D structures, we propose two descriptors that capture the characteristics of spinal curvature in 3D. The descriptors are the geometric curvature (GC) and curvature angle (CA), which are independent of the orientation and size of spine anatomy. We demonstrate the two descriptors that characterize the spinal curvature in 3D on 30 computed tomography (CT) images of normal spine and on a scoliotic spine. The descriptors are determined from 3D vertebral body lines, which are obtained by two different methods. The first method is based on the least-squares technique that approximates the manually identified vertebra centroids, while the second method searches for vertebra centroids in an automated optimization scheme, based on computer-assisted image analysis. Polynomial functions of the fourth and fifth degree were used for the description of normal and scoliotic spinal curvature in 3D, respectively. The mean distance to vertebra centroids was 1.1 mm (+/-0.6 mm) for the first and 2.1 mm (+/-1.4 mm) for the second method. The distributions of GC and CA values were obtained along the 30 images of normal spine at each vertebral level and show that maximal thoracic kyphosis (TK), thoracolumbar junction (TJ) and maximal lumbar lordosis (LL) on average occur at T3/T4, T12/L1 and L4/L5, respectively. The main advantage of GC and CA is that the measurements are independent of the orientation and size of the spine, thus allowing objective intra- and inter-subject comparisons. The positions of maximal TK, TJ and maximal LL can be easily identified by observing the GC and CA distributions at different vertebral levels. The obtained courses of the GC and CA for the scoliotic spine were compared to the distributions of GC and CA for the normal spines. The significant difference in values indicates that the descriptors of GC and CA may be used to detect and quantify scoliotic spinal curvatures. The proposed framework may therefore improve the understanding of spine anatomy and aid in the clinical quantitative evaluation of spinal deformities.

An important part of image-guided radiation therapy or surgery is registration of a three-dimensional (3D) preoperative image to two-dimensional (2D) images of the patient. It is expected that the accuracy and robustness of a 3D/2D image registration method do not depend solely on the registration method itself but also on the number and projections (views) of intraoperative images. In this study, we systematically investigate these factors by using registered image data, comprising of CT and X-ray images of a cadaveric lumbar spine phantom and the recently proposed 3D/2D registration method. The results indicate that the proportion of successful registrations (robustness) significantly increases when more X-ray images are used for registration.

A novel method for automated curved planar reformation (CPR) of magnetic resonance (MR) images of the spine is presented. The CPR images, generated by a transformation from image-based to spine-based coordinate system, follow the structural shape of the spine and allow the whole course of the curved anatomy to be viewed in individual cross-sections. The three-dimensional (3D) spine curve and the axial vertebral rotation, which determine the transformation, are described by polynomial functions. The 3D spine curve passes through the centres of vertebral bodies, while the axial vertebral rotation determines the rotation of vertebrae around the axis of the spinal column. The optimal polynomial parameters are obtained by a robust refinement of the initial estimates of the centres of vertebral bodies and axial vertebral rotation. The optimization framework is based on the automatic image analysis of MR spine images that exploits some basic anatomical properties of the spine. The method was evaluated on 21 MR images from 12 patients and the results provided a good description of spine anatomy, with mean errors of 2.5 mm and 1.7 degrees for the position of the 3D spine curve and axial rotation of vertebrae, respectively. The generated CPR images are independent of the position of the patient in the scanner while comprising both anatomical and geometrical properties of the spine.

We present a novel method for curved planar reformation (CPR) of spine images obtained by magnetic resonance (MR) imaging. CPR images, created via a transformation from image-based to spine-based coordinate system, follow the structural shape of the spine and allow the whole course of the curved structure to be viewed in a single image. The spine-based coordinate system is defined on the 3D spine curve and on the axial vertebral rotation, both described by polynomial models. The 3D spine curve passes through the centers of vertebral bodies, and the axial vertebral rotation determines the rotation of vertebral spinous processes around the spine. The optimal polynomial parameters are found in an optimization framework, based on image analysis. The method was evaluated on 19 MR images of the spine from 10 patients.

Medical image acquisition devices provide a vast amount of anatomical and functional information, which facilitate and improve diagnosis and patient treatment, especially when supported by modern quantitative image analysis methods. However, modality specific image artifacts, such as the phenomena of intensity inhomogeneity in magnetic resonance images (MRI), are still prominent and can adversely affect quantitative image analysis. In this paper, numerous methods that have been developed to reduce or eliminate intensity inhomogeneities in MRI are reviewed. First, the methods are classified according to the inhomogeneity correction strategy. Next, different qualitative and quantitative evaluation approaches are reviewed. Third, 60 relevant publications are categorized according to several features and analyzed so as to reveal major trends, popularity, evaluation strategies and applications. Finally, key evaluation issues and future development of the inhomogeneity correction field, supported by the results of the analysis, are discussed.

This paper presents a novel automatic framework to perform 3D face recognition. The proposed method uses a Simulated Annealing-based approach (SA) for range image registration with the Surface Interpenetration Measure (SIM), as similarity measure, in order to match two face images. The authentication score is obtained by combining the SIM values corresponding to the matching of four different face regions: circular and elliptical areas around the nose, forehead, and the entire face region. Then, a modified SA approach is proposed taking advantage of invariant face regions to better handle facial expressions. Comprehensive experiments were performed on the FRGC v2 database, the largest available database of 3D face images composed of 4,007 images with different facial expressions. The experiments simulated both verification and identification systems and the results compared to those reported by state-of-the-art works. By using all of the images in the database, a verification rate of 96.5 percent was achieved at a False Acceptance Rate (FAR) of 0.1 percent. In the identification scenario, a rank-one accuracy of 98.4 percent was achieved. To the best of our knowledge, this is the highest rank-one score ever achieved for the FRGC v2 database when compared to results published in the literature.

The aging population and the growing amount of medical data have increased the need for automated tools in the neurology departments. Although the researchers have been developing computerized methods to help the medical expert, these efforts have primarily emphasized improving the effectiveness in single patient data, such as computing a brain lesion size. However, patient-to-patient comparison that should help improve diagnosis and therapy has not received as much attention. To this effect, this paper introduces a fast and robust region-of-interest retrieval method for brain magnetic resonance (MR) images. We make several contributions to the domains of brain MR image analysis, and search and retrieval: 1)We show the potential and robustness of local structure information in the search and retrieval of brain MR images. 2)We provide analysis of two complementary features, local binary patterns and Kanade-Lucas-Tomasi feature points, and their comparison with a baseline method. 3)We show that incorporating spatial context in the features substantially improves accuracy. 4)We automatically extract dominant local binary patterns and demonstrate their effectiveness relative to the conventional local binary pattern approach. Comprehensive experiments on real and simulated datasets revealed that dominant local binary patterns with spatial context is robust to geometric deformations and intensity variations and have high accuracy and speed even in pathological cases. The proposed method can not only aid the medical expert in disease diagnosis, or be used in scout (localizer) scans for optimization of acquisition parameters, but also support low power handheld devices.

PURPOSE: Robust and accurate automated co-registration of the coronary arteries in 3D CTA and 2D X-ray angiography during percutaneous coronary interventions (PCI), in order to present a fused visualization. METHODS: A novel vesselness-based similarity measure was developed, that avoids an explicit segmentation of the X-ray image. A stochastic optimizer searches the optimal registration using the similarity measure. RESULTS: Both simulated data and clinical data were used to investigate the accuracy and capture range of the proposed method. The experiments show that the proposed method outperforms the iterative closest point method in terms of accuracy (average residual error of 0.42 mm vs. 1.44 mm) and capture range (average 71.1 mm/20.3 degrees vs. 14.1 mm/5.2 degrees ). CONCLUSION: The proposed method has proven to be accurate and the capture range is ample for usage in PCI. Especially the absence of an explicit segmentation of the interventionally acquired X-ray images considerably aids the robustness of the method.

A novel similarity measure for registering magnetic resonance (MR) and computed tomography (CT) images has been designed and built. MR-CT registration methods often rely on the statistical intensity relationship between the images. The proposed similarity measure instead depends on the statistical relationship between the complex phase order between the images. By utilizing the complex phase order likelihood (CPOL) as a similarity measure, structural relationships instead of intensity relationships are explicitly used. This approach can be advantageous for MR-CT registration, where the intensities of the CT imagery have highly complex and nonlinear relationships with the intensities of corresponding MR imagery but simpler linear structural relationships. This new similarity measure has been tested on real MR-CT 3D volumes and has been evaluated based on fiducial registration error to determine alignment accuracy. Quantitative results show that CPOL is capable of achieving comparable alignment accuracy when compared to normalized mutual information, while being more robust to imaging artifacts such as noise.

We evaluated 4 volume-based automatic image registration algorithms from 2 commercially available treatment planning systems (Philips Syntegra and BrainScan). The algorithms based on cross correlation (CC), local correlation (LC), normalized mutual information (NMI), and BrainScan mutual information (BSMI) were evaluated with:(1) the synthetic computed tomography (CT) images,(2) the CT and magnetic resonance (MR) phantom images, and (3) the CT and MR head image pairs from 12 patients with brain tumors. For the synthetic images, the registration results were compared with known transformation parameters, and all algorithms achieved accuracy of submillimeter in translation and subdegree in rotation. For the phantom images, the registration results were compared with those provided by frame and marker-based manual registration. For the patient images, the results were compared with anatomical landmark-based manual registration to qualitatively determine how the results were close to a clinically acceptable registration. NMI and LC outperformed CC and BSMI, with the sense of being closer to a clinically acceptable result. As for the robustness, NMI and BSMI outperformed CC and LC. A guideline of image registration in our institution was given, and final visual assessment is necessary to guarantee reasonable results.

The purpose of this study was to evaluate the potential for use of image volume based registration (IVBaR) to aid in measurement of changes in the tumor during chemotherapy of breast cancer. Successful IVBaR could aid in the detection of such changes in response to neoadjuvant chemotherapy and potentially be useful for routine breast cancer screening and diagnosis. IVBaR was employed in a new method of automated estimation of tumor volume in studies following the radiologist identification of the tumor region in the prechemotherapy scan. The authors have also introduced a new semiautomated method for validation of registration based on Doppler ultrasound (U.S.) signals that are independent of the grayscale signals used for registration. This Institutional Review Board approved study was conducted on 10 patients undergoing chemotherapy and 14 patients with a suspicious/unknown mass scheduled to undergo biopsy. Reasonably reproducible mammographic positioning and nearly whole breast U.S. imaging were achieved. The image volume was registered offline with a mutual information cost function and global interpolation based on a thin-plate spline using MIAMI FUSE software developed at the University of Michigan. The success and accuracy of registration of the three dimensional (3D) U.S. image volume were measured by means of mean registration error (MRE). IVBaR was successful with MRE of 4.3 +/- 1.7 mm in 9 out of 10 reproducibility automated breast ultrasound (ABU) studies and in 12 out of 17 ABU image pairs collected before, during, or after 115 +/- 14 days of chemotherapy. Semiautomated tumor volume estimation was performed on registered image volumes giving 86 +/- 8% mean accuracy compared to the radiologist hand-segmented tumor volume on seven cases. Doppler studies yielded fractional volume of color pixels in the region surrounding the lesion and its change with changing breast compression. The Doppler study of patients with detectable blood flow included five patients with suspicious masses and three undergoing chemotherapy. Spatial alignment of the 3D blood vessel data from the Doppler studies provided independent measures for the validation of registration. In 15 Doppler image volume pairs scanned with differing breast compression, the mean centerline separation value was 1.5 +/- 0.6 mm, while MRE based on a few identifiable structural points common to the two grayscale image volumes was 1.1 +/- 0.6 mm. Another measure, the overlap ratio of blood vessels, was shown to increase from 0.32 to 0.59 (+84%) with IVBaR for pairs at various compression levels. These results show that successful registration of ABU scans may be accomplished for comparison and integration of information.

In this paper, an automatic histogram threshold approach based on a fuzziness measure is presented. This work is an improvement of an existing method. Using fuzzy logic concepts, the problems involved in finding the minimum of a criterion function are avoided. Similarity between gray levels is the key to find an optimal threshold. Two initial regions of gray levels, located at the boundaries of the histogram, are defined. Then, using an index of fuzziness, a similarity process is started to find the threshold point. A significant contrast between objects and background is assumed. Previous histogram equalization is used in small contrast images. No prior knowledge of the image is required.

In this article, the authors evaluate a merit function for 2D/3D registration called stochastic rank correlation (SRC). SRC is characterized by the fact that differences in image intensity do not influence the registration result; it therefore combines the numerical advantages of cross correlation (CC)-type merit functions with the flexibility of mutual-information-type merit functions. The basic idea is that registration is achieved on a random subset of the image, which allows for an efficient computation of Spearman's rank correlation coefficient. This measure is, by nature, invariant to monotonic intensity transforms in the images under comparison, which renders it an ideal solution for intramodal images acquired at different energy levels as encountered in intrafractional kV imaging in image-guided radiotherapy. Initial evaluation was undertaken using a 2D/3D registration reference image dataset of a cadaver spine. Even with no radiometric calibration, SRC shows a significant improvement in robustness and stability compared to CC. Pattern intensity, another merit function that was evaluated for comparison, gave rather poor results due to its limited convergence range. The time required for SRC with 5% image content compares well to the other merit functions; increasing the image content does not significantly influence the algorithm accuracy. The authors conclude that SRC is a promising measure for 2D/3D registration in IGRT and image-guided therapy in general.

We propose a novel Bayesian registration formulation in which image location is represented as a latent random variable. Location is marginalized to determine the maximum a priori (MAP) transform between images, which results in registration that is more robust than the alternatives of omitting locality (i.e. global registration) or jointly maximizing locality and transform (i.e. iconic registration). A mathematical link is established between the Bayesian registration formulation and the mutual information (MI) similarity measure. This leads to a novel technique for selecting informative image regions for registration, based on the MI of image intensity and spatial location. Experimental results demonstrate the effectiveness of the marginalization formulation and the MI-based region selection technique for ultrasound (US) to magnetic resonance (MR) registration in an image-guided neurosurgical application.

Non-rigid image registration is a challenging task in medical image analysis. In recent years, there are two essential issues. First, intensity similarity is not necessarily equivalent to anatomical similarity when the anatomical correspondences between subject and template images are established. Second, the registration algorithm should be robust against monotonic gray-level transformation when aligning anatomical structures in the presence of bias fields. In this paper, a new feature based non-rigid registration method is proposed to deal with these two problems. The proposed method is based on a new type of image feature, called Uniform Spherical Structure Pattern (USSP). USSP encodes voxel-wise interaction information and geometric properties of anatomical structures. It is computationally efficient, rotation invariant and theoretically monotonic gray-level transformation invariant. The USSP feature is integrated with the Markov random field (MRF) discrete labeling framework to define energy function for registration in this paper. If the segmentation results are available, explicit anatomical correspondence can be established as an additional energy term. The energy function is optimized via the alpha-expansion algorithms. The proposed method is compared with three widely used non-rigid registration methods on both simulated and real databases obtained from BrainWeb and IBSR. Experimental results demonstrate that the proposed method achieves the highest registration accuracy among all the compared methods.