Registration Error and Intensity Similarity Based Label Fusion for Segmentation

Background

Label fusion is a core step of Multi-Atlas Segmentation (MAS), which has a decisive effect on segmentation results. Although existed strategies using image intensity or image shape to fuse labels have got acceptable results, there is still necessity for further performance improvement. Here, we propose a new label fusion strategy, which considers the joint information of intensity and registration quality.

Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands

Histology volume reconstruction facilitates the study of 3D shape and volume change of an organ at the level of macrostructures made up of cells. It can also be used to investigate and validate novel techniques and algorithms in volumetric medical imaging and therapies. Creating 3D high-resolution atlases of different organs^1,2,3 is another application of histology volume reconstruction. This provides a resource for investigating tissue structures and the spatial relationship between various cellular features. We present an image registration approach for histology volume reconstruction, which uses a set of optical blockface images. The reconstructed histology volume represents a reliable shape of the processed specimen with no propagated post-processing registration error. The Hematoxylin and Eosin (H&E) stained sections of two mouse mammary glands were registered to their corresponding blockface images using boundary points extracted from the edges of the specimen in histology and blockface images. The accuracy of the registration was visually evaluated. The alignment of the macrostructures of the mammary glands was also visually assessed at high resolution.This study delineates the different steps of this image registration pipeline, ranging from excision of the mammary gland through to 3D histology volume reconstruction. While 2D histology images reveal the structural differences between pairs of sections, 3D histology volume provides the ability to visualize the differences in shape and volume of the mammary glands.     

Magnetic Resonance Imaging of the Newborn Brain: Automatic Segmentation of Brain Images into 50 Anatomical Regions

We studied methods for the automatic segmentation of neonatal and developing brain images into 50 anatomical regions, utilizing a new set of manually segmented magnetic resonance (MR) images from 5 term-born and 15 preterm infants imaged at term corrected age called ALBERTs. Two methods were compared: individual registrations with label propagation and fusion; and template based registration with propagation of a maximum probability neonatal ALBERT (MPNA). In both cases we evaluated the performance of different neonatal atlases and MPNA, and the approaches were compared with the manual segmentations by means of the Dice overlap coefficient. Dice values, averaged across regions, were 0.81±0.02 using label propagation and fusion for the preterm population, and 0.81±0.02 using the single registration of a MPNA for the term population. Segmentations of 36 further unsegmented target images of developing brains yielded visibly high-quality results. This registration approach allows the rapid construction of automatically labeled age-specific brain atlases for neonates and the developing brain.     

Bayesian estimation of probabilistic atlas for tissue segmentation

Automatic anatomical brain image segmentation is still a challenge. In particular, algorithms have to address the partial volume effect (PVE) as well as the variability of the gray level of internal brain structures which may appear closer to gray matter (GM) than white matter (WM). Atlas based segmentation is one solution as it brings prior information. For such tasks, probabilistic atlases are very useful as they take account of the PVE information. In this paper, we provide a detailed analysis of a generative statistical model based on dense deformable templates that represents several tissue types observed in medical images. The inputs are gray level data whereas our atlas is composed of both an estimation of the deformation metric and probability maps of each tissue (called class). This atlas is used to guide the tissue segmentation of new images. Experiments are shown on brain T1 MRI datasets. This method only requires approximate pre-registration, as the latter is done jointly with the segmentation. Note however that an approximate registration is a reasonable pre-requisite given the application.     

Anatomical atlas in the context of head and neck radiotherapy and its use to automatic segmentation

In this paper we present a methodology to form an anatomical atlas based on the analysis of dense deformation fields recovered by the Morphons non-rigid registration algorithm. The methodology is based on measuring the bending energy required to register the whole database to a reference, and the atlas is the one image in the database which yields the smallest bending energy when taken as reference. The suitability of our atlas is demonstrated in the context of head and neck radiotherapy through its application to a database with thirty-one computed tomography (CT) images of the head and neck region. In head and neck radiotherapy, CT is the most frequently used modality for the segmentation of organs at risk and clinical target volumes. One challenge brought by the use of CT images is the presence of important artifacts caused by dental implants. The presence of such artifacts hinders the use of intensity averages, thus severely limiting the application of most atlas building techniques described in the literature in this context. The results presented in the paper show that our bending energy model faithfully represents the shape variability of patients in the head and neck region; they also show its good performance in segmentation of volumes of interest in radiotherapy. Moreover, when compared to other atlases of similar performance in automatic segmentation, our atlas presents the desirable feature of not being blurred after intensity averaging.     

3-Dimensional Diffusion Tensor Imaging (DTI) Atlas of the Rat Brain

Anatomical atlases play an important role in the analysis of neuroimaging data in rodent neuroimaging studies. Having a high resolution, detailed atlas not only can expand understanding of rodent brain anatomy, but also enables automatic segmentation of new images, thus greatly increasing the efficiency of future analysis when applied to new data. These atlases can be used to analyze new scans of individual cases using a variety of automated segmentation methods. This project seeks to develop a set of detailed 3D anatomical atlases of the brain at postnatal day 5 (P5), 14 (P14), and adults (P72) in Sprague-Dawley rats. Our methods consisted of first creating a template image based on fixed scans of control rats, then manually segmenting various individual brain regions on the template. Using itk-SNAP software, subcortical and cortical regions, including both white matter and gray matter structures, were manually segmented in the axial, sagittal, and coronal planes. The P5, P14, and P72 atlases had 39, 45, and 29 regions segmented, respectively. These atlases have been made available to the broader research community.     

Rapid pedobarographic image registration based on contour curvature and optimization

Image registration, the process of optimally aligning homologous structures in multiple images, has recently been demonstrated to support automated pixel-level analysis of pedobarographic images and, subsequently, to extract unique and biomechanically relevant information from plantar pressure data. Recent registration methods have focused on robustness, with slow but globally powerful algorithms. In this paper, we present an alternative registration approach that affords both speed and accuracy, with the goal of making pedobarographic image registration more practical for near-real-time laboratory and clinical applications. The current algorithm first extracts centroid-based curvature trajectories from pressure image contours, and then optimally matches these curvature profiles using optimization based on dynamic programming. Special cases of disconnected images (that occur in high-arched subjects, for example) are dealt with by introducing an artificial spatially linear bridge between adjacent image clusters. Two registration algorithms were developed: a ‘geometric’ algorithm, which exclusively matched geometry, and a ‘hybrid’ algorithm, which performed subsequent pseudo-optimization. After testing the two algorithms on 30 control image pairs considered in a previous study, we found that, when compared with previously published results, the hybrid algorithm improved overlap ratio (p=0.010), but both current algorithms had slightly higher mean-squared error, assumedly because they did not consider pixel intensity. Nonetheless, both algorithms greatly improved the computational efficiency (25±8 and 53±9 ms per image pair for geometric and hybrid registrations, respectively). These results imply that registration-based pixel-level pressure image analyses can, eventually, be implemented for practical clinical purposes.     

超声图像处理中Snake模型研究

Snake模型是一种基于高层信息的有效目标轮廓提取算法,其优点是作用过程及最后结果的目标轮廓是一条完整的曲线,从而引起广泛的关注。鉴于医学超声图像的信噪比较低,用经典的边缘提取算法无法得到较好的结果,因此人们将Snake模型进行了各种各样的改进,并且越来越多地将它运用到医学超声图像处理中来。本文对乳腺超声图像进行阈值分割、形态滤波等一系列预处理后,将改进的Snake模型对乳腺超声图像进行肿瘤的边缘提取,得到了比较好的结果。     

Multisegmental image fusion of the spine]

Fusion of medical images is a technique that permits the correlation of homologous anatomical structures in different imaging modalities on the basis of a spatial transformation of the data sets. CT and MRI of the spine provide complementary information of possible relevance for diagnostic and therapeutic decisions. Methods enabling a multisegmental CT-MRI fusion of the spine were developed. These solve the problem of altered spatial relationships of the individual anatomical structures due to differing patient positioning in successive data acquisitions. Routine clinical CT and MRI data of a thoracic section of the spine were obtained and transferred to a PC-workstation. Following segmentation of the CT-data, landmarks for each individual vertebra were defined in the CT and MRI data. For each individual vertebra the algorithm we developed then carried out a rigid registration of the CT information to the MR data. The fused data sets were presented as colour-coded images or on the basis of dynamic variation of transparency. To assess registration precision, fiducial registration errors (FRE) and target registration errors (TRE) were calculated. The algorithm permitted multi-segmental image fusion of the spine. The average time required for defining the landmarks was 22 seconds per landmark for CT, and 34 seconds per landmark for MR. The average FRE was 1.53 mm. The TRE for the vertebrae was less than 2 mm. The colour-coded images were particularly suitable for assessing the contours of the anatomical structures, whereas dynamic variation of the transparency of overlapping CT images enabled a better overall assessment of the spatial relationship of the anatomical structures. The algorithm permits precise multi-segmental fusion of CT and MR of the spine, which was not possible using current fusion-algorithms due to variations in the spatial orientation of the anatomical structures caused by different positioning of the axial skeleton in successive examinations.     

Registration of fundus images for generating wide field composite images of the retina ]

The composition of retinal images presents high demands to the applied methods. Substantially different lighting conditions between the images, glarings and fade-outs within one image, large textureless regions and non-linear distortions are the main challenges. We present a fully automatic algorithm for the registration of images of the human retina and their overlay to wide field montage images combining area-based and point-based approaches. The algorithm combines an area-based as well as a point-based approach for determining similarities between images. Various measures of similarity were investigated, where the normalized correlation coefficient was superior compared to the usual definitions of transinformation. The transformation of the images was based on a quadratic model that can be derived from the spherical surface of the retina. This model was compared to four other parameterized transformations and performed best both visually and quantitatively in terms of measured misregistration. Problems may occur if the images are extremely defocused or contain very little relevant structural information.     

Using Manifold Learning for Atlas Selection in Multi-Atlas Segmentation

Multi-atlas segmentation has been widely used to segment various anatomical structures. The success of this technique partly relies on the selection of atlases that are best mapped to a new target image after registration. Recently, manifold learning has been proposed as a method for atlas selection. Each manifold learning technique seeks to optimize a unique objective function. Therefore, different techniques produce different embeddings even when applied to the same data set. Previous studies used a single technique in their method and gave no reason for the choice of the manifold learning technique employed nor the theoretical grounds for the choice of the manifold parameters. In this study, we compare side-by-side the results given by 3 manifold learning techniques (Isomap, Laplacian Eigenmaps and Locally Linear Embedding) on the same data set. We assess the ability of those 3 different techniques to select the best atlases to combine in the framework of multi-atlas segmentation. First, a leave-one-out experiment is used to optimize our method on a set of 110 manually segmented atlases of hippocampi and find the manifold learning technique and associated manifold parameters that give the best segmentation accuracy. Then, the optimal parameters are used to automatically segment 30 subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). For our dataset, the selection of atlases with Locally Linear Embedding gives the best results. Our findings show that selection of atlases with manifold learning leads to segmentation accuracy close to or significantly higher than the state-of-the-art method and that accuracy can be increased by fine tuning the manifold learning process.     

Optical-CT Dual-Modality Mapping Base on DRR Registration

Optical-CT dual-modality imaging requires the mapping between 2D fluorescence images and 3D body surface light flux. In this paper, we proposed an optical-CT dual-modality image mapping algorithm based on the Digitally Reconstructed Radiography (DRR) registration. In the process of registration, a series of DRR images were computed from CT data using the ray casting algorithm. Then, the improved HMNI similarity strategy based on Hausdorff distance was used to complete the registration of the white-light optical images and DRR virtual images. According to the corresponding relationship obtained by the image registration and the Lambert’s cosine law based on the pin-hole imaging model, the 3D light intensity distribution on the surface of the object could be solved. The feasibility and effectiveness of the mapping algorithm are verified by the irregular phantom and mouse experiments.     

Estimation of the deformations induced by articulated bodies: Registration of the spinal column

We present a new non-rigid registration algorithm estimating the displacement field generated by articulated bodies. Indeed the bony structures between different patient images may rigidly move while other tissues may deform in a more complex way. Our algorithm tracks the displacement induced in the column by a movement of the patient between two acquisitions. The volumetric deformation field in the whole body is then inferred from those displacements using a linear elastic biomechanical finite element model. We demonstrate in this paper that this method provides accurate results on 3D sets of computed tomography (CT), MR and positron emission tomography (PET) images and that the results of the registration algorithm show significant decreases in the mean, min and max errors.     

Multi-Contrast Multi-Atlas Parcellation of Diffusion Tensor Imaging of the Human Brain

In this paper, we propose a novel method for parcellating the human brain into 193 anatomical structures based on diffusion tensor images (DTIs). This was accomplished in the setting of multi-contrast diffeomorphic likelihood fusion using multiple DTI atlases. DTI images are modeled as high dimensional fields, with each voxel exhibiting a vector valued feature comprising of mean diffusivity (MD), fractional anisotropy (FA), and fiber angle. For each structure, the probability distribution of each element in the feature vector is modeled as a mixture of Gaussians, the parameters of which are estimated from the labeled atlases. The structure-specific feature vector is then used to parcellate the test image. For each atlas, a likelihood is iteratively computed based on the structure-specific vector feature. The likelihoods from multiple atlases are then fused. The updating and fusing of the likelihoods is achieved based on the expectation-maximization (EM) algorithm for maximum a posteriori (MAP) estimation problems. We first demonstrate the performance of the algorithm by examining the parcellation accuracy of 18 structures from 25 subjects with a varying degree of structural abnormality. Dice values ranging 0.8–0.9 were obtained. In addition, strong correlation was found between the volume size of the automated and the manual parcellation. Then, we present scan-rescan reproducibility based on another dataset of 16 DTI images – an average of 3.73%, 1.91%, and 1.79% for volume, mean FA, and mean MD respectively. Finally, the range of anatomical variability in the normal population was quantified for each structure.     

Designing a winter bird atlas field methodology: issues of time and space in sampling and interactions with habitat

Faunal and floral atlases can provide invaluable information on species distributions and relative abundance, and have played a key role in detecting and diagnosing population changes. Most atlases rely on simple but effective field methods that can be employed by large numbers of volunteer surveyors so as to make the best use of their finite effort. For ornithological atlases, such methods have been modified to better quantify relative abundance in the breeding season. Here, we evaluate how effort should be invested in space (number of samples) and time (time per sample) in order to optimise species lists and relative abundance estimation for birds in Britain and Ireland in winter. Species accumulation curves show only slight differences in the ability to derive complete species lists whether effort is invested in time or in space. However, relative abundance precision is markedly improved if effort is invested in space at the expense of time. These results give clear guidance on the general way in which methods should be devised (i.e. more samples, each for less time), but analyses relating the shape of species accumulation curves to habitat diversity suggest the precise trade-off between surveying in space and time varies between landscapes. Designers of atlases must therefore optimise their study design for the landscape of the particular region in question.     

eHUGS: Enhanced Hierarchical Unbiased Graph Shrinkage for Efficient Groupwise Registration

Effective and efficient spatial normalization of a large population of brain images is critical for many clinical and research studies, but it is technically very challenging. A commonly used approach is to choose a certain image as the template and then align all other images in the population to this template by applying pairwise registration. To avoid the potential bias induced by the inappropriate template selection, groupwise registration methods have been proposed to simultaneously register all images to a latent common space. However, current groupwise registration methods do not make full use of image distribution information for more accurate registration. In this paper, we present a novel groupwise registration method that harnesses the image distribution information by capturing the image distribution manifold using a hierarchical graph with its nodes representing the individual images. More specifically, a low-level graph describes the image distribution in each subgroup, and a high-level graph encodes the relationship between representative images of subgroups. Given the graph representation, we can register all images to the common space by dynamically shrinking the graph on the image manifold. The topology of the entire image distribution is always maintained during graph shrinkage. Evaluations on two datasets, one for 80 elderly individuals and one for 285 infants, indicate that our method can yield promising results.     

Intraoperative Biomechanical Registration of the Liver: Does the Heterogeneity of the Liver Matter?

Background: Preoperative images such as computed tomography scans or magnetic resonance imaging contain lots of valuable information that are not easily available for surgeons during an operation. To help the clinicians better target the structures of interest during an intervention, many registration methods that align preoperative images onto the intraoperative view of the organs have been developed. For important organ deformation, biomechanically-based registration has proven to be a method of choice.Method: Using an existing biomechanically-based registration algorithm for laparoscopic liver surgery we investigate in this paper the influence of the heterogeneity of the liver on the registration result.Results: No statistical difference in the results was found between the registration performed with the homogeneous model and the one carried out with the heterogeneous model.Conclusion: As the use of an heterogeneous model does not improve significantly the registration result and increase the computation time we recommend to perform the type of registration task described in the paper with a simplified homogeneous model.     

A Robust and Accurate Two-Step Auto-Labeling Conditional Iterative Closest Points (TACICP) Algorithm for Three-Dimensional Multi-Modal Carotid Image Registration

Atherosclerosis is among the leading causes of death and disability. Combining information from multi-modal vascular images is an effective and efficient way to diagnose and monitor atherosclerosis, in which image registration is a key technique. In this paper a feature-based registration algorithm, Two-step Auto-labeling Conditional Iterative Closed Points (TACICP) algorithm, is proposed to align three-dimensional carotid image datasets from ultrasound (US) and magnetic resonance (MR). Based on 2D segmented contours, a coarse-to-fine strategy is employed with two steps: rigid initialization step and non-rigid refinement step. Conditional Iterative Closest Points (CICP) algorithm is given in rigid initialization step to obtain the robust rigid transformation and label configurations. Then the labels and CICP algorithm with non-rigid thin-plate-spline (TPS) transformation model is introduced to solve non-rigid carotid deformation between different body positions. The results demonstrate that proposed TACICP algorithm has achieved an average registration error of less than 0.2mm with no failure case, which is superior to the state-of-the-art feature-based methods.     

Pair-wise vs group-wise registration in statistical shape model construction: representation of physiological and pathological variability of bony surface morphology

Statistical shape models (SSM) of bony surfaces have been widely proposed in orthopedics, especially for anatomical bone modeling, joint kinematic analysis, staging of morphological abnormality, and pre- and intra-operative shape reconstruction. In the SSM computation, reference shape selection, shape registration and point correspondence computation are fundamental aspects determining the quality (generality, specificity and compactness) of the SSM. Such procedures can be made critical by the presence of large morphological dissimilarities within the surfaces, not only because of anthropometrical variability but also mainly due to pathological abnormalities. In this work, we proposed a SW pipeline for SSM construction based on pair-wise (PW) shape registration, which requires the a-priori selection of the reference shape, and on a custom iterative point correspondence algorithm. We addressed large morphological deformations in five different bony surface sets, namely proximal femur, distal femur, patella, proximal fibula and proximal tibia, extracted from a retrospective patient dataset. The technique was compared to a method from the literature, based on group-wise (GW) shape registration. As a main finding, the proposed technique provided generalization and specificity median errors, for all the five bony regions, lower than 2?mm. The comparative analysis provided basically similar results. Particularly, for the distal femur that was the shape affected by the largest pathological deformations, the differences in generalization, specificity and compactness were lower than 0.5?mm, 0.5?mm, and 1%, respectively. We can argue the proposed pipeline, along with the robust correspondence algorithm, is able to compute high-quality SSM of bony shapes, even affected by large morphological variability.