Medical Image Computing and Computer-Assisted Intervention – MICCAI 2006

This paper presents a new segmentation algorithm which combines active shape model and robust point matching techniques. It can use any simple feature detector to extract a large number of feature points in the image. Robust point matching is then used to search for the correspondences between feature and model points while the model is being deformed along the modes of variation of the active shape model. Although the algorithm is generic, it is particularly suited for medical imaging applications where prior knowledge is available. The value of the proposed method is examined with two different medical imaging modalities (Ultrasound, MRI) and in both 2D and 3D. The experiments have shown that the proposed algorithm is immune to missing feature points and noise. It has demonstrated significant improvements when compared to RPM-TPS and ASM alone.

Since the upturn of intravascular ultrasound (IVUS) as an imaging technique for the coronary artery system, much research has been done to simplify the complicated analysis of the resulting images. In this study, an attempt to develop an automatic tissue characterization algorithm for IVUS images was done. The first step was the extraction of texture features. The resulting feature space was used for classification, constructing a likelihood map to represent different coronary plaques. The information in this map was organized using a recently developed [1] geodesic snake formulation, the so-called Stop & Go snake. The novelty of our study lies in this last step, as it was the first time to apply the Stop & Go snake to segment IVUS images.

This paper presents a new algorithm for the semi-automatic segmentation of the prostate from B-mode trans-rectal ultrasound (TRUS) images. The segmentation algorithm first uses image warping to make the prostate shape elliptical. Measurement points along the prostate boundary, obtained from an edge-detector, are then used to find the best elliptical fit to the warped prostate. The final segmentation result is obtained by applying a reverse warping algorithm to the elliptical fit. This algorithm was validated using manual segmentation by an expert observer on 17 midgland, pre-operative, TRUS images. Distance-based metrics between the manual and semi-automatic contours showed a mean absolute difference of 0.67 ± 0.18mm, which is significantly lower than inter-observer variability. Area-based metrics showed an average sensitivity greater than 97% and average accuracy greater than 93%. The proposed algorithm was almost two times faster than manual segmentation and has potential for real-time applications.

Cardiac catheter ablation is a minimally invasive medical procedure to treat patients with heart rhythm disorders. It is useful to know the positions of the catheters and electrodes during the intervention, e.g. for the automatization of cardiac mapping. Our goal is therefore to develop a robust image analysis method that can detect the catheters in X-ray fluoroscopy images. Our method uses steerable tensor voting in combination with a catheter-specific multi-step extraction algorithm. The evaluation on clinical fluoroscopy images shows that especially the extraction of the catheter tip is successful and that the use of tensor voting accounts for a large increase in performance.

One critical issue in the context of image restoration is the problem of noise removal while keeping the integrity of relevant image information. Denoising is a crucial step to increase image conspicuity and to improve the performances of all the processings needed for quantitative imaging analysis. The method proposed in this paper is based on an optimized version of the Non Local (NL) Means algorithm. This approach uses the natural redundancy of information in image to remove the noise. Tests were carried out on synthetic datasets and on real 3T MR images. The results show that the NL-means approach outperforms other classical denoising methods, such as Anisotropic Diffusion Filter and Total Variation.

This paper presents an evaluation of the performance of a three-dimensional Active Shape Model (ASM) to segment the liver in 48 clinical CT scans. The employed shape model is built from 32 samples using an optimization approach based on the minimum description length (MDL). Three different gray-value appearance models (plain intensity, gradient and normalized gradient profiles) are created to guide the search. The employed segmentation techniques are ASM search with 10 and 30 modes of variation and a deformable model coupled to a shape model with 10 modes of variation. To assess the segmentation performance, the obtained results are compared to manual segmentations with four different measures (overlap, average distance, RMS distance and ratio of deviations larger 5mm). The only appearance model delivering usable results is the normalized gradient profile. The deformable model search achieves the best results, followed by the ASM search with 30 modes. Overall, statistical shape modeling delivers very promising results for a fully automated segmentation of the liver.

Although magnetic resonance imaging is considered to be non-invasive, there is at least one effect on the patient which has to be monitored: The heating which is generated by absorbed radio frequency (RF) power. It is described using the specific absorption rate (SAR). In order to obey legal limits for these SAR values, the scanner’s duty cycle has to be adjusted. The limiting factor depends on the patient’s position with respect to the scanner. Detection of this position allows a better adjustment of the RF power resulting in an improved scan performance and image quality. In this paper, we propose real-time methods for accurately detecting the patient’s position with respect to the scanner. MR data of thirteen test persons acquired using a new “move during scan” protocol which provides low resolution MR data during the initial movement of the patient bed into the scanner, is used to validate the detection algorithm. When being integrated, our results would enable automatic SAR optimization within the usual acquisition workflow at no extra cost.

In model-based segmentation, automated region identification is achieved via registration of novel data to a pre-determined model. The desired structure is typically generated via manual tracing within this model. When model-based segmentation is applied to human cortical data, problems arise if left-right comparisons are desired. The asymmetry of the human cortex requires that both left and right models of a structure be composed in order to effectively segment the desired structures. Paradoxically, defining a model in both hemi-spheres carries a likelihood of introducing bias to one of the structures. This paper describes a novel technique for creating a symmetric average model in which both hemispheres are equally represented and thus left-right comparison is possible. This work is an extension of that proposed by Guimond et al [1]. Hippocampal segmentation is used as a test-case in a cohort of 118 normal eld-erly subjects and results are compared with expert manual tracing.

Image diffusion can smooth away noise and small-scale structures while retaining important features, thereby enhancing the performances of many image processing algorithms such as image compression, segmentation and recognition. In this paper, we present a novel diffusion algorithm for which the filtering kernels vary according to the perceptual saliency of boundaries in the input images. The boundary saliency is estimated through a saliency measure which is generally determined by curvature changes, intensity gradient and the interaction of neighboring vectors. The connection between filtering kernels and perceptual saliency makes it possible to remove small-scale structures and preserves significant boundaries adaptively. The effectiveness of the proposed approach is validated by experiments on various medical images including the color Chinese Visible Human data set and gray MRI brain images.

Pulse inversion ultrasound is a mechanism for preferentially displaying contrast agent in blood vessels while suppressing signal from tissue. We seek a method for identifying and segmenting areas of the liver with similar statistically significant time intensity curves. As a first step in this process, a method of weighting Rayleigh distributed ultrasound image data before principal components analysis is presented. Simulation studies show that relative mean squared error can be reduced by 14% when the correct number of dimensions in selected. Our method is tested on an

in vitro

ultrasound phantom showing slightly increased error suppression, and is demonstrated on a clinical liver scan, showing decreased correlation between signals in the low intensity range.

In this paper, we treat the problem of reducing the false positives (FP) in the automatic detection of colorectal polyps at Computer Aided Detection in Computed Tomography Colonography (CAD-CTC) as a shape-filtering task. From the extracted candidate surface, we obtain a reliable shape distribution function and analyse it in the Fourier domain and use the resulting spectral data to classify the candidate surface as belonging to a polyp or a non-polyp class. The developed shape filtering scheme is computationally efficient (takes approximately 2 seconds per dataset to detect the polyps from the colonic surface) and offers robust polyp detection with an overall false positive rate of 5.44 per dataset at a sensitivity of 100% for polyps greater than 10mm when it was applied to standard and low dose CT data.

We examined the echotexture in ultrasonographic images of the wall of dominant ovulatory follicles in women during natural menstrual cycles and dominant anovulatory follicles which developed in women using oral contraceptives (OC). Ovarian follicles in women are fluid-filled structures in the ovary that contain oocytes (eggs). Dominant follicles are physiologically selected for preferential development and ovulation. Statistically significant differences between the two classes of follicles were observed for two co-occurrence matrix derived texture features and two edge-frequency based texture features which allowed accurate distinction of healthy and atretic follicles of similar diameters. Trend analysis revealed consistent turning points in time series of texture features between 3 and 4 days prior to ovulation coinciding with the time at which follicles are being biologically “prepared” for ovulation.

A new approach to fitting pharmacokinetic models to DCE-MRI data is described. The method relies on fitting individual concentration curves to a small set of basis functions and then making use of a look up table to relate the fitting coefficients to pre-calculated pharmacokinetic parameters. This is significantly faster than traditional non-linear fitting methods. Using simulated data and assuming a Tofts model, the accuracy of this direct approach is compared to the Levenberg-Marquardt algorithm. The effect of signal to noise ratio and the number of basis functions used on the accuracy is investigated. The basis fitting approach is slightly less accurate than the traditional non-linear least squares approach but the ten-fold improvement in speed makes the new technique useful as it can be used to generate pharmacokinetic maps in a clinically acceptable timeframe.

In this paper, we present a systematic investigation of the variability of the human cortical folding using a generic gyral model (GGM). The GGM consists of a fixed number of vertices that can be registered non-linearly to an individual anatomy so that for each individual we have a clearly defined set of landmarks that is spread across the cortex. This allows us to obtain a regionalized estimation of inter-subject variability. Since the GGM is stratified into different levels of depth, it also allows us to estimate variability as a function of depth. As another application of a polygonal line representation underlying the generic gyral model, we present a cortical parcellation scheme that can be used to regionalize cortical measurements.

The X-ray imaging equipment could be used to measure hemodynamic function in addition to visualizing the morphology. The parameters of specific interest are arterial blood flow and velocity. Current monoplane X-ray systems can perform 3D reconstruction of the arterial tree as well as to capture the propagation of the injected contrast agent on a sequence of 2D angiograms. We combine the 2D digital subtraction angiography sequence with the mechanically registered 3D volume of the vessel tree. From 3D vessel tree we extract each vessel and obtain its centerline and cross-section area. We get our velocity estimation from 2D sequence by comparing time-density signals measured at different ends of the projected vessel. From the average velocity and cross-section area we get the average blood flow estimate for each vessel. The algorithm described here is applied to datasets from real neuroradiological studies.

Quantitative analysis of computed tomographic (CT) images of the lungs is becoming increasingly useful in the medical and surgical management of subjects with Chronic Obstructive Pulmonary Disease (COPD). Current methods for the assessment of airway wall work well in idealized models of the airway. We propose a new method for airway wall detection based on phase congruency. This method does not rely on either a specific model of the airway or the point spread function of the scanner. Our results show that our method gives a better localization of the airway wall than ”full width at a half max” and is less sensitive to different reconstruction kernels and radiation doses.

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.

In large multi-center studies it is important to quantify data variations due to differences between sites and over time. Here we investigate inter-site variability in signal to noise ratio (SNR), percent integral uniformity (PIU), width and height using the American College of Radiology (ACR) phantom scans from the NIHPD project. Longitudinal variations are also analyzed. All measurements are fully automated. Our results show that the mean SNR, PIU and the 2 metric values were statistically different across sites. The maximum mean difference in diameter across sites was 2 mm (1.1%), and the maximum mean difference in height was 2.5 mm (1.7%). Over time, an average drift of 0.4 mm per year was observed for the diameter while a drift of 0.5 mm per year was observed for the height. Trends observed over time often depended not only on site, but also on modality and scanner manufacturer.

Evaluating registration algorithms is difficult due to the lack of gold standard in most clinical procedures. The

bronze standard

is a real-data based statistical method providing an alternative registration reference through a computationally intensive image database registration procedure. We propose in this paper an efficient implementation of this method through a grid-interfaced workflow enactor enabling the concurrent processing of hundreds of image registrations in a couple of hours only. The performances of two different grid infrastructures were compared. We computed the accuracy of 4 different rigid registration algorithms on longitudinal MRI images of brain tumors. Results showed an average subvoxel accuracy of 0.4 mm and 0.15 degrees in rotation.

Wireless endoscopy is a very recent and at the same time unique technique allowing to visualize and study the occurrence of contractions and to analyze the intestine motility. Feature extraction is essential for getting efficient patterns to detect contractions in wireless video endoscopy of small intestine. We propose a novel method based on anisotropic image filtering and efficient statistical classification of contraction features. In particular, we apply the image gradient tensor for mining informative skeletons from the original image and a sequence of descriptors for capturing the characteristic pattern of contractions. Features extracted from the endoluminal images were evaluated in terms of their discriminatory ability in correct classifying images as either belonging to contractions or not. Classification was performed by means of a support vector machine classifier with a radial basis function kernel. Our classification rates gave sensitivity of the order of 90.84% and specificity of the order of 94.43% respectively. These preliminary results highlight the high efficiency of the selected descriptors and support the feasibility of the proposed method in assisting the automatic detection and analysis of contractions.

A novel approach for generating a set of features derived from properties of patterns of curvature is introduced as a part of a computer aided colonic polyp detection system. The resulting sensitivity was 84% with 4.8 false positives per volume on an independent test set of 72 patients (56 polyps). When used in conjunction with other features, it allowed the detection system to reach an overall sensitivity of 94% with a false positive rate of 4.3 per volume.

A new method is introduce for the three-dimensional (3D) reconstruction of the coronary stents in-vivo utilizing two-dimensional projection images acquired during rotational angiography (RA). The method is based on the application of motion compensated techniques to the acquired angiograms resulting in a temporal snapshot of the stent within the cardiac cycle. For the first time results of 3D reconstructed coronary stents in vivo, with high spatial resolution are presented. The proposed method allows for a comprehensive and unique quantitative 3D assessment of stent expansion that rivals current x-ray and intravascular ultrasound techniques.

Laser photocoagulation is a proven procedure to treat various pathologies of the retina. Challenges such as motion compensation, correct energy dosage, and avoiding incidental damage are responsible for the still low success rate. They can be overcome with improved instrumentation, such as a fully automatic laser photocoagulation system.

In this paper, we present a core image processing element of such a system, namely a novel approach for retina mosaicing. Our method relies on recent developments in region detection and feature description to automatically fuse retina images. In contrast to the state-of-the-art the proposed approach works even for retina images with no discernable vascularity. Moreover, an efficient scheme to determine the blending masks of arbitrarily overlapping images for multi-band blending is presented.

In this paper, we present an original method that aims at parcellating the cortical surface in regions functionally meaningful, from individual anatomy. The parcellation is obtained using an anatomically constrained surface-based coordinate system from which we define a complete partition of the surface. The aim of our method is to exhibit a new way to describe the cortical surface organization, in both anatomical and functional terms. The method is described together with results applied to a functional somatotopy experiments.

This paper describes a novel approach to in-vivo measurement of brain surface folding in clinically acquired neonatal MR image data, which allows evaluation of surface curvature within subregions of the cortex. This paper addresses two aspects of this problem. Firstly: normalization of folding measures to provide area-independent evaluation of surface folding over arbitrary subregions of the cortex. Secondly: automated parcellation of the cortex at a particular developmental stage, based on an approximate spatial normalization of previously developed anatomical boundaries. The method was applied to seven premature infants (age 28-37 weeks) from which gray matter and gray-white matter interface surfaces were extracted. Experimental results show that previous folding measures are sensitive to the size of the surface of analysis, and that the area independent measures proposed here provide significant improvements. Such a system provides a tool to allow the study of structural development in the neonatal brain within specific functional subregions, which may be critical in identifying later neurological impairment.

This study presents a novel automatic approach for the identification of anatomical brain structures in magnetic resonance images (MRI). The method combines a fast multiscale multi-channel three dimensional (3D) segmentation algorithm providing a rich feature vocabulary together with a support vector machine (SVM) based classifier. The segmentation produces a full hierarchy of segments, expressed by an irregular pyramid with only linear time complexity. The pyramid provides a rich, adaptive representation of the image, enabling detection of various anatomical structures at different scales. A key aspect of the approach is the thorough set of multiscale measures employed throughout the segmentation process which are also provided at its end for clinical analysis. These features include in particular the prior probability knowledge of anatomic structures due to the use of an MRI probabilistic atlas. An SVM classifier is trained based on this set of features to identify the brain structures. We validated the approach using a gold standard real brain MRI data set. Comparison of the results with existing algorithms displays the promise of our approach.

Traditional techniques for statistical fMRI analysis are often based on thresholding of individual voxel values or averaging voxel values over a region of interest. In this paper we present a mixture-based response-surface technique for extracting and characterizing spatial clusters of activation patterns from fMRI data. Each mixture component models a local cluster of activated voxels with a parametric surface function. A novel aspect of our approach is the use of Bayesian nonparametric methods to automatically select the number of activation clusters in an image. We describe an MCMC sampling method to estimate both parameters for shape features and the number of local activations at the same time, and illustrate the application of the algorithm to a number of different fMRI brain images.

Diffusion tensor and functional MRI data provide insight into function and structure of the human brain. However, connectivity analysis between functional areas is still a challenge when using traditional fiber tracking techniques. For this reason, alternative approaches incorporating the entire tensor information have emerged. Based on previous research employing pathfinding for connectivity analysis, we present a novel search grid and an improved cost function which essentially contributes to more precise paths. Additionally, implementation aspects are considered making connectivity analysis very efficient which is crucial for surgery planning. In comparison to other algorithms, the presented technique is by far faster while providing connections of comparable quality. The clinical relevance is demonstrated by reconstructed connections between motor and sensory speech areas in patients with lesions located in between.

A new method for diffusion tensor MRI (DT-MRI) regularization is presented that relies on graph diffusion. We represent a DT image using a weighted graph, where the weights of edges are functions of the geodesic distances between tensors. Diffusion across this graph with time is captured by the heat-equation, and the solution, i.e. the heat kernel, is found by exponentiating the Laplacian eigen-system with time. Tensor regularization is accomplished by computing the Riemannian weighted mean using the heat kernel as its weights. The method can efficiently remove noise, while preserving the fine details of images. Experiments on synthetic and real-world datasets illustrate the effectiveness of the method.

We present a two-step process including white matter atlas generation and automatic segmentation. Our atlas generation method is based on population fiber clustering. We produce an atlas which contains high-dimensional descriptors of fiber bundles as well as anatomical label information. We use the atlas to automatically segment tractography in the white matter of novel subjects and we present quantitative results (FA measurements) in segmented white matter regions from a small population. We demonstrate reproducibility of these measurements across scans. In addition, we introduce the idea of using clustering for automatic matching of anatomical structures across hemispheres.

Individual white matter fibers cannot be resolved by current magnetic resonance (MR) technology. Many fibers of a fiber bundle will pass through an individual volume element (voxel). Individual visualized fiber tracts are thus the result of interpolation on a relatively coarse voxel grid, and an infinite number of them may be generated in a given volume by interpolation. This paper aims at creating a level set representation of a fiber bundle to describe this apparent continuum of fibers. It further introduces a coordinate system warped to the fiber bundle geometry, allowing for the definition of geometrically meaningful fiber bundle measures.

We present a method for automatically finding correspondence in Diffusion Tensor Imaging (DTI) from deformable registration to a common atlas. The registration jointly produces an average DTI atlas, which is unbiased with respect to the choice of a template image, along with diffeomorphic correspondence between each image. The registration image match metric uses a feature detector for thin fiber structures of white matter, and interpolation and averaging of diffusion tensors use the Riemannian symmetric space framework. The anatomically significant correspondence provides a basis for comparison of tensor features and fiber tract geometry in clinical studies and for building DTI population atlases.

A reoccurring theme in the diffusion tensor imaging literature is the per-voxel estimation of a symmetric 3 ×3 tensor describing the measured diffusion. In this work we attempt to generalize this approach by calculating 2 or 3 or up to

k

diffusion tensors for each voxel. We show that our procedure can more accurately describe the diffusion particularly when crossing fibers or fiber-bundles are present in the datasets.

Head motion is a significant source of error in fMRI activation detection and a common approach is to apply 3D volumetric rigid body motion correction techniques. However, in 2D multislice fMRI, each slice may have a distinct set of motion parameters due to inter-slice motion. Here, we apply an automated mutual information based slice-to-volume rigid body registration technique on time series data synthesized from a T

2

MRI brain dataset with simulated motion, functional activation, noise and geometric distortion. The map-slice-to-volume (MSV) technique was previously applied to patient data without ground truths for motion and activation regions. In this study, the activation images and area under the receiver operating characteristic curves for various time series datasets indicate that the MSV registration improves the activation detection capability when compared to results obtained from Statistical Parametric Mapping (SPM). The effect of temporal median filtering of motion parameters on activation detection performance was also investigated.

A new approach to group activation analysis in fMRI studies that test hypotheses focused on specific brain structures is presented and used to analyze hippocampal activation in a visual scene encoding study. The approach leverages the

cm-rep

method [10] to normalize hippocampal anatomy and project intra-subject hippocampal activation maps into a common reference space, eliminating normalization errors inherent in whole-brain approaches and guaranteeing that peaks detected in the random effects activation map are indeed associated with the hippocampus. When applied to real fMRI data, the method detects more significant hippocampal activation than the established whole-brain method.

Functional Magnetic Resonance imaging studies analyse sequences of brain volumes whose intensity changes predominantly reflect blood oxygenation level dependent (BOLD) effects. The most comprehensive signal model to date of the BOLD effect is formulated as a continuous-time system of nonlinear stochastic differential equations. In this paper we present a particle filtering method for the analysis of the BOLD system, and demonstrate it to be both accurate and robust in estimating the hidden physiological states including cerebral blood flow, cerebral blood volume, total deoxyhemoglobin content, and the flow inducing signal, from functional imaging data.

We present here a method that aims at producing representations of functional brain data on the cortical surface from functional MRI volumes. Such representations are required for subsequent cortical-based functional analysis. We propose a projection technique based on the definition, around each node of the grey/white matter interface mesh, of convolution kernels whose shape and distribution rely on the geometry of the local anatomy. For one anatomy, a set of convolution kernels is computed that can be used to project any functional data registered with this anatomy. The method is presented together with experiments on synthetic data and real statistical t-maps.

In this paper, we present algorithms to automatically detect and match landmark curves on cortical surfaces to get an optimized brain conformal parametrization. First, we propose an automatic landmark curve tracing method based on the principal directions of the local Weingarten matrix. Our algorithm obtains a hypothesized landmark curves using the Chan-Vese segmentation method, which solves a Partial Differential Equation (PDE) on a manifold with global conformal parameterization. Based on the global conformal parametrization of a cortical surface, our method adjusts the landmark curves iteratively on the spherical or rectangular parameter domain of the cortical surface along its principal direction field, using umbilic points of the surface as anchors. The landmark curves can then be mapped back onto the cortical surface. Experimental results show that the landmark curves detected by our algorithm closely resemble these manually labeled curves. Next, we applied these automatically labeled landmark curves to generate an optimized conformal parametrization of the cortical surface, in the sense that homologous features across subjects are caused to lie at the same parameter locations in a conformal grid. Experimental results show that our method can effectively help in automatically matching cortical surfaces across subjects.

We describe a new method to reconstruct human brain structures from 3D magnetic resonance brain images. Our method provides a fully automatic topology correction mechanism, thus avoiding tedious manual correction. Topological correctness is important because it is an essential prerequisite for brain atlas deformation and surface flattening. Our method uses an axis-aligned sweep through the volume to locate handles. Handles are detected by successively constructing and analyzing a directed graph. A multiple local region-growing process is used which simultaneously acts on the foreground and the background to isolate handles and tunnels. The sizes of handles and tunnels are measured, then handles are removed or tunnels filled based on their sizes. This process was used for 256 T1-weighted MR volumes.

This paper presents a method to improve the semi-automatic method for intensity inhomogeneity correction by Dawant et al. through introducing a fully automatic approach to reference points generation, which is based on order statistics and integrates information from the fine to coarse scale representations of the input image. The method has been validated and compared with two popular methods, N3 and BFC. Advantages of the proposed method are demonstrated.

Human brain atlases are indispensable tools in model-based segmentation and quantitative analysis of brain structures. However, adult brain atlases do not adequately represent the normal maturational patterns of the pediatric brain, and the use of an adult model in pediatric studies may introduce substantial bias. Therefore, we proposed to develop a digital atlas of the pediatric human brain in this study. The atlas was constructed from T1-weighted MR data set of a 9-year old, right-handed girl. Furthermore, we extracted and simplified boundary surfaces of 25 manually defined brain structures (cortical and subcortical) based on surface curvature. We constructed a 3D triangular mesh model for each structure by triangulation of the structure’s reference points. Kappa statistics (cortical, 0.97; subcortical, 0.91) indicated substantial similarities between the mesh-defined and the original volumes. Our brain atlas and structural mesh models (www.stjude.org/brainatlas) can be used to plan treatment, to conduct knowledge and model-driven segmentation, and to analyze the shapes of brain structures in pediatric patients.

In this work, we proposed a discriminative model of Alzheimer’s disease (AD) on the basis of multivariate pattern classification and functional magnetic resonance imaging (fMRI). This model used the correlation/anti-correlation coefficients of two intrinsically anti-correlated networks in resting brains, which have been suggested by two recent studies, as the feature of classification. Pseudo-Fisher Linear Discriminative Analysis (pFLDA) was then performed on the feature space and a linear classifier was generated. Using leave-one-out (LOO) cross validation, our results showed a correct classification rate of 83%. We also compared the proposed model with another one based on the whole brain functional connectivity. Our proposed model outperformed the other one significantly, and this implied that the two intrinsically anti-correlated networks may be a more susceptible part of the whole brain network in the early stage of AD.

In order to achieve diagnostically useful CT (computed tomography) images of the moving heart, the standard image reconstruction has to be modified to a phase-correlated reconstruction, which considers the motion phase of the heart and generates a quasi-static image in one defined motion phase. For that purpose a synchronization signal is needed, typically a concurrent ECG recording. Commonly, the reconstruction phase is adapted by the user to the patient-specific heart motion to improve the image quality and thus the diagnostic value. The purpose of our work is to automatically identify the optimal reconstruction phase for cardiac CT imaging with respect to motion artifacts. We provide a solution for a patient- and heart rate-independent detection of the optimal phase in the cardiac cycle which shows a minimum of cardiac movement. We validated our method by the correlation with the reconstruction phase selected visually on the basis of ECG-triggering and used for the medical diagnosis. The mean difference between both reconstruction phases was 12.5 % with respect to a whole cardiac motion cycle indicating a high correlation. Additionally, reconstructed cardiac images are shown which confirm the results expressed by the correlation measurement and in some cases even indicating an improvement using the proposed method.

Freehand 3D ultrasound can be acquired without a position sensor by finding the separations of pairs of frames using information in the images themselves. Previous work has not considered how to reconstruct entirely freehand data, which can exhibit irregularly spaced frames, non-monotonic out-of-plane probe motion and significant in-plane motion. This paper presents reconstruction methods that overcome these limitations and are able to robustly reconstruct freehand data. The methods are assessed on freehand data sets and compared to reconstructions obtained using a position sensor.

MR imaging of the heart valve leaflets is a challenging problem due to their large movements throughout the cardiac cycle. This paper presents a motion-compensated imaging approach with COMB tagging for valve imaging. It involves an automatic method for tracking the full 3D motion of the valve plane so as to provide a motion-tracked acquisition scheme. Super-resolution enhancement is then applied to the slice-select direction so that the partial volume effect is minimised.

In vivo

results have shown that in terms of slice positioning, the method has equivalent accuracy to that of a manual approach whilst being quicker and more consistent. The use of multiple parallel COMB tags will permit adaptive imaging that follows tissue motion. This will have significant implications for quantification of myocardial perfusion and tracking anatomy, functions that are traditionally difficult in MRI.

This paper describes a rapid method for intraoperative recovery of liver motion and deformation due to respiration by using a laparoscopic freehand 3D ultrasound (US) system. Using the proposed method, 3D US images of the liver can be extended to 4D US images by acquiring additional several sequences of 2D US images during a couple of respiration cycles. Time-varying 2D US images are acquired on several sagittal image planes and their 3D positions and orientations are measured using a laparoscopic ultrasound probe to which a miniature magnetic 3D position sensor is attached. During the acquisition, the probe is assumed to move together with the liver surface. In-plane 2D deformation fields and respiratory phase are estimated from the time-varying 2D US images, and then the time-varying 3D deformation fields on the sagittal image planes are obtained by combining 3D positions and orientations of the image planes. The time-varying 3D deformation field of the volume is obtained by interpolating the 3D deformation fields estimated on several planes. The proposed method was evaluated by

in vivo

experiments using a pig liver.

We present a model for the numerical simulation of radio frequency (RF) ablation of tumors with mono- or bipolar probes. This model includes the electrostatic equation and a variant of the well-known bio-heat transfer equation for the distribution of the electric potential and the induced heat. The equations are nonlinearly coupled by material parameters that change with temperature, dehydration and damage of the tissue. A fixed point iteration scheme for the nonlinear model and the spatial discretization with finite elements are presented. Moreover, we incorporate the effect of evaporation of water from the cells at high temperatures using a predictor-corrector like approach. The comparison of the approach to a real ablation concludes the paper.

Digital brain atlases can be used in conjuction with magnetic resonance imaging (MRI) and computed tomography (CT) for planning and guidance during neurosurgery. Digital atlases are advantageous since they can be warped nonlinearly to fit each patient’s unique anatomy. Functional neurosurgery with implantation of deep brain stimulating (DBS) electrodes requires accurate targeting, and has become a popular surgical technique in Parkinsonian patients. In this paper, we present a method for integrating postoperative data from subthalamic (STN) DBS implantation into an antomical atlas of the basal ganglia and thalamus. The method estimates electrode position from post-operative magnetic resonance imaging (MRI) data. These electrodes are then warped back into the atlas space and are modelled in three dimensions. The average of these models is then taken to show the region where the majority of STN DBS electrodes were implanted. The group with more favorable post-operative results was separated from the group which responded to the STN DBS implantation procedure less favourably to create a probablisitic distribution of DBS in the STN electrodes.

We present a novel method to visualize registration uncertainty and a simple study to motivate the use of uncertainty visualization in computer–assisted surgery. Our visualization method resulted in a statistically significant reduction in the number of attempts required to localize a target, and a statistically significant reduction in the number of targets that our subjects failed to localize. Most notably, our work addresses the existence of uncertainty in guidance and offers a first step towards helping surgeons make informed decisions in the presence of imperfect data.

A rapid approach to monitor ablative therapy through optimizing shape and elasticity parameters is introduced. Our motivating clinical application is targeting and intraoperative monitoring of hepatic tumor thermal ablation, but the method translates to the generic problem of encapsulated stiff masses (solid organs, tumors, ablated lesions, etc.) in ultrasound imaging. The approach involves the integration of the following components: a biomechanical computational model of the tissue, a correlation approach to estimate/track tissue deformation, and an optimization method to solve the inverse problem and recover the shape parameters in the volume of interest. Successful conver-gence and reliability studies were conducted on simulated data. Then ex-vivo studies were performed on 18 ex-vivo bovine liver samples previously ablated under ultrasound monitoring in controlled laboratory environment. While B-mode ultrasound does not clearly identify the development of necrotic lesions, the proposed technique can potentially segment the ablation zone. The same framework can also yield both partial and full elasticity reconstruction.

This paper develops a volumetric quantification approach of the airway wall in multi-detector computed tomography (MDCT), exploiting a 3D segmentation methodology based on patient-specific deformable mesh model. A comparative study carried out with respect to a reference 2D/3D surface quantification technique underlines the clinical interest of the proposed approach in assessing airway remodeling in asthmatics and in evaluating the efficiency of therapeutic protocols.

In this paper we propose a new diagnostic feature for Alzheimer’s Disease (AD) which is based on assessment of the degree of inter-hemispheric asymmetry using Single Photon Emission Computed Tomography (SPECT). The asymmetry measure used represents differences in 3D perfusion image patterns in the cerebral hemispheres. We start from the simplest descriptors of brain perfusion such as the mean intensity within pairs of brain lobes, gradually increasing the resolution up to five-dimensional co-occurrence matrices. Evaluation of the method was performed using SPECT scans of 79 subjects including 42 patients with clinical diagnosis of AD and 37 controls. It was found that combination of intensity and gradient features in co-occurrence matrices captures significant differences in asymmetry values between AD and normal controls (

p

<0.00003 for all cerebral lobes). Our results suggest that the asymmetry feature is useful for discriminating AD patients from normal controls as detected by SPECT.

Deep brain stimulation (DBS) is an established therapy for the treatment of movement disorders, and has shown promising results for the treatment of a wide range of other neurological disorders. However, little is known about the mechanism of action of DBS or the volume of brain tissue affected by stimulation. We have developed methods that use anatomical and diffusion tensor MRI (DTI) data to predict the volume of tissue activated (VTA) during DBS. We co-register the imaging data with detailed finite element models of the brain and stimulating electrode to enable anatomically and electrically accurate predictions of the spread of stimulation. One critical component of the model is the DTI tensor field that is used to represent the 3-dimensionally anisotropic and inhomogeneous tissue conductivity. With this system we are able to fuse structural and functional information to study a relevant clinical problem: DBS of the subthalamic nucleus for the treatment of Parkinson’s disease (PD). Our results show that inclusion of the tensor field in our model caused significant differences in the size and shape of the VTA when compared to a homogeneous, isotropic tissue volume. The magnitude of these differences was proportional to the stimulation voltage. Our model predictions are validated by comparing spread of predicted activation to observed effects of oculomotor nerve stimulation in a PD patient. In turn, the 3D tissue electrical properties of the brain play an important role in regulating the spread of neural activation generated by DBS.

Haemodynamics, and in particular wall shear stress, is thought to play a critical role in the progression and rupture of intracranial aneurysms. A novel method is presented that combines image-based wall motion estimation obtained through non-rigid registration with computational fluid dynamics (CFD) simulations in order to provide realistic intra-aneurysmal flow patterns and understand the effects of deforming walls on the haemodynamic patterns. In contrast to previous approaches, which assume rigid walls or

ad hoc

elastic parameters to perform the CFD simulations, wall compliance has been included in this study through the imposition of measured wall motions. This circumvents the difficulties in estimating personalized elasticity properties. Although variations in the aneurysmal haemodynamics were observed when incorporating the wall motion, the overall characteristics of the wall shear stress distribution do not seem to change considerably. Further experiments with more cases will be required to establish the clinical significance of the observed variations.

Acute rejection is the most common reason of graft failure after kidney transplantation, and early detection is crucial to survive the transplanted kidney function. In this paper, we introduce a new approach for the automatic classification of normal and acute rejection transplants from Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI). The proposed algorithm consists of three main steps; the first step isolates the kidney from the surrounding anatomical structures by evolving a deformable model based on two density functions; the first function describes the distribution of the gray level inside and outside the kidney region and the second function describes the prior shape of the kidney. In the second step, a new nonrigid registration approach is employed to account for the motion of the kidney due to patient breathing. To validate our registration approach, we use a simulation of deformations based on biomechanical modelling of the kidney tissue using the finite element method (F.E.M.). Finally, the perfusion curves that show the transportation of the contrast agent into the tissue are obtained from the cortex and used in the classification of normal and acute rejection transplants. Applications of the proposed approach yield promising results that would, in the near future, replace the use of current technologies such as nuclear imaging and ultrasonography, which are not specific enough to determine the type of kidney dysfunction.

Computer-aided detection (CAD) has become increasingly common in recent years as a tool in catching breast cancer in its early, more treatable stages. More and more breast centers are using CAD as studies continue to demonstrate its effectiveness. As the technology behind CAD improves, so do its results and its impact on society. In trying to improve the sensitivity and specificity of CAD algorithms, a good deal of work has been done on feature extraction, the generation of mathematical representations of mammographic features which can help distinguish true cancerous lesions from false ones. One feature that is not currently seen in the literature that physicians rely on in making their decisions is location within the breast. This is a difficult feature to calculate as it requires a good deal of prior knowledge as well as some way of accounting for the tremendous variability present in breast shapes. In this paper, we present a method for the generation and implementation of a probabilistic breast cancer atlas. We then validate this method on data from the Digital Database for Screening Mammography (DDSM).

In this paper, we present a learning-based method for the detection and segmentation of 3D free-form tubular structures, such as the rectal tubes in CT colonoscopy. This method can be used to reduce the false alarms introduced by rectal tubes in current polyp detection algorithms. The method is hierarchical, detecting parts of the tube in increasing order of complexity, from tube cross sections and tube segments to the whole flexible tube. To increase the speed of the algorithm, candidate parts are generated using a voting strategy. The detected tube segments are combined into a flexible tube using a dynamic programming algorithm. Testing the algorithm on 210 unseen datasets resulted in a tube detection rate of 94.7% and 0.12 false alarms per volume. The method can be easily retrained to detect and segment other tubular 3D structures.

Over the past years many computer aided diagnosis (CAD) schemes have been presented for the detection of colonic polyps in CT Colonography. The vast majority of these methods (implicitly) model polyps as approximately spherical protrusions. Polyp shape and size varies greatly, however and is often far from spherical. We propose a shape and size invariant method to detect suspicious regions. The method works by locally deforming the colon surface until the second principal curvature is smaller than or equal to zero. The amount of deformation is a quantitative measure of the ’protrudeness’. The deformation field allows for the computation of various additional features to be used in supervised pattern recognition. It is shown how only a few features are needed to achieve 95% sensitivity at 10 false positives (FP) per dataset for polyps larger than 6 mm.

This paper presents a model-based technique for lesion detection in colon CT scans that uses analytical shape models to map the local shape curvature at individual voxels to anatomical labels. Local intensity profiles and curvature information have been previously used for discriminating between simple geometric shapes such as spherical and cylindrical structures. This paper introduces novel analytical shape models for colon-specific anatomy, viz. folds and polyps, built by combining parts with simpler geometric shapes. The models better approximate the actual shapes of relevant anatomical structures while allowing the application of model-based analysis on the simpler model parts. All parameters are derived from the analytical models, resulting in a simple voxel labeling scheme for classifying individual voxels in a CT volume. The algorithm’s performance is evaluated against expert-determined ground truth on a database of 42 scans and performance is quantified by free-response receiver-operator curves.

Lung cancer remains an ongoing problem resulting in substantial deaths in the United States and the world. Within the United states, cancer of the lung and bronchus are the leading causes of fatal malignancy and make up 32% of the cancer deaths among men and 25% of the cancer deaths among women. Five year survival is low, (14%), but recent studies are beginning to provide some hope that we can increase survivability of lung cancer provided that the cancer is caught and treated in early stages. These results motivate revisiting the concept of lung cancer screening using thin slice multidetector computed tomography (MDCT) protocols and automated detection algorithms to facilitate early detection. In this environment, resources to aid Computer Aided Detection (CAD) researchers to rapidly develop and harden detection and diagnostic algorithms may have a significant impact on world health. The National Cancer Institute (NCI) formed the Lung Imaging Database Consortium (LIDC) to establish a resource for detecting, sizing, and characterizing lung nodules. This resource consists of multiple CT chest exams containing lung nodules that seveal radiologists manually countoured and characterized. Consensus on the location of the nodule boundaries, or even on the existence of a nodule at a particular location in the lung was not enforced, and each contour is considered a possible nodule. The researcher is encouraged to develop measures of ground truth to reconcile the multiple radiologist marks. This paper analyzes these marks to determine radiologist agreement and to apply statistical tools to the generation of a nodule ground truth. Features of the resulting consensus and individual markings are analyzed.

Acromegaly is a rare disorder which affects about 50 of every million people. The disease typically causes swelling of the hands, feet, and face, and eventually permanent changes to areas such as the jaw, brow ridge, and cheek bones. The disease is often missed by physicians and progresses beyond where it might if it were identified and treated earlier. We consider a semi-automated approach to detecting acromegaly, using a novel combination of support vector machines (SVMs) and a morphable model. Our training set consists of 24 frontal photographs of acromegalic patients and 25 of disease-free subjects. We modelled each subject’s face in an analysis-by-synthesis loop using the three-dimensional morphable face model of Blanz and Vetter. The model parameters capture many features of the 3D shape of the subject’s head from just a single photograph, and are used

directly

for classification. We report encouraging results of a classifier built from the training set of real human subjects.

Current diagnosis of prostatic adenocarcinoma is done by manual analysis of biopsy tissue samples for tumor presence. However, the recent advent of whole slide digital scanners has made histopathological tissue specimens amenable to computer-aided diagnosis (CAD). In this paper, we present a CAD system to assist pathologists by automatically detecting prostate cancer from digitized images of prostate histological specimens. Automated diagnosis on very large high resolution images is done via a multi-resolution scheme similar to the manner in which a pathologist isolates regions of interest on a glass slide. Nearly 600 image texture features are extracted and used to perform pixel-wise Bayesian classification at each image scale to obtain corresponding likelihood scenes. Starting at the lowest scale, we apply the AdaBoost algorithm to combine the most discriminating features, and we analyze only pixels with a high combined probability of malignancy at subsequent higher scales. The system was evaluated on 22 studies by comparing the CAD result to a pathologist’s manual segmentation of cancer (which served as ground truth) and found to have an overall accuracy of 88%. Our results show that (1) CAD detection sensitivity remains consistently high across image scales while CAD specificity increases with higher scales, (2) the method is robust to choice of training samples, and (3) the multi-scale cascaded approach results in significant savings in computational time.

Subject-specific physiological motion modeling combined with low-dimensional real-time sensing can provide effective prediction of acyclic tissue deformation particularly due to respiration. However, real-time sensing signals used for predictive motion modeling can be strongly coupled with each other but poorly correlated with respiratory induced cardiac deformation. This paper explores a systematic framework based on sequential feature selection for optimal sensor placement so as to achieve maximal model sensitivity and prediction accuracy in response to the entire range of tissue deformation. The proposed framework effectively resolves the problem encountered by traditional regression methods in that the latent variables from both the input and output of the regression model are used to establish their inner relationships. Detailed numerical analysis and

in vivo

results are provided, which demonstrate the potential clinical value of the technique.

Registration of 3D segmented cardiac images with tracked electrophysiological data has been previously investigated for use in cardiac mapping and navigation systems. However, dynamic cardiac 4D (3D + time) registration methods do not presently exist. This paper introduces two new 4D registration methods based on the popular iterative closest point (ICP) algorithm that may be applied to dynamic 3D shapes. The first method averages the transformations of the 3D ICP on each phase of the dynamic data, while the second finds the closest point pairs for the data in each phase and performs a least squares fit between all the pairs combined. Experimental results show these methods yield more accurate transformations compared to using a traditional 3D approach (4D errors: Translation 0.4mm, Rotation 0.45° vs. 3D errors: Translation 1.2mm, Rotation 1.3°) while also increasing capture range and success rate.

A novel strategy is presented to recover cardiac electrical excitation pattern from tomographic medical image sequences. The geometrical/physical representation of the heart and the dense motion field of the myocardium are first derived from imaging data through segmentation and motion recovery. The myocardial active forces are then calculated through the law of force equilibrium from the motion field, realized with a stochastic multiframe algorithm. Since tissue active forces are physiologically driven by electrical excitations, we can readily relate the pattern of active forces to the pattern of electrical propagation in myocardium, where spatial regularization is enforced. Experiments are conducted on three-dimensional synthetic data and canine magnetic resonance image sequence with favorable results.

This paper reports on pilot laboratory experiments with a recently proposed surgical procedure for percutaneous screw insertion into fractured scaphoid bones using ultrasound guidance. The experiments were intended to determine the operator variability of the procedure and its performance in comparison with a traditional pinning procedure using fluoroscopy. In the proposed procedure, a three-dimensional surface model is created from pre-operative computed tomography images and intra-operatively registered to the patient using ultrasound images. A graphical interface that communicates with an optical camera tracking the surgical tools, guides the surgeon during the procedure in real time. The results of the experiments revealed non-significant differences between operators for the error in the entry location of the drill hole (p=0.90); however, significant differences for the exit location (p<0.05). Comparison with the traditional pinning procedure shows that the outcome of the recently proposed procedure appears to be more consistent.

Tissue engineering integrates the principles of engineering and life sciences toward the design, construction, modification and growth of biological substitutes that restore, maintain, or improve tissue function. The structural integrity and ultimate functionality of such tissue analogs is defined by scaffolds- porous, three-dimensional

"trellis-like"

structures that, on implantation, provide a viable environment to regenerate damaged tissues. The orthogonal scaffold fabrication methods currently employed can be broadly classified into two categories: (a) conventional, irreproducible, stochastic techniques producing reasonably biomorphic scaffold architecture, and (b) rapidly emerging, repeatable, computer-controlled techniques producing straight edged "

contra naturam"

scaffold architecture. In this paper, we present the results of the first attempt in an image-based scaffold modeling and optimization strategy that synergistically exploits the orthogonal fabrication techniques to create repeatable, biomorphic scaffolds with optimal scaffold morphology. Motivated by the use of Gaussian random fields (GRF) to model cosmological structure formation, we use appropriately ordered and clipped stacks of GRF to model the three-dimensional pore-solid scaffold labyrinths. Image-based metrology, fabrication and mechanical characterization of these scaffolds reveal the possibility of enabling the previously elusive deployment of promising benchside tissue analogs to the clinical bedside.

This paper explores the use of deformable mesh for registration of microscopic iris image sequences. The registration, as an effort for stabilizing and rectifying images corrupted by motion artifacts, is a crucial step toward leukocyte tracking and motion characterization for the study of immune systems. The image sequences are characterized by locally nonlinear deformations, where an accurate analytical expression can not be derived through modeling of image formation. We generalize the existing deformable mesh and formulate it in a probabilistic framework, which allows us to conveniently introduce local image similarity measures, to model image dynamics and to maintain a well-defined mesh structure and smooth deformation through appropriate regularization. Experimental results demonstrate the effectiveness and accuracy of the algorithm.

No current non-invasive method is capable of assessing the efficacy of brain tumor therapy early during treatment. We outline an approach that evaluates tumor activity via statistical analysis of vessel shape using vessels segmented from MRA. This report is the first to describe the changes in vessel shape that occur during treatment of metastatic brain tumors as assessed by sequential MRA. In this preliminary study of 16 patients undergoing treatment for metastatic breast cancer we conclude that vessel shape may predict tumor response several months in advance of traditional methods.

In the present paper, we propose a new pulse measurement methodology based on thermal imaging (contact-free). The method capitalizes both on the thermal undulation produced by the traveling pulse as well as the periodic expansion of the compliant vessel wall. The paper reports experiments on 34 subjects, where it compares the performance of the new pulse measurement method to the one we reported previously. The measurements were ground-truthed through a piezo-electric sensor. Statistical analysis reveals that the new imaging methodology is more accurate and robust than the previous one. Its performance becomes nearly perfect, when the vessel is not obstructed by a thick fat deposit.

Recent research in biology has indicated correlations between the movement patterns of functional sites (such as replication sites in DNA) and zones of genetic activity within a nucleus. A detailed study and analysis of the motion dynamics of these sites can reveal an interesting insight into their role in DNA replication and function. In this paper, we propose a suite of novel techniques to determine, analyze, and interpret the mobility patterns of functional sites. Our algorithms are based on interesting ideas from theoretical computer science and database theory and provide for the first time the tools to interpret the seemingly stochastic motion patterns of the functional sites within the nucleus in terms of a set of tractable ‘patterns’ which can then be analyzed to understand their biological significance.

Multidimensional fluorescence imaging is a powerful molecular imaging modality that is emerging as an important tool in the study of biological tissues. Due to the large volume of multi-spectral data associated with the technique, it is often difficult to find the best combination of parameters to maximize the contrast between different tissue types. This paper presents a novel framework for the characterization of tissue compositions based on the use of time resolved fluorescence imaging without the explicit modeling of the decays. The composition is characterized through soft clustering based on manifold embedding for reducing the dimensionality of the datasets and obtaining a consistent differentiation scheme for determining intrinsic constituents of the tissue. The proposed technique has the benefit of being fully automatic, which could have significant advantages for automated histopathology and increasing the speed of intraoperative decisions. Validation of the technique is carried out with both phantom data and tissue samples of the human pancreas.

Diffusion weighted magnetic resonance (DWMR or DW) imaging is a fast evolving technique to investigate the connectivity of brain white matter by measuring the self-diffusion of the water molecules in the tissue. Registration is a key step in group analysis of the DW images that may lead to understanding of functional and structural variability of the normal brain, understanding disease process, and improving neurosurgical planning. In this paper, we present a new method for registering DW images. The method works directly on the diffusion weighted images without using tensor reconstruction, fiber tracking, and fiber clustering. Therefore, the performance of the method does not rely on the accuracy and robustness of these steps. Moreover, since all the information in the original diffusion weighted images is used for registration, the results of the method is robust to imaging noise. We demonstrate the method on intra-subject registration with an affine transform using DW images acquired on the same scanner with the same imaging protocol. Extension to deformable registration for images acquired on different scanners and/or with different imaging protocols is also discussed.

Rigid registration of pre-operative surgical plans to intra-operative coordinates of a patient is an important step in computer-assisted orthopaedic surgery. A good measure for registration accuracy is the target registration error (TRE) which is the distance after registration between a pair of corresponding points not used in the registration process. However, TRE is not a deterministic value, since there is always error in the localized features (points) utilized in the registration. In this situation, the distribution of TRE carries more information than TRE by itself. Previously, the distribution of TRE has been estimated with the accuracy of the first-order approximation. In this paper, we analytically approximate the TRE distribution up to at least the second-order accuracy based on the Unscented Kalman Filter algorithm.

A new method for elastic mono-modal image registration for adaptive fractionated radiotherapy is presented. Elastic registration is a prerequisite for many medical applications in diagnosis, therapy planning, and therapy. Especially for adaptive radiotherapy efficient and accurate registration is required. Therefore, we developed a fast block matching algorithm for robust image registration. Anatomical landmarks are automatically selected at tissue borders and relocated in the frequency domain. A smooth interpolation is calculated by modified thin-plate splines with local impact. The concept of the algorithm allows different handling of different image structures. Thus, more features were included, like handling of discontinuities (e. g. air cavities in the intestinal track or rectum, observable in only one image), which can not be registered in a conventional way. The planning CT as well as delineated structures of target volume and organs at risks are transformed according to deviations observed in daily acquired verification CTs prior each dose fraction. This way, the time consuming repeated delineation, a prerequisite for adaptive radiotherapy, is avoided. The total calculation time is below 5 minutes and the accurateness is higher than voxel precision, which allows to use this tool in the clinical workflow. We present results of prostate, head-and-neck, and paraspinal tumors with verification by manually selected landmarks. We think this registration technique is not only suitable for adaptive radiotherapy, but also for other applications which require fast registration and possibilities to process special structures (e. g. discontinuities) in a different way.

This paper presents a deformable registration method to co-register histological images with MR images of the same prostate. By considering various distortion and cutting artifacts in histological images and also fundamentally different nature of histological and MR images, our registration method is thus guided by two types of landmark points that can be reliably detected in both histological and MR images, i.e., prostate boundary points, and internal salient points that can be identified by a scale-space analysis method. The similarity between these automatically detected landmarks in histological and MR images are defined by geometric features and normalized mutual information, respectively. By optimizing a function, which integrates the similarities between landmarks with spatial constraints, the correspondences between the landmarks as well as the deformable transformation between histological and MR images can be simultaneously obtained. The performance of our proposed registration algorithm has been evaluated by various designed experiments. This work is part of a larger effort to develop statistical atlases of prostate cancer using both imaging and histological information, and to use these atlases for optimal biopsy and therapy planning.

We present a new algorithm for affine registration of diffusion tensor magnetic resonance (DT-MR) images. The method is based on a new formulation of a point-wise tensor similarity measure, which weights directional and magnitude information differently depending on the type of diffusion. The method is compared to a reference method, which uses normalized mutual information (NMI), calculated either from a fractional anisotropy (FA) map or a

T

2

-weighted MR image. The registration methods are applied to real and simulated DT-MR images. Visual assessment is done for real data and for simulated data, registration accuracy is defined. The results show that the proposed method outperforms the reference method.

We propose and test analytic equations for approximating expected fiducial and surface target registration error (TRE). The equations are derived from a spatial stiffness model of registration. The fiducial TRE equation is equivalent to one presented by [1]. We believe that the surface TRE equation is novel, and we provide evidence from computer simulations to support the accuracy of the approximation.

This paper presents a method for tracking a bronchoscope based on motion prediction and image registration from multiple initial starting points as a function of a bronchoscope navigation system. We try to improve performance of bronchoscope tracking based on image registration using multiple initial guesses estimated using motion prediction. This method basically tracks a bronchoscopic camera by image registration between real bronchoscopic images and virtual ones derived from CT images taken prior to the bronchoscopic examinations. As an initial guess for image registration, we use multiple starting points to avoid falling into local minima. These initial guesses are computed using the motion prediction results obtained from the Kalman filter’s output. We applied the proposed method to nine pairs of X-ray CT images and real bronchoscopic video images. The experimental results showed significant performance in continuous tracking without using any positional sensors.

Measurements of cup alignment after total hip replacement (THR) surgery are typically performed on postoperative pelvic radiographs. Radiographic measurement of cup orientation depends on the position and orientation of the pelvis on the X-ray table, and its variability could introduce significant measurement errors. We have developed a tool to accurately measure 3D implant orientation from postoperative antero-posterior radiographs by registering to preoperative CT scans. The purpose of this study is to experimentally and clinically validate the automatic CT/X-ray matching algorithm by comparing the X-ray based measurements of cup orientation with direct 3D measurements from postoperative CT scans. The mean measurement errors (± stdev) found in this study were 0.4°±0.8° for abduction and 0.6°±0.8° for version. In addition, radiographic pelvic orientation measurements demonstrated a wide range of inter-subject variability, with pelvic flexion ranging from –5.9° to 11.2°.

We have developed a novel model-to-image registration technique which aligns a 3-dimensional model of vasculature with two semi-orthogonal fluoroscopic projections. Our vascular registration method is used to intra-operatively initialize the alignment of a catheter and a pre-operative vascular model in the context of image-guided TIPS (Transjugular, Intrahepatic, Portosystemic Shunt formation) surgery. Registration optimization is driven by the intensity information from the projection pairs at sample points along the centerlines of the model. Our algorithm shows speed, accuracy and consistency given clinical data.

Image registration is a fundamental problem in medical imaging. It is especially challenging in lung images compared, for example, with the brain. The challenges include large anatomical variations of human lung and a lack of fixed landmarks inside the lung. This paper presents a new method for lung HRCT image registration. It employs a landmark-based global transformation and a novel ray-tracing-based lung surface registration. The proposed surface registration method has two desirable properties: 1) it is fully reversible, and 2) it ensures that the registered lung will be inside the target lung. We evaluated the registration performance by applying it to lung regions mapping. Tested on 46 scans, the registered regions were 89% accurate compared with the ground-truth.

We introduce an elastic registration approach which is based on a physical deformation model and uses Gaussian elastic body splines (GEBS). We formulate an extended energy functional related to the Navier equation under Gaussian forces which also includes landmark localization uncertainties. These uncertainties are characterized by weight matrices representing anisotropic errors. Since the approach is based on a physical deformation model, cross-effects in elastic deformations can be taken into account. Moreover, we have a free parameter to control the locality of the transformation for improved registration of local geometric image differences. We demonstrate the applicability of our scheme based on 3D CT images from the Truth Cube experiment, 2D MR images of the brain, as well as 2D gel electrophoresis images. It turns out that the new scheme achieves more accurate results compared to previous approaches.

In this paper we propose a method for the nonrigid registration of contrast-enhanced dynamic sequences of magnetic resonance(MR) images. The algorithm has been developed with accuracy in mind, but also has a clinically viable execution time (i.e. a few minutes) as a goal. The algorithm is driven by multiresolution optical flow with the brightness consistency assumption relaxed, subject to a regularized best-fit within a family of transforms. The particular family of transforms we have employed uses a grid of control points and trilinear interpolation. We present validation results from a study simulating non-rigid deformation by a biomechanical model of the breast, with simulated uptake of a contrast agent. We further present results from applying the algorithm as part of a routine breast cancer screening protocol.

With the growing acceptance of nonrigid registration as a useful tool to perform clinical research, and in particular group studies, the storage space needed to hold the resulting transforms is deemed to become a concern for vector field based approaches, on top of the traditional computation time issue. In a recent study we lead, which involved the registration of more than 22,000 pairs of T

1

MR volumes, this constrain appeared critical indeed. In this paper, we propose to decompose the vector field on a wavelet basis, and let the registration algorithm minimize the number of non-zero coefficients by introducing an

L

1

penalty. This enables a sparse representation of the vector field which, unlike parametric representations, does not confine the estimated transform into a small parametric space with a fixed uniform smoothness : nonzero wavelet coefficients are optimally distributed depending on the data. Furthermore, we show that the iconic feature registration framework allows to embed the non-differentiable

L

1

penalty into a

C

1

energy that can be efficiently minimized by standard optimization techniques.

In this paper we propose a diffeomorphic non-rigid registration algorithm based on free-form deformations (FFDs) which are modelled by B-splines. In contrast to existing non-rigid registration methods based on FFDs the proposed diffeomorphic non-rigid registration algorithm based on free-form deformations (FFDs) which are modelled by B-splines. To construct a diffeomorphic transformation we compose a sequence of free-form deformations while ensuring that individual FFDs are one-to-one transformations. We have evaluated the algorithm on 20 normal brain MR images which have been manually segmented into 67 anatomical structures. Using the agreement between manual segmentation and segmentation propagation as a measure of registration quality we have compared the algorithm to an existing FFD registration algorithm and a modified FFD registration algorithm which penalises non-diffeomorphic transformations. The results show that the proposed algorithm generates diffeomorphic transformations while providing similar levels of performance as the existing FFD registration algorithm in terms of registration accuracy.

Nonlinear image registration is a prerequisite for a variety of medical image analysis tasks. A frequently used registration method is based on manually or automatically derived point landmarks leading to a sparse displacement field which is densified in a thin-plate spline (TPS) framework. A large problem of TPS interpolation/approximation is the requirement for evenly distributed landmark correspondences over the data set which can rarely be guaranteed by landmark matching algorithms. We propose to overcome this problem by combining the sparse correspondences with intensity-based registration in a generic nonlinear registration scheme based on the calculus of variations. Missing landmark information is compensated by a stronger intensity term, thus combining the strengths of both approaches. An explicit formulation of the generic framework is derived that constrains an intra-modality intensity data term with a regularization term from the corresponding landmarks and an anisotropic image-driven displacement regularization term. An evaluation of this algorithm is performed comparing it to an intensity- and a landmark-based method. Results on four synthetically deformed and four clinical thorax CT data sets at different breathing states are shown.

We present a new approach for the registration of breast MR images, which are acquired at different time points for observation of lesion evolution. In this registration problem, it is of utmost importance to correct only for differences in patient positioning and to preserve other diagnostically important differences between both images, resulting from anatomical and pathological changes between both acquisitions. Classical free form deformation algorithms are therefore less suited, since they allow too large local volume changes and their deformation is not biomechanically based. Instead of adding constraints or penalties to these methods in order to restrict unwanted deformations, we developed a truly biomechanically based registration method where the position of skin and muscle surface are used as the only boundary conditions. Results of our registration method show an important improvement in correspondence between the reference and the deformed floating image, without introducing physically implausible deformations and within a short computational time.

A particular problem in image registration arises for multi-modal images taken from different imaging devices and/or modalities. Starting in 1995, mutual information has shown to be a very successful distance measure for multi-modal image registration. However, mutual information has also a number of well-known drawbacks. Its main disadvantage is that it is known to be highly non-convex and has typically many local maxima.

This observation motivate us to seek a different image similarity measure which is better suited for optimization but as well capable to handle multi-modal images. In this work we investigate an alternative distance measure which is based on normalized gradients and compare its performance to Mutual Information. We call the new distance measure Normalized Gradient Fields (NGF).

A new approach to align an image of a medical object with a given prototype is proposed. Visual appearance of the images, after equalizing their signals, is modelled with a new Markov-Gibbs random field with pairwise interaction model. Similarity to the prototype is measured by a Gibbs energy of signal co-occurrences in a characteristic subset of pixel pairs derived automatically from the prototype. An object is aligned by an affine transformation maximizing the similarity by using an automatic initialization followed by gradient search. Experiments confirm that our approach aligns complex objects better than popular conventional algorithms.

Ultrasound is a universal guidance tool for many medical procedures, whereas it is of poor image quality and resolution. Merging high-contrast image information from other image modalities enhances the guidance capability of ultrasound. However, few registration methods work well for it. In this paper we present a surface-to-image registration technique for mono- or multimodal medical data concerning ultrasound. This approach is able to automatically register the object surface to its counterpart in image volume. Three similarity measurements are investigated in the rigid registration experiments of the pubic arch in transrectal ultrasound images. It shown that the selection of the similarity function is related to the ultrasound characteristics of the object to be registered.

Freehand 3D ultrasound systems acquire sets of B-Mode ultrasound images tagged with position information obtained by a tracking device. For both further processing and clinical use of these ultrasound slice images scattered in space, it is often required to reconstruct them into 3D-rectilinear grid arrays. We propose new efficient methods for this so-called ultrasound spatial compounding using a backward-warping paradigm. They allow to establish 3D-volumes from any scattered freehand ultrasound data with superior quality / speed properties with respect to existing methods. In addition, arbitrary MPR slices can be reconstructed directly from the freehand ultrasound slice set, without the need of an extra volumetric reconstruction step. We qualitatively assess the reconstruction quality and quantitatively compare our compounding method to other algorithms using ultrasound data of the neck and liver. The usefulness of direct MPR reconstruction for multimodal image registration is demonstrated as well.

In this paper a novel approach for the registration and segmentation of dynamic contrast enhanced renal MR images is presented. This integrated method is motivated by the observation of the reciprocity between registration and segmentation in 4D time-series images. Fully automated Fourier-based registration with sub-voxel accuracy and semi-automated time-series segmen-tation were intertwined to improve the accuracy in a multi-step fashion. We have tested our algorithm on several real patient data sets. Clinical validation showed remarkable and consistent agreement between the proposed method and manual segmentation by experts.

This paper proposes a novel method for clinical triple-region image segmentation using a single level set function. Triple-region image segmentation finds wide application in the computer aided X-ray, CT, MRI and ultrasound image analysis and diagnosis. Usually multiple level set functions are used consecutively or simultaneously to segment triple-region medical images. These approaches are either time consuming or suffer from the convergence problems. With the new proposed triple-regions level set energy modelling, the triple-region segmentation is handled within the two region level set framework where only one single level set function needed. Since only a single level set function is used, the segmentation is much faster and more robust than using multiple level set functions. Adapted to the clinical setting, individual principal component analysis and a support vector machine classifier based clinical acceleration scheme are used to accelerate the segmentation. The clinical acceleration scheme takes the strengths of both machine learning and the level set method while limiting their weaknesses to achieve automatic and fast clinical segmentation. Both synthesized and practical images are used to test the proposed method. These results show that the proposed method is able to successfully segment the triple-region using a single level set function. Also this segmentation is very robust to the placement of initial contour. While still quickly converging to the final image, with the clinical acceleration scheme, our proposed method can be used during pre-processing for automatic computer aided diagnosis and surgery.

We present a method for semi-automatic segmentation of the liver from CT scans. True 3D interaction with haptic feedback is used to facilitate initialization, i.e., seeding of a fast marching algorithm. Four users initialized 52 datasets and the mean interaction time was 40 seconds. The segmentation accuracy was verified by a radiologist. Volume measurements and segmentation precision show that the method has a high reproducibility.

PET imagery is a valuable oncology tool for characterizing lesions and assessing lesion response to therapy. These assessments require accurate delineation of the lesion. This is a challenging task for clinicians due to small tumor sizes, blurred boundaries from the large point-spread-function and respiratory motion, inhomogeneous uptake, and nearby high uptake regions. These aspects have led to great variability in lesion assessment amongst clinicians. In this paper, we describe a segmentation algorithm for PET lesions which yields objective segmentations without operator variability. The technique is based on the mean shift algorithm, applied in a spherical coordinate frame to yield a directional assessment of foreground and background and a varying background model. We analyze the algorithm using clinically relevant hybrid digital phantoms and illustrate its effectiveness relative to other techniques.

We present a new method for automatic segmentation of heterogeneous image data, which is very common in medical image analysis. The main contribution of the paper is a mathematical formulation for incorporating soft model assignments into the calculation of affinities, which are traditionally model free. We integrate the resulting model-aware affinities into the multilevel

segmentation by weighted aggregation

algorithm. We apply the technique to the task of detecting and segmenting brain tumor and edema in multimodal MR volumes. Our results indicate the benefit of incorporating model-aware affinities into the segmentation process for the difficult case of brain tumor.

A new physically justified adaptive probabilistic model of blood vessels on magnetic resonance angiography (MRA) images is proposed. The model accounts for both laminar (for normal subjects) and turbulent blood flow (in abnormal cases like anemia or stenosis) and results in a fast algorithm for extracting a 3D cerebrovascular system from the MRA data. Experiments with synthetic and 50 real data sets confirm the high accuracy of the proposed approach.

Recent work shows that diffusion tensor imaging (DTI) can help resolving thalamic nuclei based on the characteristic fiber orientation of the corticothalamic/thalamocortical striations within each nucleus. In this paper we describe a novel segmentation method based on spectral clustering. We use Markovian relaxation to handle spatial information in a natural way, and we explicitly minimize the normalized cut criteria of the spectral clustering for a better optimization. Using this modified spectral clustering algorithm, we can resolve the organization of the thalamic nuclei into groups and subgroups solely based on the voxel affinity matrix, avoiding the need for explicitly defined cluster centers. The identification of nuclear subdivisions can facilitate localization of functional activation and pathology to individual nuclear subgroups.

Segmentations of MR images of the human brain can be generated by propagating an existing atlas label volume to the target image. By fusing multiple propagated label volumes, the segmentation can be improved. We developed a model that predicts the improvement of labelling accuracy and precision based on the number of segmentations used as input. Using a cross-validation study on brain image data as well as numerical simulations, we verified the model. Fit parameters of this model are potential indicators of the quality of a given label propagation method or the consistency of the input segmentations used.

Segmentation of the human cerebral cortex from MRI has been subject of much attention during the last decade. Methods based on active surfaces for representing and extracting the cortical boundaries have shown promising results. We present an active surface method, that extracts the inner and outer cortical boundaries using a combination of different vector fields and a local weighting method based on the intrinsic properties of the deforming surface. Our active surface model deforms polygonal meshes to fit the boundaries of the cerebral cortex using a force balancing scheme. As a result of the local weighting strategy and a self-intersection constraint, the method is capable of modelling tight sulci where the image edge is missing or obscured. The performance of the method is evaluated using both real and simulated MRI data.

Brain MRI segmentation remains a challenging problem in spite of numerous existing techniques. To overcome the inherent difficulties associated with this segmentation problem, we present a new method of information integration in a graph based framework. In addition to image intensity, tissue priors and local boundary information are integrated into the edge weight metrics in the graph. Furthermore, inhomogeneity correction is incorporated by adaptively adjusting the edge weights according to the intermediate inhomogeneity estimation. In the validation experiments of simulated brain MRIs, the proposed method outperformed a segmentation method based on iterated conditional modes (ICM), which is a commonly used optimization method in medical image segmentation. In the experiments of real neonatal brain MRIs, the results of the proposed method have good overlap with the manual segmentations by human experts.

The accuracy and precision of segmentations of medical images has been difficult to quantify in the absence of a “ground truth” or reference standard segmentation for clinical data. Although physical or digital phantoms can help by providing a reference standard, they do not allow the reproduction of the full range of imaging and anatomical characteristics observed in clinical data.

An alternative assessment approach is to compare to segmentations generated by domain experts. Segmentations may be generated by raters who are trained experts or by automated image analysis algorithms. Typically these segmentations differ due to intra-rater and inter-rater variability. The most appropriate way to compare such segmentations has been unclear.

We present here a new algorithm to enable the estimation of performance characteristics, and a true labeling, from observations of segmentations of imaging data where segmentation labels may be ordered or continuous measures. This approach may be used with, amongst others, surface, distance transform or level set representations of segmentations, and can be used to assess whether or not a rater consistently over-estimates or under-estimates the position of a boundary.

The segmentation problem appears in most medical imaging applications. Many research groups are pushing toward a whole body segmentation based on atlases. With a similar objective, we propose a general framework to segment several structures. Rather than inventing yet another segmentation algorithm, we introduce inter-structure spatial dependencies to work with existing segmentation algorithms. Ranking the structures according to their dependencies, we end up with a hierarchical approach that improves each individual segmentation and provides automatic initializations. The best ordering of the structures can be learned off-line. We apply this framework to the segmentation of several structures in brain MR images.

A probabilistic model was proposed in this research for fully-automated segmentation of liver region in non-contrast X-ray torso CT images. This probabilistic model was composed of two kinds of probability that show the location and density (CT number) of the liver in CT images. The probability of the liver on the spatial location was constructed from a number of CT scans in which the liver regions were pre-segmented manually as gold standards. The probability of the liver on density was estimated specifically using a Gaussian function. The proposed probabilistic model was used for automated liver segmentation from non-contrast CT images. 132 cases of the CT scans were used for the probabilistic model construction and then this model was applied to segment liver region based on a leave-one-out method. The performances of the probabilistic model were evaluated by comparing the segmented liver with the gold standard in each CT case. The validity and usefulness of the proposed model were proved.

In this paper, an automatic knowledge-based framework for level set segmentation of 3D calvarial tumors from Computed Tomography images is presented. Calvarial tumors can be located in both soft and bone tissue, occupying wide range of image intensities, making automatic segmentation and computational modeling a challenging task. The objective of this study is to analyze and validate different approaches in intensity priors modeling with an attention to multiclass problems. One, two, and three class Gaussian mixture models and a discrete model are evaluated considering probability density modeling accuracy and segmentation outcome. Segmentation results were validated in comparison to manually segmented golden standards, using analysis in ROC (Receiver Operating Curve) space and Dice similarity coefficient.

It is widely accepted in the medical community that breast tissue density is an important risk factor for the development of breast cancer. Thus, the development of reliable automatic methods for classification of breast tissue is justified and necessary. Although different approaches in this area have been proposed in recent years, only a few are based on the BIRADS classification standard. In this paper we review different strategies for extracting features in tissue classification systems, and demonstrate, not only the feasibility of estimating breast density using automatic computer vision techniques, but also the benefits of segmentation of the breast based on internal tissue information. The evaluation of the methods is based on the full MIAS database classified according to BIRADS categories, and agreement between automatic and manual classification of 82% was obtained.

In view of the great effects of osteoporosis on public health, it would be of great value to be able to measure the three-dimensional structure of trabecular bone in vivo as a means to diagnose and quantify the disease. The aim of this work was to implement a method for quantitative characterisation of trabecular bone structure using clinical CT.

Several previously described parameters have been calculated from volumes acquired with a 64-slice clinical scanner. Using automated region growing, distance transforms and three-dimensional thinning, measures describing the number, thickness and spacing of bone trabeculae was obtained. Fifteen bone biopsies were analysed. The results were evaluated using micro-CT as reference.

For most parameters studied, the absolute values did not agree well with the reference method, but several parameters were closely correlated with the reference method. The shortcomings appear to be due to the low resolution and high noise level. However, the high correlation found between clinical CT and micro-CT measurements suggest that it might be possible to monitor changes in the trabecular structure in vivo.

Fully automatic, completely reliable segmentation in medical images is an unrealistic expectation with today’s technology. However, many automatic segmentation algorithms may achieve a near-correct solution, incorrect only in a small region. For these situations, an interactive editing tool is required, ideally in 3D, that is usually left to a manual correction. We formulate the editing task as an energy minimization problem that may be solved with a modified version of either graph cuts or the random walker 3D segmentation algorithms. Both algorithms employ a seeded user interface, that may be used in this scenario for a user to seed erroneous voxels as belonging to the foreground or the background. In our formulation, it is unnecessary for the user to specify both foreground and background seeds.

The Graph Cuts method of interactive segmentation has become very popular in recent years. This method performs at interactive speeds for smaller images/volumes, but an unacceptable amount of storage and computation time is required for the large images/volumes common in medical applications. The Banded Graph Cut (BGC) algorithm was proposed to drastically increase the computational speed of Graph Cuts, but is limited to the segmentation of large, roundish objects. In this paper, we propose a modification of BGC that uses the information from a Laplacian pyramid to include thin structures into the band. Therefore, we retain the computational efficiency of BGC while providing quality segmentations on thin structures. We make quantitative and qualitative comparisons with BGC on images containing thin objects. Additionally, we show that the new parameter introduced in our modification provides a smooth transition from BGC to traditional Graph Guts.

The quantitative assessment of neck lymph nodes in the context of malign tumors requires an efficient segmentation technique for lymph nodes in tomographic 3D datasets. We present a Stable 3D Mass-Spring Model for lymph node segmentation in CT datasets. Our model for the first time represents concurrently the characteristic gray value range, directed contour information as well as shape knowledge, which leads to a much more robust and efficient segmentation process. Our model design and segmentation accuracy are both evaluated with lymph nodes from clinical CT neck datasets.

An automatic method for lung nodule segmentation from computed tomography (CT) data is presented that is different from previous work in several respects. Firstly, it is supervised; it learns how to obtain a reliable segmentation from examples in a training phase. Secondly, the method provides a soft, or probabilistic segmentation, thus taking into account the uncertainty inherent in this segmentation task. The method is trained and tested on a public data set of 23 nodules for which soft labelings are available. The new method is shown to outperform a previously published conventional method. By merely changing the training data, non-solid nodules can also be segmented.

This paper considers the problem of automatic classification of textured tissues in 3D MRI. More specifically, it aims at validating the use of features extracted from the phase of the MR signal to improve texture discrimination in bone segmentation. This extra information provides better segmentation, compared to using magnitude only features. We also present a novel multiscale scheme to improve the speed of pixel based classification algorithm, such as support vector machines. This algorithm dramatically increases the speed of the segmentation process by an order of magnitude through a reduction of the number of pixels that needs to be classified in the image.

Retinal vessel segmentation is an essential step of the diagnoses of various eye diseases. In this paper, we propose an automatic, efficient and unsupervised method based on gradient matrix, the normalized cut criterion and tracking strategy. Making use of the gradient matrix of the Lucas-Kanade equation, which consists of only the first order derivatives, the proposed method can detect a candidate window where a vessel possibly exists. The normalized cut criterion, which measures both the similarity within groups and the dissimilarity between groups, is used to search a local intensity threshold to segment the vessel in a candidate window. The tracking strategy makes it possible to extract thin vessels without being corrupted by noise. Using a multi-resolution segmentation scheme, vessels with different widths can be segmented at different resolutions, although the window size is fixed. Our method is tested on a public database. It is demonstrated to be efficient and insensitive to initial parameters.

We propose a method for atrophy simulation in structural MR images based on finite-element methods, providing data for objective evaluation of atrophy measurement techniques. The modelling of diffuse global and regional atrophy is based on volumetric measurements from patients with known disease and guided by clinical knowledge of the relative pathological involvement of regions. The consequent biomechanical readjustment of structures is modelled using conventional physics-based techniques based on tissue properties and simulating plausible deformations with finite-element methods. Tissue characterization is performed by means of the meshing of a labelled brain atlas, creating a reference volumetric mesh, and a partial volume tissue model is used to reduce the impact of the mesh discretization. An example of simulated data is shown and a visual evaluation protocol used by experts has been developed to assess the degree of realism of the simulated images. First results demonstrate the potential of the proposed methodology.

In medical imaging, parameterized 3D surface models are of great interest for anatomical modeling and visualization, statistical comparisons of anatomy, and surface-based registration and signal processing. Here we introduce a parameterization method based on algebraic functions. By solving the Yamabe equation with the Ricci flow method, we can conformally map a brain surface to a multi-hole disk. The resulting parameterizations do not have any singularities and are intrinsic and stable. To illustrate the technique, we computed parameterizations of several types of anatomical surfaces in MRI scans of the brain, including the hippocampi and the cerebral cortices with various landmark curves labeled. For the cerebral cortical surfaces, we show the parameterization results are consistent with selected landmark curves and can be matched to each other using constrained harmonic maps. Unlike previous planar conformal parameterization methods, our algorithm does not introduce any singularity points. It also offers a method to explicitly match landmark curves between anatomical surfaces such as the cortex, and to compute conformal invariants for statistical comparisons of anatomy.

The concept of the Logarithm of the Odds (LogOdds) is frequently used in areas such as artificial neural networks, economics, and biology. Here, we utilize LogOdds for a shape representation that demonstrates desirable properties for medical imaging. For example, the representation encodes the shape of an anatomical structure as well as the variations within that structure. These variations are embedded in a vector space that relates to a probabilistic model.

We apply our representation to a voxel based segmentation algorithm. We do so by embedding the manifold of Signed Distance Maps (SDM) into the linear space of LogOdds. The LogOdds variant is superior to the SDM model in an experiment segmenting 20 subjects into subcortical structures.

We also use LogOdds in the non-convex interpolation between space conditioned distributions. We apply this model to a longitudinal schizophrenia study using quadratic splines. The resulting time-continuous simulation of the schizophrenic aging process has a higher accuracy then a model based on convex interpolation.

This paper introduces a new similarity measure for multi-modal image registration task. The measure is based on the generalized survival exponential entropy (GSEE) and mutual information (GSEE-MI). Since GSEE is estimated from the cumulative distribution function instead of the density function, it is observed that the interpolation artifact is reduced. The method has been tested on four real MR-CT data sets. The experimental results show that the GSEE-MI-based method is more robust than the conventional MI-based method. The accuracy is comparable for both methods.