Medical Image Computing and Computer-Assisted Intervention – MICCAI 2007

In algorithms for processing diffusion tensor images, two common ingredients are interpolating tensors, and measuring the distance between them. We propose a new class of interpolation paths for tensors, termed

geodesic-loxodromes

, which explicitly preserve clinically important tensor attributes, such as mean diffusivity or fractional anisotropy, while using basic differential geometry to interpolate tensor orientation. This contrasts with previous Riemannian and Log-Euclidean methods that preserve the determinant. Path integrals of tangents of geodesic-loxodromes generate novel measures of over-all difference between two tensors, and of difference in shape and in orientation.

The presence of Rician noise in magnetic resonance imaging (MRI) introduces systematic errors in diffusion tensor imaging (DTI) measurements. This paper evaluates gradient direction schemes and tensor estimation routines to determine how to achieve the maximum accuracy and precision of tensor derived measures for a fixed amount of scan time. We present Monte Carlo simulations that quantify the effect of noise on diffusion measurements and validate these simulation results against appropriate in-vivo images. The predicted values of the systematic and random error caused by imaging noise are essential both for interpreting the results of statistical analysis and for selecting optimal imaging protocols given scan time limitations.

We present a methodology to achieve 3D high resolution

in-utero

fetal brain DTI that shows excellent ADC as well as promising FA maps. After continuous DTI scanning to acquire a repeated series of parallel slices with 15 diffusion directions, image registration is used to realign the images to correct for fetal motion. Once aligned, the diffusion images are treated as irregularly sampled data where each voxel is associated with an appropriately rotated diffusion direction, and used to estimate the diffusion tensor on a regular grid. The method has been tested successful on eight fetuses and has been validated on adults imaged at 1.5T.

Magnetic resonance diffusion imaging (dMRI) has become an established research tool for the investigation of tissue structure and orientation. In this paper, we present a method for real time processing of diffusion tensor and Q-ball imaging. The basic idea is to use Kalman filtering framework to fit either the linear tensor or Q-ball model. Because the Kalman filter is designed to be an incremental algorithm, it naturally enables updating the model estimate after the acquisition of any new diffusion-weighted volume. Processing diffusion models and maps during ongoing scans provides a new useful tool for clinicians, especially when it is not possible to predict how long a subject may remain still in the magnet.

In this paper, we present a novel approach for the segmentation of white matter tracts based on Finsler active contours. This technique provides an optimal measure of connectivity, explicitly segments the connecting fiber bundle, and is equipped with a metric which is able to utilize the directional information of high angular resolution data. We demonstrate the effectiveness of the algorithm for segmenting the cingulum bundle.

Real-time three-dimensional (RT3D) echocardiography is a new imaging modality that presents the unique opportunity to visualize the complex three-dimensional (3 -D) shape and the motion of left ventricle (LV)

in vivo

. To take advantage of this opportunity, automatic segmentation of LV myocardium is essential. While there are a variety of efforts on the segmentation of LV endocardial (ENDO) boundaries, the segmentation of epicardial (EPI) boundaries is still problematic. In this paper, we present a new approach of coupled-surfaces propagation to address this problem. Our method is motivated by the idea that the volume of the myocardium is close to being constant during a cardiac cycle and takes this tight coupling as an important constraint. We employ two surfaces, each driven by the image-derived information that takes into account the ultrasound physics by modeling speckle using shifted Rayleigh distribution while maintaining the coupling. By evolving two surfaces simultaneously, the final representation of myocardium is thus achieved. Results from 328 sets of RT3D echocardiographic data are evaluated against the outlines of three observers. We show that the results from automatic segmentation are comparable to those from manual segmentation.

To quantitatively predict coronary artery diseases, automated analysis may be preferred to current visual assessment of left ventricular (LV) wall motion. In this paper, a novel automated classification method is presented which uses shape models with localized variations. These sparse shape models were built from four-chamber and two-chamber echocardiographic sequences using principal component analysis and orthomax rotations. The resulting shape parameters were then used to classify local wall-motion abnormalities of LV segments. Various orthomax criteria were investigated. In all cases, higher classification correctness was achieved using significantly less shape parameters than before rotation. Since pathologies are typically spatially localized, many medical applications involving local classification should benefit from orthomax parameterizations.

This paper presents a method for registering 3D intracardiac echo (ICE) to pre-operative images. A magnetic tracking sensor is integrated on the ICE catheter tip to provide the 3D location and orientation. The user guides the catheter into the patient heart to acquire a series of ultrasound images covering the anatomy of the heart chambers. An automatic intensity-based registration algorithm is applied to align these ultrasound images with pre-operative images. One of the important applications is to help electrophysiology doctors to treat complicated atrial fibrillation cases. After registration, the doctor can see the position and orientation of the ICE catheter and other tracked catheters inside the heart anatomy in real time. The image guidance provided by this technique may increase the ablation accuracy and reduce the amount of time for the electrophysiology procedures. We show successful image registration results from animal experiments.

While Minimally Invasive Surgery (MIS) offers great benefits to patients compared with open surgery surgeons suffer from a restricted field-of-view and obstruction from instruments. We present a novel method for 3D reconstruction of soft tissue, which can provide a wider field-of-view with 3D information for surgeons, including restoration of missing data. The paper focuses on the use of Structure from Motion (SFM) techniques to solve the missing data problem and application of competitive evolutionary agents to improve the robustness to missing data and outliers. The method has been evaluated with synthetic data, images from a phantom heart model, and

in vivo

MIS image sequences using the da Vinci telerobotic surgical system.

Off-pump Coronary Artery Bypass Grafting (CABG) is still today a technically difficult procedure. In fact, the mechanical stabilizers used to locally suppress the heart excursion have been demonstrated to exhibit significant residual motion. We therefore propose a novel active stabilizer which is able to compensate for this residual motion. The interaction between the heart and a mechanical stabilizer is first assessed in vivo on an animal model. Then, the principle of active stabilization, based on the high speed vision-based control of a compliant mechanism, is presented. In vivo experimental results are given using a prototype which structure is compatible with a minimally invasive approach.

An atlas-based automated liver segmentation method from 3D CT images is described. The method utilizes two types of atlases, that is, the probabilistic atlas (PA) and statistical shape model (SSM). Voxel-based segmentation with PA is firstly performed to obtain a liver region, and then the obtained region is used as the initial region for subsequent SSM fitting to 3D CT images. To improve reconstruction accuracy especially for largely deformed livers, we utilize a multi-level SSM (ML-SSM). In ML-SSM, the whole shape is divided into patches, and principal component analysis is applied to each patches. To avoid the inconsistency among patches, we introduce a new constraint called the adhesiveness constraint for overlap regions among patches. In experiments, we demonstrate that segmentation accuracy improved by using the initial region obtained with PA and the introduced constraint for ML-SSM.

This paper presents a new atlas-based segmentation framework for the delineation of major regions in magnetic resonance brain images employing an atlas of the global topological structure as well as a statistical atlas of the regions of interest. A segmentation technique using fast marching methods and tissue classification is proposed that guarantees strict topological equivalence between the segmented image and the atlas. Experimental validation on simulated and real brain images shows that the method is accurate and robust.

As demands on hospital efficiency increase, there is a stronger need for automatic analysis, recovery, and modification of surgical workflows. Even though most of the previous work has dealt with higher level and hospital-wide workflow including issues like document management, workflow is also an important issue within the surgery room. Its study has a high potential, e.g., for building context-sensitive operating rooms, evaluating and training surgical staff, optimizing surgeries and generating automatic reports.

In this paper we propose an approach to segment the surgical workflow into phases based on temporal synchronization of multidimensional state vectors. Our method is evaluated on the example of laparoscopic cholecystectomy with state vectors representing tool usage during the surgeries. The discriminative power of each instrument in regard to each phase is estimated using AdaBoost. A boosted version of the Dynamic Time Warping (DTW) algorithm is used to create a surgical reference model and to segment a newly observed surgery. Full cross-validation on ten surgeries is performed and the method is compared to standard DTW and to Hidden Markov Models.

In functional connectivity analysis, networks of interest are defined based on correlation with the mean time course of a user-selected ‘seed’ region. In this work we propose to simultaneously estimate the optimal representative time courses that summarize the fMRI data well and the partition of the volume into a set of disjoint regions that are best explained by these representative time courses. Our approach offers two advantages. First, is removes the sensitivity of the analysis to the details of the seed selection. Second, it substantially simplifies group analysis by eliminating the need for a subject-specific threshold at which correlation values are deemed significant. This unsupervised technique generalizes connectivity analysis to situations where candidate seeds are difficult to identify reliably or are unknown. Our experimental results indicate that the functional segmentation provides a robust, anatomically meaningful and consistent model for functional connectivity in fMRI.

We present a robotically assisted prostate brachytherapy system and test results in training phantoms. The system consists of a transrectal ultrasound (TRUS) and a spatially co-registered robot integrated with an FDA-approved commercial treatment planning system. The salient feature of the system is a small parallel robot affixed to the mounting posts of the template. The robot replaces the template interchangeably and uses the same coordinate system. Established clinical hardware, workflow and calibration are left intact. In these experiments, we recorded the first insertion attempt without adjustment. All clinically relevant locations were reached. Non-parallel needle trajectories were achieved. The pre-insertion transverse and rotational errors (measured with Polaris optical tracker relative to the template’s coordinate frame) were 0.25mm (STD=0.17mm) and 0.75° (STD=0.37°). The needle tip placement errors measured in TRUS were 1.04mm (STD=0.50mm). The system is in Phase-I clinical feasibility and safety trials, under Institutional Review Board approval.

Multi-modality fusion imaging for targeted prostate biopsy is difficult because of prostate motion during the biopsy procedure. A closed-loop control mechanism is proposed to improve the efficacy and safety of the biopsy procedure, which uses real-time ultrasound and spatial tracking as feedback to adjust the registration between a preoperative 3D image (e.g. MRI) and real-time ultrasound images. The spatial tracking data is used to initialize the image-based registration between intraoperative ultrasound images and a preoperative ultrasound volume. The preoperative ultrasound volume is obtained using a 2D sweep and manually registered to the MRI dataset before the biopsy procedure. The accuracy of the system is 2.3±0.9 mm in phantom studies. The results of twelve patient studies show that prostate motion can be effectively compensated using closed-loop control.

The fusion of 3D freehand ultrasound with CT and CTA has benefits for a variety of clinical applications, however a lot of manual work is usually required for correct registration. We developed new methods that allow one to simulate medical ultrasound from CT in real-time, reproducing the majority of ultrasonic imaging effects. The second novelty is a robust similarity measure that assesses the correlation of a combination of multiple signals extracted from CT with ultrasound, without knowing the influence of each signal. This serves as the foundation of a fully automatic registration, which aligns a freehand ultrasound sweep with the corresponding 3D modality using a rigid or an affine transformation model, without any manual interaction. We also present the used initialization, global and local parameter optimization schemes, and validation on abdominal CTA and ultrasound imaging of 10 patients.

A 1.5 mm magnetic sphere was navigated automatically inside the carotid artery of a living swine. The propulsion force, tracking and real-time capabilities of a Magnetic Resonance Imaging (MRI) system were integrated into a closed loop control platform. The sphere was released using an endovascular catheter approach. Specially developed software is responsible for the tracking, propulsion, event timing and closed loop position control in order to follow a 10 roundtrips preplanned trajectory on a distance of 5 cm inside the right carotid artery of the animal. Experimental protocol linking the technical aspects of this

in vivo

assay is presented. In the context of this demonstration, many challenges which provide insights about concrete issues of future nanomedical interventions and interventional platforms have been identified and addressed.

The assessment and diagnosis of breast cancer with ultrasound is a challenging problem due to the low contrast between cancer masses and benign tissue. Due to this low contrast it has proven to be difficult to achieve reliable segmentation results on breast cancer masses. An autoregressive model has been employed to filter out of the backscattered RF-signal from a tissue harmonic image which is not degraded by harmonic leakage. Measurements on the filtered image have shown a significant (up to 45 %) increase in contrast between cancer masses and benign tissue.

This paper introduces an outlier rejection and signal reconstruction method for high angular resolution diffusion weighted imaging. The approach is based on the thresholding of Laplacian measurements over the sphere of the apparent diffusion coefficient profiles defined for a given set of gradient directions. Exemplary results are presented.

In Diffusion Tensor Imaging (DTI), differently oriented fiber bundles inside one voxel are incorrectly modeled by a single tensor. High Angular Resolution Diffusion Imaging (HARDI) aims at using more complex models, such as a two-tensor model, for estimating two fiber bundles.

We propose a new method for creating experimental phantom data of fiber crossings, by mixing the DWI-signals from high FA-regions with different orientation. The properties of these experimental phantoms approach the conditions of real data. These phantoms can thus serve as a ‘ground truth’ in validating crossing reconstruction algorithms. The angular resolution of a dual tensor model is determined using series of crossings, generated under different angles. An angular resolution of 0.6

π

was found in data scanned with a diffusion weighting parameter

b

=1000 s/mm

2

. This resolution did not change significantly in experiments with

b

=3000 and 5000 s/mm

2

, keeping the scanning time constant.

CT-images acquired by mobile C-arm devices can contain artefacts caused by positioning errors. We propose a data driven method based on iterative 3D-reconstruction and 2D/3D-registration to correct projection data inconsistencies. With a 2D/3D-registration algorithm, transformations are computed to align the acquired projection images to a previously reconstructed volume. In an iterative procedure, the reconstruction algorithm uses the results of the registration step. This algorithm also reduces small motion artefacts within 3D-reconstructions. Experiments with simulated projections from real patient data show the feasibility of the proposed method. In addition, experiments with real projection data acquired with an experimental robotised C-arm device have been performed with promising results.

We present large deformation diffeomorphic metric curve mapping (LDDMM-Curve) for registering cortical hemispheres. We showed global cortical hemisphere matching and evaluated the mapping accuracy in five subregions of the cortex in fourteen MRI scans.

We present a novel system for 3-D visualisation of the heart and coronary arteries. Binary tags (generated offline) are combined with value-gradient transfer functions (specified online) allowing for interactive visualisation, while relaxing the offline segmentation criteria. The arteries are roughly segmented using a Hessian-based line filter and the pericardial cavity using a Fast Marching active contour. A comparison of different contour initialisations reveals that simple geometric shapes (such as spheres or extruded polygons) produce suitable results.

Intravascular ultrasound (IVUS) is an invasive imaging modality capable of providing cross-sectional images of the interior of a blood vessel in real time and at normal video framerates (10-30 frames/s). Low contrast between the features of interest in the IVUS imagery remains a confounding factor in IVUS analysis; it would be beneficial therefore to have a method capable of detecting certain physical features imaged under IVUS in an automated manner. We present such a method and apply it to the detection of blood. While blood detection algorithms are not new in this field, we deviate from traditional approaches to IVUS signal characterization in our use of 1-class learning. This eliminates certain problems surrounding the need to provide “foreground” and “background” (or, more generally,

n

-class) samples to a learner. Applied to the blood-detection problem on 40 MHz recordings made

in vivo

in swine, we are able to achieve ~95% sensitivity with ~90% specificity at a radial resolution of ~600

μ

m.

This paper presents a novel method for phase unwrapping for phase sensitive reconstruction in MR imaging. The unwrapped phase is obtained by integrating the phase gradient by solving a Poisson equation. An efficient solver, which has been made publicly available, is used to solve the equation. The proposed method is demonstrated on a fat quantification MRI task that is a part of a prospective study of fat accumulation. The method is compared to a phase unwrapping method based on region growing. Results indicate that the proposed method provides more robust unwrapping. Unlike region growing methods, the proposed method is also straight-forward to implement in 3D.

We propose to carry out cooperatively both tissue and structure segmentations by distributing a set of

local

and

cooperative

models in a unified MRF framework. Tissue segmentation is performed by partitionning the volume into subvolumes where local MRFs are estimated in cooperation with their neighbors to ensure consistency. Local estimation fits precisely to the local intensity distribution and thus handles nonuniformity of intensity without any bias field modelization. Structure segmentation is performed via local MRFs that integrate localization constraints provided by

a priori

general fuzzy description of brain anatomy. Structure segmentation is not reduced to a postprocessing step but cooperates with tissue segmentation to gradually and conjointly improve models accuracy. The evaluation was performed using phantoms and real 3T brain scans. It shows good results and in particular robustness to nonuniformity and noise with a low computational cost.

With the advent of biomarkers such as

11

C-PIB and the increase in use of PET, automated methods are required for processing and analyzing datasets from research studies and in clinical settings. A common preprocessing step is the calculation of standardized uptake value ratio (SUVR) for inter-subject normalization. This requires segmented grey matter (GM) for VOI refinement. However

11

C-PIB uptake is proportional to amyloid build up leading to inhomogeneities in intensities, especially within GM. Inhomogeneities present a challenge for clustering and pattern classification based approaches to PET segmentation as proposed in current literature.

In this paper we modify a MR image segmentation technique based on expectation maximization for

11

C-PIB PET segmentation. A priori probability maps of the tissue types are used to initialize and enforce anatomical constraints. We developed a Bézier spline based inhomogeneity correction techniques that is embedded in the segmentation algorithm and minimizes inhomogeneity resulting in better segmentations of

11

C-PIB PET images. We compare our inhomogeneity with a global polynomial correction technique and validate our approach using co-registered MRI segmentations.

A mapping of unit vectors onto a 5D hypersphere is used to model and partition ODFs from HARDI data. This mapping has a number of useful and interesting properties and we make a link to interpretation of the second order spherical harmonic decompositions of HARDI data. The paper presents the working theory and experiments of using a von Mises-Fisher mixture model for directional samples. The MLE of the second moment of the HvMF pdf can also be related to fractional anisotropy. We perform error analysis of the estimation scheme in single and multi-fibre regions and then show how a penalised-likelihood model selection method can be employed to differentiate single and multiple fibre regions.

An optimal 3-D graph search approach designed for simultaneous multiple surface detection is extended to allow for varying smoothness and surface interaction constraints instead of the traditionally used constant constraints. We apply the method to the intraretinal layer segmentation of 24 3-D optical coherence tomography (OCT) images, learning the constraints from examples in a leave-one-subject-out fashion. Introducing the varying constraints decreased the mean unsigned border positioning errors (mean error of 7.3 ± 3.7

μ

m using varying constraints compared to 8.3 ± 4.9

μ

m using constant constraints and 8.2 ± 3.5

μ

m for the inter-observer variability).

In this paper, we propose a fully automatic method for the coupled 3D localization and segmentation of lower abdomen structures. We apply it to the joint segmentation of the prostate and bladder in a database of CT scans of the lower abdomen of male patients. A flexible approach on the bladder allows the process to easily adapt to high shape variation and to intensity inhomogeneities that would be hard to characterize (due, for example, to the level of contrast agent that is present). On the other hand, a statistical shape prior is enforced on the prostate. We also propose an adaptive non–overlapping constraint that arbitrates the evolution of both structures based on the availability of strong image data at their common boundary. The method has been tested on a database of 16 volumetric images, and the validation process includes an assessment of inter–expert variability in prostate delineation, with promising results.

Disease-specific understanding of echocardiographic sequences requires accurate characterization of spatio-temporal motion patterns. In this paper we present a method of automatic extraction and matching of spatio-temporal patterns from cardiac echo videos. Specifically, we extract cardiac regions (chambers and walls) using a variation of multiscale normalized cuts that combines motion estimates from deformable models with image intensity. We then derive spatio-temporal trajectories of region measurements such as wall motion, volume and thickness. The region trajectories are then matched to infer the similarities in disease labels of patients. Validation results on patient data sets collected from many hospitals are presented.

Because of their physiological meaningfulness, cardiac physiome models have been used as constraints to recover patient information from medical images. Although the results are promising, the parameters of the physiome models are not patient-specific, and thus affect the clinical relevance of the recovered information especially in pathological cases. In view of this problem, we incorporate patient information from body surface potential maps in the physiome model to provide a more patient-specific while physiological plausible guidance, which is further coupled with patient measurements derived from structural images to recover the cardiac geometry and deformation simultaneously. Experiments have been conducted on synthetic data to show the benefits of the framework, and on real human data to show its practical potential.

We propose a highly automated approach to the point correspondence problem for anatomical shapes in medical images. Manual landmarking is performed on a

small subset

of the shapes in the study, and a machine learning approach is used to elucidate the characteristic

shape and appearance

features at each landmark. A classifier trained using these features defines a cost function that drives key landmarks to anatomically meaningful locations after MDL-based correspondence establishment. Results are shown for artificial examples as well as real data.

In this paper we propose a new approach for false positive reduction in the field of mammographic mass detection. The goal is to distinguish between the true recognized masses and the ones which actually are normal parenchyma. Our proposal is based on Local Binary Patterns (LBP) for representing salient micro-patterns and preserving at the same time the spatial structure of the masses. Once the descriptors are extracted, Support Vector Machines (SVM) are used for classifying the detected masses. We test our proposal using a set of 1792 suspicious regions of interest extracted from the DDSM database. Exhaustive experiments illustrate that LBP features are effective and efficient for false positive reduction even at different mass sizes, a critical aspect in mass detection systems. Moreover, we compare our proposal with current methods showing that LBP obtains better performance.

This paper presents a novel segmentation-based approach for fiber-tract extraction in diffusion-tensor (DT) images. Typical tractography methods, incorporating thresholds on fractional anisotropy and fiber curvature to terminate tracking, can face serious problems arising from partial voluming and noise. For these reasons, tractography often fails to extract thin tracts with sharp changes in orientation, e.g. the

cingulum

. Unlike tractography—which disregards the information in the tensors that were previously tracked—the proposed method extracts the cingulum by exploiting the statistical coherence of tensors in the entire structure. Moreover, the proposed segmentation-based method allows

fuzzy

class memberships to optimally extract information within partial-volumed voxels. Unlike typical fuzzy-segmentation schemes employing Gaussian models that are biased towards ellipsoidal clusters, the proposed method

models the manifolds

underlying the classes by incorporating nonparametric data-driven statistical models. Furthermore, it exploits the nonparametric model to capture the

spatial continuity and structure

of the fiber bundle. The results on real DT images demonstrate that the proposed method extracts the cingulum bundle significantly more accurately as compared to tractography.

In this paper, we introduce an adaptive model-based segmentation framework, in which edge and region information are integrated and used adaptively while a solid model deforms toward the object boundary. Our 3D segmentation method stems from Metamorphs deformable models [1]. The main novelty of our work is in that, instead of performing segmentation in an entire 3D volume, we propose model-based segmentation in an adaptively changing subvolume of interest. The subvolume is determined based on appearance statistics of the evolving object model, and within the subvolume, more accurate and object-specific edge and region information can be obtained. This local and adaptive scheme for computing edges and object region information makes our segmentation solution more efficient and more robust to image noise, artifacts and intensity inhomogeneity. External forces for model deformation are derived in a variational framework that consists of both edge-based and region-based energy terms, taking into account the adaptively changing environment. We demonstrate the performance of our method through extensive experiments using cardiac MR and liver CT images.

We propose a new segmentation algorithm based on competing fuzzy connectedness theory, which is then used for visualizing coronary arteries in 3D CT angiography (CTA) images. The major difference compared to other fuzzy connectedness algorithms is that an additional data structure, the connectedness tree, is constructed at the same time as the seeds propagate. In preliminary evaluations, accurate result have been achieved with very limited user interaction. In addition to improving computational speed and segmentation results, the fuzzy connectedness tree algorithm also includes automated extraction of the vessel centerlines, which is a promising approach for creating curved plane reformat (CPR) images along arteries’ long axes.

In this paper, an original approach for the segmentation of tensor fields is proposed. Based on the modeling of the data by means of Gaussian mixtures directly in the tensor domain, this technique presents a wide range of applications in medical image processing, particularly for Diffusion Tensor Magnetic Resonance Imaging (DT-MRI). The performance of the segmentation method proposed is shown through the segmentation of the corpus callosum from a dataset of 32 DT-MRI volumes. Comparison with a recent and related segmentation approach is favorable to our method, showing its capability for the automatic extraction of anatomical structures in the white matter.

We present a new segmentation approach for the myocardium in gated and non-gated perfusion SPECT images. To this end, we represent the epi- and endocardium by separate signed distance functions and couple them by a soft constraint to give explicit control over the wall thickness. By an explicit modeling of the basal plane, the volume of the blood pool as well as the myocardium are determinable. Furthermore, prior shape information is incorporated by applying a kernel density estimation on a number of expert segmentations in a low-dimensional PCA subspace. Thereby, information along the time axis is fully taken into account by employing 4-dimensional embedding functions.

In this paper we present a new methodology for cardiac motion tracking in tagged MRI using nonrigid image registration based on subdivision surfaces and subdivision lattices. We use two sets of registrations to do the motion tracking. First, a set of surface registrations is used to create and initially align the subdivision model of the left ventricle with short-axis and long-axis MR images. Second, a series of volumetric registrations are used to perform the motion tracking and to reconstruct the 4D cardiac motion field from the tagged MR images. The motion of a point in the myocardium over time is calculated by registering the images taken during systole to the set of reference images taken at end-diastole. Registration is achieved by optimizing the positions of the vertices in the base lattice so that the mutual information of the images being registered is maximized. The presented method is validated using a cardiac motion simulator and we also present strain measurements obtained from a group of normal volunteers.

We extend our static multimodal nonrigid registration [1] to a spatio-temporal (2D+T) co-registration of a real-time 3D ultrasound and a cardiovascular MR sequence. The motivation for our research is to assist a clinician to automatically fuse the information from multiple imaging modalities for the early diagnosis and therapy of cardiac disease. The deformation field between both sequences is decoupled into spatial and temporal components. Temporal alignment is firstly performed to re-slice both sequences using a differential registration method. Spatial alignment is then carried out between the frames corresponding to the same temporal position. The spatial deformation is modeled by the polyaffine transformation whose anchor points (or control points) are automatically detected and refined by calculating a local mis-match measure based on phase mutual information. The spatial alignment is built in an adaptive multi-scale framework to maximize the phase-based similarity measure by optimizing the parameters of the polyaffine transformation. Results demonstrate that this novel method can yield an accurate registration to particular cardiac regions.

In this paper, we explore the use of fiber bundles extracted from diffusion MR images for a nonlinear registration algorithm. We employ a white matter atlas to automatically label major fiber bundles and to establish correspondence between subjects. We propose a polyaffine framework to calculate a smooth and invertible nonlinear warp field based on these correspondences, and derive an analytical solution for the reorientation of the tensor fields under the polyaffine transformation. We demonstrate our algorithm on a group of subjects and show that it performs comparable to a higher dimensional nonrigid registration algorithm.

Current clinical and research neuroimaging protocols acquire images using multiple modalities, for instance, T1, T2, diffusion tensor and cerebral blood flow magnetic resonance images (MRI). These multivariate datasets provide unique and often complementary anatomical and physiological information about the subject of interest. We present a method that uses fused multiple modality (scalar and tensor) datasets to perform intersubject spatial normalization. Our multivariate approach has the potential to eliminate inconsistencies that occur when normalization is performed on each modality separately. Furthermore, the multivariate approach uses a much richer anatomical and physiological image signature to infer image correspondences and perform multivariate statistical tests. In this initial study, we develop the theory for Multivariate Symmetric Normalization (MVSyN), establish its feasibility and discuss preliminary results on a multivariate statistical study of 22q deletion syndrome.

Accurate registration of cortical structures plays a fundamental role in statistical analysis of brain images across population. This paper presents a novel framework for the non-rigid intersubject brain surface registration, using conformal structure and spherical thin-plate splines. By resorting to the conformal structure, complete characteristics regarding the intrinsic cortical geometry can be retained as a mean curvature function and a conformal factor function defined on a canonical, spherical domain. In this transformed space, spherical thin-plate splines are firstly used to explicitly match a few prominent homologous landmarks, and in the meanwhile, interpolate a global deformation field. A post-optimization procedure is then employed to further refine the alignment of minor cortical features based on the geometric parameters preserved on the domain. Our experiments demonstrate that the proposed framework is highly competitive with others for brain surface registration and population-based statistical analysis. We have applied our method in the identification of cortical abnormalities in PET imaging of patients with neurological disorders and accurate results are obtained.

Hypothesis testing is an important way to detect the statistical difference between two populations. In this paper, we use the Fisher permutation and bootstrap tests to differentiate hippocampal shape between genders. These methods are preferred to traditional hypothesis tests which impose assumptions on the distribution of the samples. An efficient algorithm is adopted to rapidly perform the

exact

tests. We extend this algorithm to multivariate data by projecting the original data onto an “informative direction” to generate a scalar test statistic. This “informative direction” is found to preserve the original discriminative information. This direction is further used in this paper to isolate the discriminative shape difference between classes from the individual variability, achieving a visualization of shape discrepancy.

We propose a novel kidney segmentation approach based on the graph cuts technique. The proposed approach depends on both image appearance and shape information. Shape information is gathered from a set of training shapes. Then we estimate the shape variations using a new distance probabilistic model which approximates the marginal densities of the kidney and its background in the variability region using a Poisson distribution refined by positive and negative Gaussian components. To segment a kidney slice, we align it with the training slices so we can use the distance probabilistic model. Then its gray level is approximated with a LCG with sign-alternate components. The spatial interaction between the neighboring pixels is identified using a new analytical approach. Finally, we formulate a new energy function using both image appearance models and shape constraints. This function is globally minimized using

s

/

t

graph cuts to get the optimal segmentation. Experimental results show that the proposed technique gives promising results compared to others without shape constraints.

Non-linear attenuation of the Arterial Input Function (AIF) is a major problem in first-pass MR perfusion imaging due to the high concentration of the contrast agent in the blood pool. This paper presents a technique to reconstruct the true AIF using signal intensities in the myocardium and the attenuated AIF based on a Hierarchical Bayesian Model (HBM). With the proposed method, both the AIF and the response function are modeled as smoothed functions by using Bayesian penalty splines (P-Splines). The derived AIF is then used to estimate the impulse response of the myocardium based on deconvolution analysis. The proposed technique is validated both with simulated data using the MMID4 model and ten

in vivo

data sets for estimating myocardial perfusion reserve rates. The results demonstrate the ability of the proposed technique in accurately reconstructing the desired AIF for myocardial perfusion quantification. The method does not involve any MRI pulse sequence modification, and thus is expected to have wider clinical impact.

This paper presents a fast image synthesis procedure for elastic volumes under deformation. Given the node displacements of a mesh and the 3D image voxel data of an undeformed volume, the method maps the image plane pixels to be synthesized from the deformed configuration back to the nominal pre-deformed configuration, where the pixel intensities are obtained easily through interpolation in the regular-grid structure of the voxel volume. For smooth interpolation, this mapping requires the identification of the mesh element enclosing each image pixel. To accelerate this

point location

procedure, a fast method of marking the image pixels is employed by finding the intersection of the mesh and the image, and marking this intersection on the image pixels using

Bresenham’s line drawing algorithm

. A deformable tissue phantom was constructed, it was modeled using the finite element method, and its 3D ultrasound volume was acquired in its undeformed state. Actual B-mode images of the phantom under deformation by the ultrasound probe were then compared with the corresponding synthesized images simulated for the same deformations. Results show that realistic images can be synthesized in real-time using the proposed technique.

In the last decade several methods for extracting the human cerebral cortex from magnetic resonance images have been proposed. Studies comparing these methods have been few. In this study we compare a recent cortical extraction method with FreeSurfer, which has been widespread in the scientific community during recent years. The comparison is performed using realistic phantoms generated from surfaces extracted from original brain scans. The geometrical accuracy of the reconstructed surfaces is compared to the surfaces extracted from the original scan. We found that our method is comparable with FreeSurfer in terms of accuracy, and in some cases it performs better. In terms of speed our method is more than 25 times faster.

The performance of robotic assisted minimally invasive beating heart surgery is a challenging task due to the rhythmic motion of the heart, which hampers delicate tasks such as small vessel anastomosis. In this paper, a virtual motion compensation scheme is proposed for stabilizing images from the surgical site. The method uses vision based 3D tracking to accurately infer cardiac surface deformation and augmented reality for rendering a motion stabilized view for improved surgical performance. The method forgoes the need of fiducial markers and can be integrated with the existing master-slave robotic consoles. The proposed technique is validated with both simulated surgical scenes with known ground truth and

in vivo

data acquired from a TECAB procedure. The experimental results demonstrate the potential of the proposed technique in performing microscale tasks in a moving frame of reference with improved precision and repeatability.

Numerous studies have demonstrated the efficacy of image-guided needle-based therapy and biopsy in the management of prostate cancer. The accuracy of traditional prostate interventions performed using transrectal ultrasound (TRUS) is limited by image fidelity, needle template guides, needle deflection and tissue deformation. Magnetic Resonance Imaging (MRI) is an ideal modality for guiding and monitoring such interventions due to its excellent visualization of the prostate, its sub-structure and surrounding tissues. We have designed a comprehensive robotic assistant system that allows prostate biopsy and brachytherapy procedures to be performed entirely inside a 3T closed MRI scanner. We present a detailed design of the robotic manipulator and an evaluation of its usability and MR compatibility.

This paper introduces a new method for navigated spine surgery using a stereoscopic video see-through head-mounted display (HMD) and an optical tracking system. Vertebrae are segmented from volumetric CT data and visualized in-situ. A surgical drilling device is virtually extended with a mirror for intuitive planning of the drill canal, control of drill direction and insertion depth. The first designated application for the virtually extended drilling device is the preparation of canals for pedicle screw implantation in spine surgery. The objective of surgery is to install an internal fixateur for stabilization of injured vertebrae. We invited five surgeons of our partner clinic to test the system with realistic replica of lumbar vertebrae and compared the new approach with the classical, monitor-based navigation system providing three orthogonal slice views on the operation site. We measured time of procedure and scanned the drilled vertebrae with CT to verify accuracy of drilling.

Target registration error (TRE) refers to the uncertainty in localizing a point of interest after a point-based registration is performed. Common in medical image registration, the metric is typically represented as a root-mean-square statistic. In the late 1990s, a statistical model was developed based on the rigid body definition of the fiducial markers and the localization error associated in measuring the fiducials. The statistical model assumed that the fiducial localizer error was isotropic, but recently the model was reworked to handle anisotropic fiducial localizer error (FLE).

In image guided surgery, the statistical model is used to predict the surgical tool tip tracking accuracy associated with optical spatial measurement systems for which anisotropic FLE models are required. However, optical tracking systems often track the surgical tools relative to a patient based reference tool. Here the formulation for modeling the TRE of a surgical probe relative to a reference frame is developed mathematically and evaluated using a Monte Carlo simulation. The effectiveness of the statistical model is directly related to the FLE model, the fiducial marker design and the distance from centroid to target.

An important part of image-guided radiation therapy or surgery is registration of a three-dimensional (3D) preoperative image to two-dimensional (2D) images of the patient. It is expected that the accuracy and robustness of a 3D/2D image registration method do not depend solely on the registration method itself but also on the number and projections (views) of intraoperative images. In this study, we systematically investigate these factors by using registered image data, comprising of CT and X-ray images of a cadaveric lumbar spine phantom and the recently proposed 3D/2D registration method [1], [2]. The results indicate that the proportion of successful registrations (robustness) significantly increases when more X-ray images are used for registration.

In abdominal surgery, a laparoscopic ultrasound transducer is commonly used to detect lesions such as metastases. The determination and visualization of position and orientation of its flexible tip in relation to the patient or other surgical instruments can be of much help to (novice) surgeons utilizing the transducer intraoperatively. This difficult subject has recently been paid attention to by the scientific community [1,2,3,4,5,6]. Electromagnetic tracking systems can be applied to track the flexible tip. However, the magnetic field can be distorted by ferromagnetic material. This paper presents a new method based on optical tracking of the laparoscope and magneto-optic tracking of the transducer, which is able to automatically detect field distortions. This is used for a smooth augmentation of the B-scan images of the transducer directly on the camera images in real time.

This paper proposes an evaluation of a novel calibration method for an optically tracked oblique laparoscope. We present the necessary tools to track an oblique scope and a camera model which includes changes to the intrinsic camera parameters thereby extending previously proposed methods. Because oblique scopes offer a wide ‘virtual’ view on the surgical field, the method is of great interest for augmented reality guidance of laparoscopic interventions using an oblique scope.

The model and an approximated version are evaluated in an extensive validation study. Using 5 sets of 40 calibration images, we compare both camera models (i.e. model and approximation) and 2 interpolation schemes. The selected model and interpolation scheme reaches an average accuracy of 2.60 pixel and an equivalent 3D error of 0.60 mm.

Finally, we present initial experience of the presented approach with an oblique scope and optical tracking in a clinical setup. During a laparoscopic rectum resection surgery the setup was used to augment the scene with a model of the pelvis. The method worked properly and the attached probes did not interfere with normal procedure.

Navigated bronchoscopy has been developed by various groups within the last decades. Systems based on CT data and electromagnetic tracking enable the visualization of the position and orientation of the bronchoscope, forceps, and biopsy tools within CT data. Therefore registration between the tracking space and the CT volume is required. Standard procedures are based on point-based registration methods that require selecting corresponding natural landmarks in both coordinate systems by the examiner. We developed a novel algorithm for a fully automatic registration procedure in navigated bronchoscopy based on the trajectory recorded during routine examination of the airways at the beginning of an intervention. The proposed system provides advantages in terms of an unchanged medical workflow and high accuracy. We compared the novel method with point-based and ICP-based registration. Experiments demonstrate that the novel method transforms up to 97% of tracking points inside the segmented airways, which was the best performance compared to the other methods.

This paper presents an automatic surgical target and trajectory identification technique for planning deep brain stimulation (DBS) procedures. The probabilistic functional maps, constructed from population-based actual stimulating field information and intra-operative electrophysiological activities, were integrated into a neurosurgical visualization and navigation system to facilitate the surgical planning and guidance. In our preliminary studies, we compared the actual surgical target locations and trajectories established by an experienced stereotactic neurosurgeon with those automatically planned using our probabilistic functional maps on 10 subthalamic nucleus (STN) DBS procedures. The average displacement between the surgical target locations in both groups was 1.82mm with a standard deviation of 0.77mm. The difference between the surgical trajectories was 3.1 º and 2.3 º in the lateral-to-medial and anterior-to-posterior orientations respectively.

We present software engineering methods to provide free open-source software for MR-guided therapy. We report that graphical representation of the surgical tools, interconnectively with the tracking device, patient-to-image registration, and MRI-based thermal mapping are crucial components of MR-guided therapy in sharing such software. Software process includes a network-based distribution mechanism by multi-platform compiling tool CMake, CVS, quality assurance software DART. We developed six procedures in four separate clinical sites using proposed software engineering and process, and found the proposed method is feasible to facilitate multicenter clinical trial of MR-guided therapies. Our future studies include use of the software in non-MR-guided therapies.

We present an iterative bootstrapping framework to create and analyze statistical atlases of bony anatomy such as the human pelvis from a large collection of CT data sets. We create an initial tetrahedral mesh representation of the target anatomy and use deformable intensity-based registration to create an initial atlas. This atlas is used as prior information to assist in deformable registration/segmentation of our subject image data sets, and the process is iterated several times to remove any bias from the initial choice of template subject and to improve the stability and consistency of mean shape and variational modes. We also present a framework to validate the statistical models. Using this method, we have created a statistical atlas of full pelvis anatomy with 110 healthy patient CT scans. Our analysis shows that any given pelvis shape can be approximated up to an average accuracy of 1.5036 mm using the first 15 principal modes of variation. Although a particular intensity-based deformable registration algorithm was used to produce these results, we believe that the basic method may be adapted readily for use with any registration method with broadly similar characteristics.

This paper introduces a new benchmark study of evaluating landmark-based shape correspondence used for statistical shape analysis. Different from previous shape-correspondence evaluation methods, the proposed benchmark first generates a large set of synthetic shape instances by randomly sampling a specified ground-truth statistical shape model. We then run the test shape-correspondence algorithms on these synthetic shape instances to construct a new statistical shape model. We finally introduce a new measure to describe the difference between this newly constructed statistical shape model and the ground truth. This new measure is then used to evaluate the performance of the test shape-correspondence algorithm. By introducing the ground-truth statistical shape model, we believe the proposed benchmark allows for a more objective evaluation of the shape correspondence than those that do not specify any ground truth.

The goal of this work is the automatic inference of frequent patterns of the cortical sulci, namely patterns that can be observed only for a subset of the population. The sulci are detected and identified using brainVISA open software. Then, each sulcus is represented by a set of shape descriptors called the 3D moment invariants. Unsupervised agglomerative clustering is performed to define the patterns. A ratio between compactness and contrast among clusters is used to select the best patterns. A pattern is considered significant when this ratio is statistically better than the ratios obtained for clouds of points following a Gaussian distribution. The patterns inferred for the left cingulate sulcus are consistent with the patterns described in the atlas of Ono.

Structural segmentations of brain MRI can be generated by propagating manually labelled atlas images from a repository to a query subject and combining them. This method has been shown to be robust, consistent and increasingly accurate with increasing numbers of classifiers. It outperforms standard atlas-based segmentation but suffers, however, from problems of scale when the number of atlases is large. For a large repository and a particular query subject, using a selection strategy to identify good classifiers is one way to address problems of scale. This work presents and compares different classifier selection strategies which are applied to a group of 275 subjects with manually labelled brain MR images. We approximate an upper limit for the accuracy or overlap that can be achieved for a particular structure in a given subject and compare this with the accuracy obtained using classifier selection. The accuracy of different classifier selection strategies are also rated against the distribution of overlaps generated by random groups of classifiers.

The creation of average anatomical atlases has been a growing area of research in recent years. It is of increased value to construct representations of, not only intensity atlases, but also their segmentation into required tissues or structures. This paper presents novel

groupwise combined segmentation and registration

approaches, which aim to simultaneously improve both the alignment of intensity images to their average shape, as well as the segmentations of structures in the average space. An iterative EM framework is used to build average 3D MR atlases of populations for which prior atlases do not currently exist: preterm infants at one- and two-years old. These have been used to quantify the growth of tissues occurring between these ages.

We develop a meshless method for simulating soft organ deformation. The method is motivated by simple, automatic model creation for real-time simulation. Our method is meshless in the sense that deformation is calculated at nodes that are not part of an element mesh. Node placement is almost arbitrary. Fully geometrically nonlinear total Lagrangian formulation is used. Geometric integration is performed over a regular background grid that does not conform to the simulation geometry. Explicit time integration is used via the central difference method. To validate the method we simulate indentation of a swine brain and compare the results to experimental data.

In cancer treatment, understanding the aggressiveness of the tumor is essential in therapy planning and patient follow-up. In this article, we present a novel method for quantifying the speed of invasion of gliomas in white and grey matter from time series of magnetic resonance (MR) images. The proposed approach is based on mathematical tumor growth models using the reaction-diffusion formalism. The quantification process is formulated by an inverse problem and solved using anisotropic fast marching method yielding an efficient algorithm. It is tested on a few images to get a first proof of concept with promising new results.

Interventional neuroradiology is a growing field of minimally invasive therapies that includes embolization of aneurysms and arterio-venous malformations, carotid angioplasty and carotid stenting, and acute stroke therapy. Treatment is performed using image-guided instrument navigation through the patient’s vasculature and requires intricate combination of visual and tactile coordination. In this paper we present a series of techniques for real-time high-fidelity simulation of angiographic studies. We focus in particular on the computation and visualization of blood flow and blood pressure distribution patterns, mixing of blood and contrast agent, and high-fidelity simulation of fluoroscopic images.

Needle placement into a patient body under guidance of ultrasound is a frequently performed procedure in clinical practice. Safe and successful performance of such procedure requires a high level of spatial reasoning and hand-eye co-ordination skills, which must be developed through intensive practice. In this paper we present a training system designed to improve the skills of interventional radiology trainees in ultrasound-guided needle placement procedures. Key issues involved in the system include surface and volumetric registration, solid texture modelling, spatial calibration, and real-time synthesis and rendering of ultrasound images. Moreover, soft tissue deformation caused by the needle movement and needle cutting is realised using a mass-spring-model approach. These have led to a realistic ultrasound simulation system, which has been shown to be a useful tool for the training of needle insertion procedures. Preliminary results of a construct evaluation study indicate the effectiveness and usefulness of the developed training system.

Cardiac arrhythmias are increasingly being treated using ablation procedures. Development of fast electrophysiological models and estimation of parameters related to conduction pathologies can aid in the investigation of better treatment strategies during Radio-frequency ablations. We present a fast electrophysiological model incorporating anisotropy of the cardiac tissue. A global-local estimation procedure is also outlined to estimate a hidden parameter (apparent electrical conductivity) present in the model. The proposed model is tested on synthetic and real data derived using XMR imaging. We demonstrate a qualitative match between the estimated conductivity parameter and possible pathology locations. This approach opens up possibilities to directly integrate modelling in the intervention room.

DBS for Parkinson’s disease involves an extensive planning to find a suitable electrode implantation path to the selected target. We have investigated the feasibility of improving the conventional planning with an automatic calculation of possible paths in 3D. This requires the segmentation of anatomical structures. Subsequently, the paths are calculated and visualized. After selection of a suitable path, the settings for the stereotactic frame are determined. A qualitative evaluation has shown that automatic avoidance of critical structures is feasible. The participating neurosurgeons estimate the time gain to be around 30 minutes.

In this paper we present an approach for blood vessel segmentation from dynamic contrast-enhanced MRI datasets of the hand joints acquired from patients with active rheumatoid arthritis. Exclusion of the blood vessels is needed for accurate visualisation of the activation events and objective evaluation of the degree of inflammation. The segmentation technique is based on statistical modelling motivated by the physiological properties of the individual tissues, such as speed of uptake and concentration of the contrast agent; it incorporates Markov random field probabilistic framework and principal component analysis. The algorithm was tested on 60 temporal slices and has shown promising results.

Consistency of MR scan planning is very important for diagnosis, especially in multi-site trials and follow-up studies, where disease progress or response to treatment is evaluated. Accurate manual scan planning is tedious and requires skillful operators. On the other hand, automated scan planning is difficult due to relatively low quality of survey images (“scouts”) and strict processing time constraints. This paper presents a novel method for automated planning of MRI scans of the spine. Lumbar and cervical examinations are considered, although the proposed method is extendible to other types of spine examinations, such as thoracic or total spine imaging. The automated scan planning (ASP) system consists of an anatomy recognition part, which is able to automatically detect and label the spine anatomy in the scout scan, and a planning part, which performs scan geometry planning based on recognized anatomical landmarks. A validation study demonstrates the robustness of the proposed method and its feasibility for clinical use.

In conventional CMR, bulk cardiac motion causes target structures to move in and out of the static acquisition plane. Due to the partial volume effect, accurate localisation of subtle features through the cardiac cycle, such as the trabeculae and papillary muscles, is difficult. This problem is exacerbated by the short acquisition window necessary to avoid motion blur and ghosting, especially during early systole. This paper presents an adaptive imaging approach with COMB multi-tag tracking that follows true 3D motion of the myocardium so that the same tissue slice is imaged throughout the cine acquisition. The technique is demonstrated with motion-compensated multi-slice imaging of ventricles, which allows for tracked visualisation and analysis of the trabeculae and papillary muscles for the first time. This enables novel

in-vivo

measurement of circumferential and radial strain for trabeculation and papillary muscle contractility. These statistics will facilitate the evaluation of diseases such as mitral valve insufficiency and ischemic heart disease. The adaptive imaging technique will also have significant implications for CMR in general, including motion-compensated quantification of myocardial perfusion and blood flow, and motion-correction of sequences with long acquisition windows.

Breast cancer accounts for about 30% of all cancers and 15% of all cancer deaths in women in the United States. Advances in computer assisted diagnosis (CAD) holds promise for early detecting and staging disease progression. In this paper we introduce a Grid-enabled CAD to perform automatic analysis of imaged histopathology breast tissue specimens. More than 100,000 digitized samples (1200×1200 pixels) have already been processed on the Grid. We have analyzed results for 3744 breast tissue samples, which were originated from four different institutions using diaminobenzidine (DAB) and hematoxylin staining. Both linear and nonlinear dimension reduction techniques are compared, and the best one (ISOMAP) was applied to reduce the dimensionality of the features. The experimental results show that the Gentle Boosting using an eight node CART decision tree as the weak learner provides the best result for classification. The algorithm has an accuracy of 86.02% using only 20% of the specimens as the training set.

In the context of thoracic CT-PET volume registration, we present a novel method to incorporate a breathing model in a non-linear registration procedure, guaranteeing physiologically plausible deformations. The approach also accounts for the rigid motions of lung tumors during breathing. We performed a set of registration experiments on one healthy and four pathological data sets. Initial results demonstrate the interest of this method to significantly improve the accuracy of multi-modal volume registration for diagnosis and radiotherapy applications.

For assessment of cerebrovascular diseases, it is beneficial to obtain three-dimensional (3D) information on vessel morphology and hemodynamics. Rotational angiography is routinely used to determine the 3D geometry and we propose a method to exploit the same acquisition to determine the blood flow waveform and the mean volumetric flow rate. The method uses a model of contrast agent dispersion to determine the flow parameters from the spatial and temporal development of the contrast agent concentration, represented by a flow map. Furthermore, it also overcomes artifacts due to the rotation of the c-arm using a newly introduced reliability map. The method was validated on images from a computer simulation and from a phantom experiment. With a mean error of 11.0% for the mean volumetric flow rate and 15.3% for the blood flow waveform from the phantom experiments, we conclude that the method has the potential to give quantitative estimates of blood flow parameters during cerebrovascular interventions.

In this article, we propose a framework for modeling glioma growth and the subsequent mechanical impact on the surrounding brain tissue (mass-effect) in a medical imaging context. Glioma growth is modeled via nonlinear reaction-advection-diffusion, with a two-way coupling with the underlying tissue elastic deformation. Tumor bulk and infiltration and subsequent mass-effects are not regarded separately, but captured by the model itself in the course of its evolution. Our formulation is fully Eulerian and naturally allows for updating the tumor diffusion coefficient following structural displacements caused by tumor growth/infiltration. We show that model parameters can be estimated via optimization based on imaging data, using efficient solution algorithms on regular grids. We test the model and the automatic optimization framework on real brain tumor data sets, achieving significant improvement in landmark prediction compared to a simplified purely mechanical approach.

Breast cancer detection, diagnosis and treatment increasingly involves images of the breast taken with different degrees of breast deformation. We introduce a new biomechanical modelling framework for predicting breast deformation and thus aiding the combination of information derived from the various images. In this paper, we focus on MR images of the breast under different loading conditions, and consider methods to map information between the images.

We generate subject-specific finite element models of the breast by semi-automatically fitting geometrical models to segmented data from breast MR images, and characterizing the subject-specific mechanical properties of the breast tissues. We identified the unloaded reference configuration of the breast by acquiring MR images of the breast under neutral buoyancy (immersed in water). Such imaging is clearly not practical in the clinical setting, however this previously unavailable data provides us with important data with which to validate models of breast biomechanics, and provides a common configuration with which to refer and interpret all breast images.

We demonstrate our modelling framework using a pilot study that was conducted to assess the mechanical performance of a subject-specific homogeneous biomechanical model in predicting deformations of the breast of a volunteer in a prone gravity-loaded configuration. The model captured the gross characteristics of the breast deformation with an RMS error of 4.2 mm in predicting the skin surface of the gravity-loaded shape, which included tissue displacements of over 20 mm. Internal tissue features identified from the MR images were tracked from the reference state to the prone gravity-loaded configuration with a mean error of 3.7 mm. We consider the modelling assumptions and discuss how the framework could be refined in order to further improve the tissue tracking accuracy.

This paper presents a statistical model of the liver deformation that occurs in addition to the quasi-periodic respiratory motion. Having an elastic but still compact model of this variability is an important step towards reliable targeting in radiation therapy. To build this model, the deformation of the liver at exhalation was determined for 12 volunteers over roughly one hour using 4DMRI and subsequent non-rigid registration. The correspondence between subjects was established based on mechanically relevant landmarks on the liver surface. Leave-one-out experiments were performed to evaluate the accuracy in predicting the liver deformation from partial information, such as a point tracked by ultrasound imaging. Already predictions from a single point strongly reduced the localisation errors, whilst the method is robust with respect to the exact choice of the measured predictor.

This paper focuses on the estimation of statistical atlases of 3D images by means of diffeomorphic transformations. Within a Log-Euclidean framework, the exponential and logarithm maps of diffeomorphisms need to be computed. In this framework, the Inverse Scaling and Squaring (ISS) method has been recently extended for the computation of the logarithm map, which is one of the most time demanding stages. In this work we propose to apply the Baker-Campbell-Hausdorff (BCH) formula instead. In a 3D simulation study, BCH formula and ISS method obtained similar accuracy but BCH formula was more than 100 times faster. This approach allowed us to estimate a 3D statistical brain atlas in a reasonable time, including the average and the modes of variation. Details for the computation of the modes of variation in the Sobolev tangent space of diffeomorphisms are also provided.

In this paper we present a new way of measuring brain variability based on the registration of sulcal lines sets in the large deformation framework. Lines are modelled geometrically as currents, avoiding then matchings based on point correspondences. At the end we retrieve a globally consistent deformation of the underlying brain space that best matches the lines. Thanks to this framework the measured variability is defined everywhere whereas a previous method introduced by P. Fillard requires tensors extrapolation. Evaluating both methods on the same database, we show that our new approach enables to describe different details of the variability and to highlight the major trends of deformation in the database thanks to a Tangent-PCA analysis.

In this paper, we propose a unified framework for computing atlases from manually labeled data at various degrees of “sharpness” and the joint registration-segmentation of a new brain with these atlases. In non-rigid registration, the tradeoff between warp regularization and image fidelity is typically set empirically. In segmentation, this leads to a probabilistic atlas of arbitrary “sharpness”: weak regularization results in well-aligned training images and a “sharp” atlas; strong regularization yields a “blurry” atlas. We study the effects of this tradeoff in the context of cortical surface parcellation by comparing three special cases of our framework, namely: progressive registration-segmentation of a new brain to increasingly “sharp” atlases with increasingly flexible warps; secondly, progressive registration to a single atlas with increasingly flexible warps; and thirdly, registration to a single atlas with fixed constrained warps. The optimal parcellation in all three cases corresponds to a unique balance of atlas “sharpness” and warp regularization that yield statistically significant improvements over the previously demonstrated parcellation results.

Despite being routinely required in medical applications, deformable surface registration is notoriously difficult due to large intersubject variability and complex geometry of most medical datasets. We present a general and flexible deformable matching framework based on generalized surface flows that efficiently tackles these issues through tailored deformation priors and multiresolution computations. The value of our approach over existing methods is demonstrated for automatic and user-guided cortical registration.

Clinical employment of biomechanical modelling techniques in areas of medical image analysis and surgical simulation is often hindered by conflicting requirements for high fidelity in the modelling approach and high solution speeds. We report the development of techniques for high-speed nonlinear finite element (FE) analysis for surgical simulation. We employ a previously developed nonlinear total Lagrangian explicit FE formulation which offers significant computational advantages for soft tissue simulation. However, the key contribution of the work is the presentation of a fast graphics processing unit (GPU) solution scheme for the FE equations. To the best of our knowledge this represents the first GPU implementation of a nonlinear FE solver. We show that the present explicit FE scheme is well-suited to solution via highly parallel graphics hardware, and that even a midrange GPU allows significant solution speed gains (up to 16.4×) compared with equivalent CPU implementations. For the models tested the scheme allows real-time solution of models with up to 16000 tetrahedral elements. The use of GPUs for such purposes offers a cost-effective high-performance alternative to expensive multi-CPU machines, and may have important applications in medical image analysis and surgical simulation.

A needle-tissue interaction model is an essential part of every needle insertion simulator. In this paper, a new experimental method for the modeling of needle-tissue interaction is presented. The method consists of measuring needle and tissue displacements with ultrasound, measuring needle base forces, and using a deformation simulation model to identify the parameters of a needle-tissue interaction model. The feasibility of this non-invasive approach was demonstrated in an experiment in which a brachytherapy needle was inserted into a prostate phantom. Ultrasound radio-frequency data and the time-domain cross-correlation method, often used in ultrasound elastography, were used to generate the tissue displacement field during needle insertion. A three-parameter force density model was assumed for the needle-tissue interaction. With the needle displacement, tissue displacement and needle base forces as input data, finite element simulations were carried out to adjust the model parameters to achieve a good fit between simulated and measured data.

We simulate the intravasation of liquid distention media into the systemic circulation as it occurs during hysteroscopy and transurethral resection of the prostate. A linear network flow model is extended with a correction for non-newtonian blood behaviour in small vessels and an appropriate handling of vessel compliance. We then integrate a fast lookup scheme in order to allow for real-time simulation. Cutting of tissue is accounted for by adjusting pressure boundary conditions for all cut vessels. We investigate the influence of changing distention fluid pressure settings and of the position of tissue cuts. Our simulation predicts significant intravasation only on the venous side, and just in cases when larger veins are cut. The implemented methods allow the realistic control of bleeding for short-term and the total resulting intravasation volume for long-term complication scenarios. While the simulation is fast enough to support real-time training, it is also adequate for explaining intravasation effects which were previously observed on a phenomenological level only.

The introduction of hybrid scanners has greatly increased the popularity of molecular imaging techniques. Many clinical applications benefit from combining complementary information based on the precise alignment of the two modalities. In case the alignment is inaccurate, then this crucial assumption often made for subsequent processing steps will be violated. However, this violation may not be apparent to the physician. In CT-based attenuation correction (AC) for cardiac SPECT/CT data, critical misalignments between SPECT and CT can lead to spurious perfusion defects. In this work, we focus on increasing the accuracy of rigid volume registration of cardiac SPECT/CT data by using prior knowledge. A new weighting scheme for an intensity co-occurrence prior is introduced to assure accurate and robust alignment in the local heart region. Experimental results demonstrate that the proposed method outperforms mutual information registration and shows robustness across a selection of learned distributions acquired from 15 different patients.

We developed a discrete tomography method for prostate implant reconstructions using only a limited number of X-ray projection images. A 3D voxel volume is reconstructed by back-projection and using distance maps generated from the projection images. The true seed locations are extracted from the voxel volume while false positive seeds are eliminated using a novel optimal geometry coverage model. The attractive feature of our method is that it does not require exact seed segmentation of the X-ray images and it yields near 100% correct reconstruction from only six images with an average reconstruction accuracy of 0.86 mm (std=0.46mm).

We present a new method for blind deconvolution of multiple noisy images blurred by a shift-variant point-spread-function (PSF). We focus on a setting in which several images of the same object are available, and a transformation between these images is known. This setting occurs frequently in biomedical imaging, for example in microscopy or in medical ultrasound imaging. By using the information from multiple observations, we are able to improve the quality of images blurred by a shift-variant filter,

without

prior knowledge of this filter. Also, in contrast to other work on blind and shift-variant deconvolution, in our approach no parametrization of the PSF is required. We evaluate the proposed method quantitatively on synthetically degraded data as well as qualitatively on 3D ultrasound images of liver. The algorithm yields good restoration results and proves to be robust even in presence of high noise levels in the images.

Significant research has been conducted in radiation beam gating technology to manage target and organ motions in radiotherapy treatment of cancer patients. As more and more on-board imagers are installed onto linear accelerators, fluoroscopic imaging becomes readily available at the radiation treatment stage. Thus, beam gating parameters, such as beam-on timing and beam-on window can be potentially determined by employing image registration between treatment planning CT images and fluoroscopic images. We propose a new registration method on deformable soft tissue between fluoroscopic images and DRR (Digitally Reconstructed Radiograph) images from planning CT images using active shape models. We present very promising results of our method applied to 30 clinical datasets. These preliminary results show that the method is very robust for the registration of deformable soft tissue. The proposed method can be used to determine beam-on timing and treatment window for radiation beam gating technology, and can potentially greatly improve radiation treatment quality.

Automatic classification of neurological tissues is a first step to many structural analysis pipelines. Most computational approaches are designed to extract the best possible classification results out of MR data acquired with standard clinical protocols. We observe that the characteristics of the latter owe more to the historical circumstances under which they were developed and the visual appreciation of the radiographer who acquires the images than to the optimality with which they can be classified with an automatic algorithm.

We submit that better performances could be obtained by considering the acquisition and analysis processes

conjointly

rather than optimising them independently. Here, we propose such a joint approach to MR tissue classification in the form of a fast MR sequence, which nulls the magnitude and changes the sign of the phase at the boundary between tissue types. A simple phase-based thresholding algorithm then suffices to segment the tissues. Preliminary results show promises to simplify and shorten the overall classification process.

We present a new functional magnetic resonance imaging (fMRI) analysis method that incorporates both spatial and temporal dynamics of blood-oxygen-level dependent (BOLD) signals within a region of interest (ROI). 3D moment descriptors are used to characterize the spatial changes in BOLD signals over time. The method is tested on fMRI data collected from eight healthy subjects performing a bulb-squeezing motor task with their right-hand at various frequencies. Multiple brain regions including the left cerebellum, both primary motor cortices (M1), both supplementary motor areas (SMA), left prefrontal cortex (PFC), and left anterior cingulate cortex (ACC) demonstrate significant task-related changes. Furthermore, our method is able to discriminate differences in activation patterns at the various task frequencies, whereas using a traditional intensity based method, no significant activation difference is detected. This suggests that temporal dynamics of the spatial distribution of BOLD signal provide additional information regarding task-related activation thus complementing conventional intensity-based approaches.

This paper describes a method for automatic detection of contractions in the small bowel through analyzing Wireless Capsule Endoscopic images. Based on the characteristics of contraction images, a coherent procedure that includes analyzes of the temporal and spatial features is proposed. For temporal features, the image sequence is examined to detect candidate contractions through the changing number of edges and an evaluation of similarities between the frames of each possible contraction to eliminate cases of low probability. For spatial features, descriptions of the directions at the edge pixels are used to determine contractions utilizing a classification method. The experimental results show the effectiveness of our method that can detect a total of 83% of cases. Thus, this is a feasible method for developing tools to assist in diagnostic procedures in the small bowel.

Computed tomography (CT) images of the lungs provide high resolution views of the airways. Quantitative measurements such as lumen diameter and wall thickness help diagnose and localize airway diseases, assist in surgical planning, and determine progress of treatment. Automated quantitative analysis of such images is needed due to the number of airways per patient. We present an approach involving dynamic programming coupled with boundary-specific cost functions that is capable of differentiating inner and outer borders. The method allows for precise delineation of the inner lumen and outer wall. The results are demonstrated on synthetic data, evaluated on human datasets compared to human operators, and verified on phantom CT scans to sub-voxel accuracy.

Dry Eye Syndrome is a common disease in the western world, with effects from uncomfortable itchiness to permanent damage to the ocular surface. Nevertheless, there is still no objective test that provides reliable results. We have developed a new method for the automated detection of dry areas in videos taken after instilling fluorescein in the tear film. The method consists of a multi-step algorithm to first locate the iris in each image, then align the images and finally analyze the aligned sequence in order to find the regions of interest. Since the fluorescein spreads on the ocular surface of the eye the edges of the iris are fuzzy making the detection of the iris challenging. We use RANSAC to first detect the upper and lower eyelids and then the iris. Then we align the images by finding differences in intensities at different scales and using a least squares optimization method (Levenberg-Marquardt), to overcome the movement of the iris and the camera. The method has been tested on videos taken from different patients. It is demonstrated to find the dry areas accurately and to provide a measure of the extent of the disease.

Ultrasound Myocardial Elastography (UME) and Tagged Magnetic Resonance Imaging (tMRI) are two imaging modalities that were developed in the recent years to quantitatively estimate the myocardial deformations. Tagged MRI is currently considered as the gold standard for myocardial strain mapping in vivo. However, despite the low SNR nature of ultrasound signals, echocardiography enjoys the wides- pread availability in the clinic, as well as its low cost and high temporal resolution. Comparing the strain estimation performances of the two techniques has been of great interests to the community. In order to assess the cardiac deformation across different imaging modalities, in this paper, we developed a semi-automatic intensity and gradient based registration framework that rigidly registers the 3D tagged MRIs with the 2D ultrasound images. Based on the two registered modalities, we conducted spatially and temporally more detailed quantitative strain comparison of the RF-based UME technique and tagged MRI. From the experimental results, we conclude that qualitatively the two modalities share similar overall trends. But error and variations in UME accumulate over time. Quantitatively tMRI is more robust and accurate than UME.

A major challenge in contemporary magnetic resonance imaging (MRI) lies in providing the highest resolution exam possible in the shortest acquisition period. Recently, several authors have proposed the use of L

1

-norm minimization for the reconstruction of sparse MR images from highly-undersampled k-space data. Despite promising results demonstrating the ability to accurately reconstruct images sampled at rates significantly below the Nyquist criterion, the extensive computational complexity associated with the existing framework limits its clinical practicality. In this work, we propose an alternative recovery framework based on homotopic approximation of the L

0

-norm and extend the reconstruction problem to a multiscale formulation. In addition to several interesting theoretical properties, practical implementation of this technique effectively resorts to a simple iterative alternation between bilteral filtering and projection of the measured k-space sample set that can be computed in a matter of seconds on a standard PC.

Assessing the quality of motion estimation in the lung remains challenging. We approach the problem by imaging isolated porcine lungs within an artificial thorax with four-dimensional computed tomography (4DCT). Respiratory kinematics are estimated via pairwise non-rigid registration using different metrics and image resolutions. Landmarks are manually identified on the images and used to assess accuracy by comparing known displacements to the registration-derived displacements. We find that motion quantitation becomes less precise as the inflation interval between images increases. In addition, its sensitivity to image resolution varies anatomically. Mutual information and cross-correlation perform similarly, while mean squares is significantly poorer. However, none of the metrics compensate for the difficulty of registering over a large inflation interval. We intend to use the results of these experiments to more effectively and efficiently quantify pulmonary kinematics in future, and to explore additional parameter combinations.

Characterization and quantification of diffuse parenchymal lung disease (DPLD) severity using MDCT, mainly in interstitial lung diseases and emphysema, is an important issue in clinical research for the evaluation of new therapies. This paper develops a 3D automated approach for detection and diagnosis of DPLDs (emphysema, fibrosis, honeycombing, ground glass).The proposed methodology combines multi-resolution image decomposition based on 3D morphological filtering, and graph-based classification for a full characterization of the parenchymal tissue. The very promising results obtained on a small patient database are good premises for a near implementation and validation of the proposed approach in clinical routine.

In this paper, we present an unsupervised 2D/3D reconstruction scheme combining a parameterized multiple-component geometrical model and a point distribution model, and show its application to automatically reconstruct a surface model of a proximal femur from a limited number of calibrated fluoroscopic images with no user intervention at all. The parameterized multiple-component geometrical model is regarded as a simplified description capturing the geometrical features of a proximal femur. Its parameters are optimally and automatically estimated from the input images using a particle filter based inference method. The estimated geometrical parameters are then used to initialize a point distribution model based 2D/3D reconstruction scheme for an accurate reconstruction of a surface model of the proximal femur. We designed and conducted

in vitro

and

in vivo

experiments to compare the present unsupervised reconstruction scheme to a supervised one. An average mean error of 1.2 mm was found when the supervised reconstruction scheme was used. It increased to 1.3 mm when the unsupervised one was used. However, the unsupervised reconstruction scheme has the advantage of elimination of user intervention, which holds the potential to facilitate the application of the 2D/3D reconstruction in surgical navigation.

A novel method called local shape controlled voting has been developed for spherical object detection in 3D voxel images. By introducing local shape properties into the voting procedure of normal overlap, the proposed method improves the capability of differentiating spherical objects from other structures, as the normal overlap technique only measures the ‘density’ of normal overlapping, while how the normals are distributed in 3D is not discovered. The proposed method was applied to computer aided detection of pulmonary nodules based on helical CT images. Experiments showed that this method attained a better performance compared to the original normal overlap technique.

This paper proposes to use the Laplace-Beltrami spectrum (LBS) as a global shape descriptor for medical shape analysis, allowing for shape comparisons using minimal shape preprocessing: no registration, mapping, or remeshing is necessary. The discriminatory power of the method is tested on a population of female caudate shapes of normal control subjects and of subjects with schizotypal personality disorder.

In this paper we present a framework for real-time tracking of deformable contours in volumetric datasets. The framework supports composite deformation models, controlled by parameters for contour shape in addition to global pose. Tracking is performed in a sequential state estimation fashion, using an extended Kalman filter, with measurement processing in information space to effectively predict and update contour deformations in real-time. A deformable B-spline surface coupled with a global pose transform is used to model shape changes of the left ventricle of the heart.

Successful tracking of global motion and local shape changes without user intervention is demonstrated on a dataset consisting of 21 3D echocardiography recordings. Real-time tracking using the proposed approach requires a modest CPU load of 13% on a modern computer. The segmented volumes compare to a semi-automatic segmentation tool with 95% limits of agreement in the interval 4.1 ±24.6 ml (

r

 = 0.92).

Segmentation of vessels and brain aneurysms on non-invasive and flow-sensitive phase contrast magnetic resonance angiographic (PCMRA) images is essential in the detection of vascular diseases, in particular, intracranial aneurysms. In this paper, we devise a novel method based on multi-range filters and local variances to perform segmentation of vessels and intracranial aneurysms on PCMRA images. The proposed method is validated and compared using a synthetic and numerical image volume and four clinical cases. It is experimentally shown that the proposed method is capable of segmenting vessels and aneurysms with various sizes on PCMRA images.

The segmentation of macroscopically ill-defined and highly variable structures, such as the hippocampus

Hc

and the amygdala

Am

, from MRI requires specific constraints. Here, we describe and evaluate a hybrid segmentation method that uses knowledge derived from a probabilistic atlas and from anatomical landmarks based on stable anatomical characteristics of the structures. Combined in a previously published semi-automatic segmentation method, they lead to a fast, robust and accurate fully automatic segmentation of

Hc

and

Am

. The probabilistic atlas was built from 16 young controls and registered with the ”unified segmentation” of SPM5. The algorithm was quantitatively evaluated with respect to manual segmentation on two MRI datasets: the 16 young controls, with a leave-one-out strategy, and a mixed cohort of 8 controls and 15 subjects with epilepsy with variable hippocampal sclerosis. The segmentation driven by hybrid knowledge leads to greatly improved results compared to that obtained by registration of the thresholded atlas alone: mean overlap for

Hc

on the 16 young controls increased from 78% to 87% (

p

 < 0.001) and on the mixed cohort from 73% to 82% (

p

 < 0.001) while the error on volumes decreased from 10% to 7% (

p

 < 0.005) and from 18% to 8% (

p

 < 0.001), respectively. Automatic results were better than the semi-automatic results: for the 16 young controls, average overlap increased from 84% to 87% (

p

 < 0.001) for

Hc

and from 81% to 84% (

p

 < 0.002) for

Am

, with equivalent improvements in volume error.

This paper presents a Bayesian framework for neonatal brain-tissue segmentation in

clinical

magnetic resonance (MR) images. This is a challenging task because of the low contrast-to-noise ratio and large variance in both tissue intensities and brain structures, as well as imaging artifacts and partial-volume effects in clinical neonatal scanning. We propose to incorporate a spatially adaptive likelihood model using a data-driven nonparametric statistical technique. The method initially learns an

intensity-based prior

, relying on the empirical Markov statistics from training data, using fuzzy nonlinear support vector machines (SVM). In an iterative scheme, the models adapt to spatial variations of image intensities via nonparametric density estimation. The method is effective even in the absence of anatomical atlas priors. The implementation, however, can naturally incorporate probabilistic atlas priors and Markov-smoothness priors to impose additional regularity on segmentation. The maximum-a-posteriori (MAP) segmentation is obtained within a graph-cut framework. Cross validation on clinical neonatal brain-MR images demonstrates the efficacy of the proposed method, both qualitatively and quantitatively.

We introduce a non-linear shape prior for the deformable model framework that we learn from a set of shape samples using recent manifold learning techniques. We model a category of shapes as a finite dimensional manifold which we approximate using Diffusion maps. Our method computes a Delaunay triangulation of the reduced space, considered as Euclidean, and uses the resulting space partition to identify the closest neighbors of any given shape based on its Nyström extension. We derive a non-linear shape prior term designed to attract a shape towards the shape prior manifold at given constant embedding. Results on shapes of ventricle nuclei demonstrate the potential of our method for segmentation tasks.

In medical image analysis there is frequently a need to invert dense displacement fields which map one image space to another. In this paper we describe inversion techniques and determine their accuracy in the context of 18 inter-subject brain image registrations. Scattered data interpolation (SDI) is used to initialise locally and globally consistent iterative techniques. The inverse-consistency error,

E

IC

is computed over the whole image and over 10 specific brain regions. SDI produced good results with mean (max)

E

IC

~0.02mm (2.0mm). Both iterative method produced mean errors of ~0.005mm but the globally consistent method resulted in a smaller maximum error (1.9mm compared with 1.4mm). The largest errors were in the cerebral cortex with large outlier errors in the ventricles. Simple iterative techniques are, on this evidence, able to produce reasonable estimates of inverse displacement fields provided there is good initialisation.

Registration of Diffusion Weighted (DW)-MRI datasets has been commonly achieved to date in literature by using either scalar or 2

nd

-order tensorial information. However, scalar or 2

nd

-order tensors fail to capture complex local tissue structures, such as fiber crossings, and therefore, datasets containing fiber-crossings cannot be registered accurately by using these techniques. In this paper we present a novel method for non-rigidly registering DW-MRI datasets that are represented by a field of 4

th

-order tensors. We use the Hellinger distance between the normalized 4

th

-order tensors represented as distributions, in order to achieve this registration. Hellinger distance is easy to compute, is scale and rotation invariant and hence allows for comparison of the true shape of distributions. Furthermore, we propose a novel 4

th

-order tensor re-transformation operator, which plays an essential role in the registration procedure and shows significantly better performance compared to the re-orientation operator used in literature for DTI registration. We validate and compare our technique with other existing scalar image and DTI registration methods using simulated diffusion MR data and real HARDI datasets.

Non-rigid image registration is an ill-posed yet challenging problem due to its supernormal high degree of freedoms and inherent requirement of smoothness. Graph-cuts method is a powerful combinatorial optimization tool which has been successfully applied into image segmentation and stereo matching. Under some specific constraints, graph-cuts method yields either a global minimum or a local minimum in a strong sense. Thus, it is interesting to see the effects of using graph-cuts in non-rigid image registration. In this paper, we formulate non-rigid image registration as a discrete labeling problem. Each pixel in the source image is assigned a displacement label (which is a vector) indicating which position in the floating image it is spatially corresponding to. A smoothness constraint based on first derivative is used to penalize sharp changes in displacement labels across pixels. The whole system can be optimized by using the graph-cuts method via alpha-expansions. We compare 2D and 3D registration results of our method with two state-of-the-art approaches. It is found that our method is more robust to different challenging non-rigid registration cases with higher registration accuracy.

Recent developments in freehand 3D ultrasound (US) have shown how image registration and speckle decorrelation methods can be used for 3D reconstruction instead of relying on a tracking device. Estimating elevational separation between untracked US images using speckle decorrelation is error prone due to the uncertainty that plagues the correlation measurements. In this paper, using maximum entropy estimation methods, the uncertainty is directly modeled from the calibration data normally used to estimate an average decorrelation curve. Multiple correlation measurements can then be fused within a maximum likelihood estimation framework in order to reduce the drift in elevational pose estimation over large image sequences. The approach is shown to be effective through empirical results on simulated and phantom US data.

Dynamic enhancement causes serious problems for registration of contrast enhanced breast MRI, due to variable uptakes of agent on different tissues or even same tissues in the breast. We present an iterative optimization algorithm to de-enhance the dynamic contrast-enhanced breast MRI and then register them for avoiding the effects of enhancement on image registration. In particular, the spatially varying enhancements are modeled by a Markov Random Field, and estimated by a locally smooth function with boundaries using a graph cut algorithm. The de-enhanced images are then registered by conventional B-spline based registration algorithm. These two steps benefit from each other and are repeated until the results converge. Experimental results show that our two-step registration algorithm performs much better than conventional mutual information based registration algorithm. Also, the effects of tumor shrinking in the conventional registration algorithms can be effectively avoided by our registration algorithm.

There exists a large body of literature on shape matching and registration in medical image analysis. However, most of the previous work is focused on matching particular sets of features—point-sets, lines, curves and surfaces. In this work, we forsake specific geometric shape representations and instead seek probabilistic representations—specifically Gaussian mixture models—of shapes. We evaluate a closed-form distance between two probabilistic shape representations for the general case where the mixture models differ in variance and the number of components. We then cast non-rigid registration as a deformable density matching problem. In our approach, we take one mixture density onto another by deforming the component centroids via a thin-plate spline (TPS) and also minimizing the distance with respect to the variance parameters. We validate our approach on synthetic and 3D arterial tree data and evaluate it on 3D hippocampal shapes.

We present a new method for the fast and robust computation of information theoretic similarity measures for alignment of multi-modality medical images. The proposed method defines a non-uniform, adaptive sampling scheme for estimating the entropies of the images, which is less vulnerable to local maxima as compared to uniform and random sampling. The sampling is defined using an octree partition of the template image, and is preferable over other proposed methods of non-uniform sampling since it respects the underlying data distribution. It also extends naturally to a multi-resolution registration approach, which is commonly employed in the alignment of medical images. The effectiveness of the proposed method is demonstrated using both simulated MR images obtained from the BrainWeb database and clinical CT and SPECT images.

A fundamental problem when computing statistical shape models is the determination of correspondences between the instances of the associated data set. Often, homologies between points that represent the surfaces are assumed which might lead to imprecise mean shape and variability results. We propose an approach where exact correspondences are replaced by evolving correspondence probabilities. These are the basis for a novel algorithm that computes a generative statistical shape model. We developed an unified MAP framework to compute the model parameters (’mean shape’ and ’modes of variation’) and the nuisance parameters which leads to an optimal adaption of the model to the set of observations. The registration of the model on the instances is solved using the Expectation Maximization - Iterative Closest Point algorithm which is based on probabilistic correspondences and proved to be robust and fast. The alternated optimization of the MAP explanation with respect to the observation and the generative model parameters leads to very efficient and closed-form solutions for (almost) all parameters. Experimental results on brain structure data sets demonstrate the efficiency and well-posedness of the approach. The algorithm is then extended to an automatic classification method using the k-means clustering and applied to synthetic data as well as brain structure classification problems.

In this paper we propose a weakly supervised learning algorithm for appearance models based on the minimum description length (MDL) principle. From a set of training images or volumes depicting examples of an anatomical structure, correspondences for a set of landmarks are established by group-wise registration. The approach does not require any annotation. In contrast to existing methods no assumptions about the topology of the data are made, and the topology can change throughout the data set. Instead of a continuous representation of the volumes or images, only sparse finite sets of interest points are used to represent the examples during optimization. This enables the algorithm to efficiently use distinctive points, and to handle texture variations robustly. In contrast to standard elasticity based deformation constraints the MDL criterion accounts for systematic deformations typical for training sets stemming from medical image data. Experimental results are reported for five different 2D and 3D data sets.

In this paper, we present a new segmentation evaluation method that can simulate radiologist’s subjective assessment of 3D tumor segmentation in CT images. The method uses a new metric defined as a linear combination of a set of commonly used objective metrics. The weighing parameters of the linear combination are determined by maximizing the rank correlation between radiologist’s subjective rating and objective measurements. Experimental results on 93 lesions demonstrate that the new composite metric shows better performance in segmentation evaluation than each individual objective metric. Also, segmentation rating using the composite metric compares well with radiologist’s subjective evaluation. Our method has the potential to facilitate the development of new tumor segmentation algorithms and assist large scale segmentation evaluation studies.

We propose an extended graph-shifts algorithm for image segmentation and labeling. This algorithm performs energy minimization by manipulating a dynamic hierarchical representation of the image. It consists of a set of moves occurring at different levels of the hierarchy where the types of move, and the level of the hierarchy, are chosen automatically so as to maximally decrease the energy. Extended graph-shifts can be applied to a broad range of problems in medical imaging. In this paper, we apply extended graph-shifts to the detection of pathological brain structures: (i) segmentation of brain tumors, and (ii) detection of multiple sclerosis lesions. The energy terms in these tasks are learned from training data by statistical learning algorithms. We demonstrate accurate results, precision and recall in the order of 93%, and also show that the algorithm is computationally efficient, segmenting a full 3D volume in about one minute.

Minimally invasive surgery in orthopedic field is considered to be a challenging problem with a milling robot. One objective of this study is to minimize collision of the cutting tool with soft tissue. The authors have developed a robot with redundant axis to avoid the collision so far. Some important components are modeled based on physical requirements, and a geometric optimization approach based on the model has been also proposed to improve performance. In this paper, a protective mechanism to cover the non-working part of the cutting edge is proposed to avoid soft tissue damage. Hardware and software have been developed for this application and the effectiveness of this technique was evaluated with urethane bone.