Medical Image Computing and Computer-Assisted Intervention – MICCAI 2010

This paper presents a new diffeomorphic temporal registration algorithm and its application to motion and strain quantification from a temporal sequence of 3D images. The displacement field is computed by forward eulerian integration of a non-stationary velocity field. The originality of our approach resides in enforcing time consistency by representing the velocity field as a sum of continuous spatiotemporal BSpline kernels. The accuracy of the developed diffeomorphic technique was first compared to a simple pairwise strategy on synthetic US images with known ground truth motion and with several noise levels, being the proposed algorithm more robust to noise than the pairwise case. Our algorithm was then applied to a database of cardiac 3D+t Ultrasound (US) images of the left ventricle acquired from eight healthy volunteers and three Cardiac Resynchronization Therapy (CRT) patients. On healthy cases, the measured regional strain curves provided uniform strain patterns over all myocardial segments in accordance with clinical literature. On CRT patients, the obtained normalization of the strain pattern after CRT agreed with clinical outcome for the three cases.

This paper presents a robust framework for freehand ultrasound elastography to cope with uncertainties of freehand palpation using the information from an external tracker. In order to improve the quality of the elasticity images, the proposed method selects a few image pairs such that in each pair the lateral and out-of-plane motions are minimized. It controls the strain rate by choosing the axial motion to be close to a given optimum value. The tracking data also enables fusing multiple strain images that are taken roughly from the same location. This method can be adopted for various trackers and strain estimation algorithms. In this work, we show the results for two tracking systems of electromagnetic (EM) and optical tracker. Using phantom and

ex-vivo

animal experiments, we show that the proposed techniques significantly improve the elasticity images and reduce the dependency to the hand motion of user.

To ensure accurate targeting and repeatability, 3D TRUS-guided biopsies require registration to determine coordinate transformations to (1) incorporate pre-procedure biopsy plans and (2) compensate for

inter-session

prostate motion and deformation between repeat biopsy sessions. We evaluated prostate surface- and image-based 3D-to-3D TRUS registration by measuring the TRE of manually marked, corresponding, intrinsic fiducials in the whole gland and peripheral zone, and also evaluated the error anisotropy. The image-based rigid and non-rigid methods yielded the best results with mean TREs of 2.26 mm and 1.96 mm, respectively. These results compare favorably with a clinical need for an error of less than 2.5 mm.

Breathing motion leads to a significant displacement and deformation of organs in the abdominal region. This makes the detection of the breathing phase for numerous applications necessary. We propose a new, purely image-based respiratory gating method for ultrasound. Further, we use this technique to provide a solution for breathing affected 4D ultrasound acquisitions with a wobbler probe. We achieve the gating with Laplacian eigenmaps, a manifold learning technique, to determine the low-dimensional manifold embedded in the high-dimensional image space. Since Laplacian eigenmaps assign each ultrasound frame a coordinate in low-dimensional space by respecting the neighborhood relationship, they are well suited for analyzing the breathing cycle. For the 4D application, we perform the manifold learning for each angle, and consecutively, align all the local curves and perform a curve fitting to achieve a globally consistent breathing signal. We performed the image-based gating on several 2D and 3D ultrasound datasets over time, and quantified its very good performance by comparing it to measurements from an external gating system.

Vibration-controlled transient elastography (VCTE

TM

) technique is routinely used in clinical practice to assess non-invasively the liver stiffness which is correlated to hepatic fibrosis. Adequate use of the VCTE

TM

probe requires the knowledge of the distance between the skin and the liver parenchyma. This paper compares two methods to estimate this distance using spatial variations of the spectral content of ultrasound radiofrequency (RF) lines, obtained from a probe consisting of a single element ultrasound transducer placed in front of the liver right lobe. Results on a database of 188 patients, including normal-weight and obese persons, show that the spectral variance can accurately discriminate the subcutaneous fat from the liver tissue. The proposed algorithm works in real-time and is suitable for VCTE

TM

scanning protocol setup.

Automatic delineation of the prostate boundary in transrectal ultrasound (TRUS) can play a key role in image-guided prostate intervention. However, it is a very challenging task for several reasons, especially due to the large variation of the prostate shape from the base to the apex. To deal with the problem, a new method for incrementally learning the patient-specific local shape statistics is proposed in this paper to help achieve robust and accurate boundary delineation over the entire prostate gland. The proposed method is fast and memory efficient in that new shapes can be merged into the shape statistics without recomputing using all the training shapes, which makes it suitable for use in real-time interventional applications. In our work, the learned shape statistics is incorporated into a modified sequential inference model for tracking the prostate boundary. Experimental results show that the proposed method is more robust and accurate than the active shape model using global population-based shape statistics in delineating the prostate boundary in TRUS.

Visceral fat area (VFA) has close relationship with hypertension, diabetes and cardiovascular disease, and therefore serve as a reliable indicator of these diseases. Abdominal computed tomography (CT) enables precise quantification of the VFA and has been considered as the gold standard for VFA assessment. In this paper, we develope a novel method to quickly and accurately measure the VFA with ultrasonography (US). We evaluated the novel method on five volunteers and the diagnosis procedures lasted less than 30 seconds averagely. The simulation results by our method were compared with VFA estimated by abdominal CT. The correlation coefficient between them was 0.913 for men and 0.858 for women. And the mean deviation of between VFA by CT and by our method was 19.8

cm

2

for men and 13.3

cm

2

for women.

Intravascular Ultrasound (IVUS) is an image-guiding technique for cardiovascular diagnostic, providing cross-sectional images of vessels. During the acquisition, the catheter is pulled back (pullback) at a constant speed in order to acquire spatially subsequent images of the artery. However, during this procedure, the heart twist produces a swinging fluctuation of the probe position along the vessel axis. In this paper we propose a real-time gating algorithm based on the analysis of

motion blur

variations during the IVUS sequence. Quantitative tests performed on an in-vitro ground truth data base shown that our method is superior to state of the art algorithms both in computational speed and accuracy.

Motivation:

Spinal needle injections are technically demanding procedures. The use of ultrasound image guidance without prior CT and MR imagery promises to improve the efficacy and safety of these procedures in an affordable manner.

Methodology:

We propose to create a statistical shape model of the lumbar spine and warp this atlas to patient-specific ultrasound images during the needle placement procedure. From CT image volumes of 35 patients, statistical shape model of the L3 vertebra is built, including mean shape and main modes of variation. This shape model is registered to the ultrasound data by simultaneously optimizing the parameters of the model and its relative pose. Ground-truth data was established by printing 3D anatomical models of 3 patients using a rapid prototyping. CT and ultrasound data of these models were registered using fiducial markers.

Results:

Pairwise registration of the statistical shape model and 3D ultrasound images led to a mean target registration error of 3.4 mm, while 81% of all cases yielded clinically acceptable accuracy below the 3.5 mm threshold.

In this paper we propose a fully automatic 2D prostate segmentation algorithm using fused ultrasound (US) and elastography images. We show that the addition of information from mechanical tissue properties acquired from elastography to acoustic information from B-mode ultrasound, can improve segmentation results. Gray level edge similarity and edge continuity in both US and elastography images deform an Active Shape Model. Comparison of automatic and manual contours on 107 transverse images of the prostate show a mean absolute error of 2.6 ±0.9 mm and a running time of 17.9 ±12.2 s. These results show that the combination of the high contrast elastography images with the more detailed but low contrast US images can lead to very promising results for developing an automatic 3D segmentation algorithm.

By revealing complex fiber structure through the orientation distribution function (ODF), q-ball imaging has recently become a popular reconstruction technique in diffusion-weighted MRI. In this paper, we propose an analytical dimension reduction approach to ODF maxima extraction. We show that by expressing the ODF, or any antipodally symmetric spherical function, in the common fourth order real and symmetric spherical harmonic basis, the maxima of the two-dimensional ODF lie on an analytically derived one-dimensional space, from which we can detect the ODF maxima. This method reduces the computational complexity of the maxima detection, without compromising the accuracy. We demonstrate the performance of our technique on both artificial and human brain data.

We propose a method for the creation of an anatomically and mechanically realistic brain phantom from polyvinyl alcohol cryogel (PVA-C) for validation of image processing methods for segmentation, reconstruction, registration, and denoising. PVA-C is material widely used in medical imaging phantoms for its mechanical similarities to soft tissues. The phantom was cast in a mold designed using the left hemiphere of the Colin27 brain dataset [1] and contains deep sulci, a complete insular region, and an anatomically accurate left ventricle. Marker spheres and inflatable catheters were also implanted to enable good registration and simulate tissue deformation, respectively. The phantom was designed for triple modality imaging, giving good contrast images in computed tomography, ultrasound, and magnetic resonance imaging. Multimodal data acquired from this phantom are made freely available to the image processing community (

http://pvabrain.inria.fr

) and will aid in the validation and further development of medical image processing techniques.

We present a framework for statistical analysis in large cohorts of structural brain connectivity, derived from diffusion weighted MRI. A brain network is defined between subcortical gray matter structures and a cortical parcellation obtained with FreeSurfer. Connectivity is established through minimum cost paths with an anisotropic local cost function and is quantified per connection. The connectivity network potentially encodes important information about brain structure, and can be analyzed using multivariate regression methods. The proposed framework can be used to study the relation between connectivity and e.g. brain function or neurodegenerative disease. As a proof of principle, we perform principal component regression in order to predict age and gender, based on the connectivity networks of 979 middle-aged and elderly subjects, in a 10-fold cross-validation. The results are compared to predictions based on fractional anisotropy and mean diffusivity averaged over the white matter and over the corpus callosum. Additionally, the predictions are performed based on the best predicting connection in the network. Principal component regression outperformed all other prediction models, demonstrating the age and gender information encoded in the connectivity network.

Thick-slice image acquisitions are sometimes inevitable in magnetic resonance imaging due to limitations posed by pulse sequence timing and signal-to-noise-ratio. The estimation of an isotropic high-resolution volume from thick-slice MRI scans is desired for improved image analysis and evaluation. In this article we formulate a maximum a posteriori (MAP) estimation algorithm for high-resolution volumetric MRI reconstruction. As compared to the previous techniques, this probabilistic formulation relies on a slice acquisition model and allows the incorporation of image priors. We focus on image priors based on image gradients and compare the developed MAP estimation approach to scattered data interpolation (SDI) and maximum likelihood reconstruction. The results indicate that the developed MAP estimation approach outperforms the SDI techniques and appropriate image priors may improve the volume estimation when the acquired thick-slice scans do not sufficiently sample the imaged volume. We also report applications in pediatric and fetal imaging.

This paper presents a longitudinal change detection framework for detecting relevant modifications in diffusion MRI, with application to Multiple Sclerosis (MS). The proposed method is based on multivariate statistical testings which were initially introduced for tensor population comparison. We use these methods in the context of longitudinal change detection by considering several strategies to build sets of tensors characterizing the variability of each voxel. These testing tools have been considered either for the comparison of tensor eigenvalues or eigenvectors, thus enabling to differentiate orientation and diffusivity changes. Results on simulated MS lesion evolutions and on real data are presented. Interestingly, experiments on an MS patient highlight the ability of the proposed approach to detect changes in non evolving lesions (according to conventional MRI) and around lesions (in the normal appearing white matter), which might open promising perspectives for the follow-up of the MS pathology.

Identifying mild cognitive impairment (MCI) subjects who will convert to clinical Alzheimer’s disease (AD) is important for therapeutic decisions, patient counselling and clinical trials. Hippocampal volume and rate of atrophy predict clinical decline at the MCI stage and progression to AD. In this paper, we create

p

-maps from the differences in the shape of the hippocampus between 60 normal controls and 60 AD subjects using statistical shape models, and generate different regions of interest (ROI) by thresholding the

p

-maps at different significance levels. We demonstrate increased statistical power to classify 86 MCI converters and 128 MCI stable subjects using the hippocampal atrophy rates calculated by the boundary shift integral within these ROIs.

Accurate and reliable method for measuring the thickness of human cerebral cortex provides powerful tool for diagnosing and studying of a variety of neuro-degenerative and psychiatric disorders. In these studies, capturing the

subtle

longitudinal changes of cortical thickness during pathological or physiological development is of great importance. For this purpose, in this paper, we propose a 4D cortical thickness measuring method. Different from the existing temporal-independent methods, our method fully utilizes the 4D information given by temporal serial images. Therefore, it is much more resistant to noises from the imaging and pre-processing steps. The experiments on longitudinal image datasets from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) show that our method significantly improves the longitudinal stability,

i.e.

temporal consistency, in cortical thickness measurement, which is crucial for longitudinal study. Power analysis of the correlation between cortical thickness and Mini-Mental-Status-Examination (MMSE) score demonstrated that our method generates statistically more significant results when comparing with the 3D temporal-independent thickness measuring methods.

Recently, inference of functional connectivity between brain regions using resting state fMRI (rsfMRI) data has attracted significant interests in the neuroscience community. This paper proposes a novel fiber-centered approach to study the functional connectivity between brain regions using high spatial resolution diffusion tensor imaging (DTI) and rsfMRI data. We measure the functional coherence of a fiber as the time series’ correlation of two gray matter voxels that this fiber connects. The functional connectivity strength between two brain regions is defined as the average functional coherence of fibers connecting them. Our results demonstrate that: 1) The functional coherence of fibers is correlated with the brain regions they connect; 2) The functional connectivity between brain regions is correlated with structural connectivity. And these two patterns are consistent across subjects. These results may provide new insights into the brain’s structural and functional architecture.

We introduce a generative probabilistic model for segmentation of tumors in multi-dimensional images. The model allows for different tumor boundaries in each channel, reflecting difference in tumor appearance across modalities. We augment a probabilistic atlas of healthy tissue priors with a latent atlas of the lesion and derive the estimation algorithm to extract tumor boundaries and the latent atlas from the image data. We present experiments on 25 glioma patient data sets, demonstrating significant improvement over the traditional multivariate tumor segmentation.

A rapidly increasing number of medical imaging studies is longitudinal, i.e. involves series of repeated examinations of the same individuals. This paper presents a methodology for analysis of such 4D images, with brain aging as the primary application. An adaptive regional clustering method is first adopted to construct a spatial pattern, in which a measure of correlation between morphological measurements and a continuous patient’s variable (age in our case) is used to group brain voxels into regions; Secondly, a dynamic probabilistic Hidden Markov Model (HMM) is created to statistically analyze the relationship between spatial brain patterns and hidden states; Thirdly, parametric HMM models under a bagging framework are used to capture the changes occurring with time by decoding the hidden states longitudinally. We apply this method to datasets from elderly individuals, and test the effectiveness of this spatio-temporal model in analyzing the temporal dynamics of spatial aging patterns on an individual basis. Experimental results show this method could facilitate the early detection of pathological brain change.

Cortical surface complexity is a potential structural marker for certain diseases such as schizophrenia. In this study, we developed a measure of fractal dimension (FD) calculated from lowpass-filtered spherical harmonic brain surface reconstructions. A local FD measure was also computed at each vertex in a cortical surface mesh, visualizing local variations in surface complexity over the brain surface. We analyzed the surface complexity for 87 patients with DSM-IV schizophrenia (with stable psychopathology and treated with antipsychotic medication) and 108 matched healthy controls. The global FD for the right hemisphere in the schizophrenic group was significantly lower than that in controls. Local FD maps showed that the lower complexity was mainly due to differences in the prefrontal cortex.

The paper proposes a new shape morphometry approach that combines advanced classification techniques with geometric features to identify morphological abnormalities on the brain surface. Our aim is to improve the classification accuracy in distinguishing between normal subjects and schizophrenic patients. The approach is inspired by natural language processing. Local brain surface geometric patterns are quantized to

visual words

, and their co-occurrences are encoded as

visual topic

. To do this, a generative model, the probabilistic Latent Semantic Analysis is learned from quantized shape descriptors (visual words). Finally, we extract from the learned models a generative score, that is used as input of a Support Vector Machine (SVM), defining an hybrid generative/discriminative classification algorithm. An exhaustive experimental section is proposed on a dataset consisting of MRI scans from 64 patients and 60 control subjects. Promising results are reporting by observing accuracies up to 86.13%.

Kinetic analysis is an essential tool of Positron Emission Tomography image analysis. However it requires a pure tissue time activity curve (TAC) in order to calculate the system parameters. Pure tissue TACs are particularly difficult to obtain in the brain as the low resolution of PET means almost all voxels are a mixture of tissues. Factor analysis explicitly accounts for mixing but is an underdetermined problem that can give arbitrary results. A joint factor and kinetic analysis is proposed whereby factor analysis explicitly accounts for mixing of tissues. Hence, more meaningful parameters are obtained by the kinetic models, which also ensure a less ambiguous solution to the factor analysis. The method was tested using a cylindrical phantom and the

18

F-DOPA data of a brain cancer patient.

Emphysema is one of the most widespread diseases in subjects with smoking history. The gold standard method for estimating the severity of emphysema is a lung function test, such as forced expiratory volume in first second (FEV

1

). However, several clinical studies showed that chest CT scans offer more sensitive estimates of emphysema progression. The standard CT densitometric score of emphysema is the relative area of voxels below a threshold (RA). The RA score is a global measurement and reflects the overall emphysema progression.

In this work, we propose a framework for estimation of local emphysema progression from longitudinal chest CT scans. First, images are registered to a common system of coordinates and then local image dissimilarities are computed in corresponding anatomical locations. Finally, the obtained dissimilarity representation is converted into a single emphysema progression score. We applied the proposed algorithm on 27 patients with severe emphysema with CT scans acquired five time points, at baseline, after 3, after 12, after 21 and after 24 or 30 months. The results showed consistent emphysema progression with time and the overall progression score correlates significantly with the increase in RA score.

The use of multivariate pattern recognition for the analysis of neural representations encoded in fMRI data has become a significant research topic, with wide applications in neuroscience and psychology. A popular approach is to learn a mapping from the data to the observed behavior. However, identifying the instantaneous cognitive state without reference to external conditions is a relatively unexplored problem and could provide important insights into mental processes. In this paper, we present preliminary but promising results from the application of an unsupervised learning technique to identify distinct brain states. The temporal ordering of the states were seen to be synchronized with the experimental conditions, while the spatial distribution of activity in a state conformed with the expected functional recruitment.

In neuroanatomy, automatic geometry extraction of neurons from electron microscopy images is becoming one of the main limiting factors in getting new insights into the functional structure of the brain. We propose a novel framework for tracing neuronal processes over serial sections for 3d reconstructions. The automatic processing pipeline combines the probabilistic output of a random forest classifier with geometrical consistency constraints which take the geometry of whole sections into account. Our experiments demonstrate significant improvement over grouping by Euclidean distance, reducing the split and merge error per object by a factor of two.

The deep brain nuclei play an important role in many brain functions and particularly motor control. Damage to these structures result in movement disorders such as in Parkinson’s disease or Huntington’s disease, or behavioural disorders such as Tourette syndrome. In this paper, we propose to study the connectivity profile of the deep nuclei to the motor, associative or limbic areas and we introduce a novel tool to build a probabilistic atlas of these connections to the cortex directly on the surface of the cortical mantel, as it corresponds to the space of functional interest. The tool is then applied on two populations of healthy volunteers and patients suffering from severe Huntington’s disease to produce two surface atlases of the connectivity of the basal ganglia to the cortical areas. Finally, robust statistics are used to characterize the differences of that connectivity between the two populations, providing new connectivity-based biomarkers of the pathology.

The quantification of brain asymmetries may provide biomarkers for presurgical localization of language function and can improve our understanding of neural structure-function relationships in health and disease. We propose a new method for studying the asymmetry of the white matter tracts in the entire brain, and we apply it to a preliminary study of normal subjects across the handedness spectrum. Methods for quantifying white matter asymmetry using diffusion MRI tractography have thus far been based on comparing numbers of fibers or volumes of a single fiber tract across hemispheres. We propose a generalization of such methods, where the “number of fibers” laterality measurement is extended to the entire brain using a soft fiber comparison metric. We summarize the distribution of fiber laterality indices over the whole brain in a histogram, and we measure properties of the distribution such as its skewness, median, and inter-quartile range. The whole-brain fiber laterality histogram can be measured in an exploratory fashion without hypothesizing asymmetries only in particular structures. We demonstrate an overall difference in white matter asymmetry in consistent- and inconsistent-handers: the skewness of the fiber laterality histogram is significantly different across handedness groups.

We introduce a fibre tractography framework based on a particle filter which estimates a local geometrical model of the underlying white matter tract, formulated as a ‘streamline flow’ using generalized helicoids. The method is not dependent on the diffusion model, and is applicable to diffusion tensor (DT) data as well as to high angular resolution reconstructions. The geometrical model allows for a robust inference of local tract geometry, which, in the context of the causal filter estimation, guides tractography through regions with partial volume effects. We validate the method on synthetic data and present results on two types

in vivo

data: diffusion tensors and a spherical harmonic reconstruction of the fibre orientation distribution function (fODF).

In many application of functional Magnetic Resonance Imaging (fMRI), including clinical or pharmacological studies, the definition of the location of the functional activity between subjects is crucial. While current acquisition and normalization procedures improve the accuracy of the functional signal localization, it is also important to ensure that functional foci detection yields accurate results, and reflects between-subject variability. Here we introduce a fast functional landmark detection procedure, that explicitly models the spatial variability of activation foci in the observed population. We compare this detection approach to standard statistical maps peak extraction procedures: we show that it yields more accurate results on simulations, and more reproducible results on a large cohort of subjects. These results demonstrate that explicit functional landmark modeling approaches are more effective than standard statistical mapping for brain functional focus detection.

Amyotrophic lateral sclerosis (ALS) is a progressive motor neuron disease with poor prognosis. Previous DW-MRI based studies in ALS on WM tracts showed a decrease of FA in tracts related to the motor system. Recent evidence suggests that extra-motor tracts are also affected by ALS. This paper aims to analyse the cingulum tracts of ALS patients and controls. To do so, we introduce kernel matching, a novel method to obtain optimal correspondence between the white matter tracts. The orientation of tract tensors in atlas space as well as the global tract shape are employed as prior information. The method proved successful to reduce the large variance of tensor shape features along the cinguli emanating from registration errors. Only after applying the proposed kernel matching method we found a significant increase in the tensor norm of both cinguli. We hypothesize that the degeneration of fibers increases tensor norm.

Gaussian smoothing of images is an important step in Voxel-based Analysis and Statistical Parametric Mapping (VBA-SPM); it accounts for registration errors and integrates imaging signals from a region around each voxel being analyzed. However, it has also become a limitation of VBA-SPM based methods, since it is often chosen empirically, non-optimally, and lacks spatial adaptivity to the shape and spatial extent of the region of interest. In this paper, we propose a new framework, named Optimally-Discriminative Voxel-Based Analysis (ODVBA), for determining the optimal spatially adaptive smoothing of images, followed by applying voxel-based group analysis. In ODVBA, Nonnegative Discriminative Projection is applied locally to get the direction that best discriminates between two groups, e.g. patients and controls; this direction is equivalent to local filtering by an optimal kernel whose coefficients define the optimally discriminative direction. By considering all the neighborhoods that contain a given voxel, we then compose this information to produce the statistic for each voxel. Permutation tests are finally used to obtain the statistical significance. The experiments on Mild Cognitive Impairment (MCI) study have shown the effectiveness of the framework.

Landmark-based 3D hippocampal shape classification involves high-dimensional descriptor space, many noisy and redundant features, and a very small number of training samples. Feature selection becomes critical in this situation, because it not only improves classification performance, but also identifies the regions that contribute more to shape discrimination. This work identifies the drawbacks of SVM-RFE, and proposes a novel class-separability-based feature selection approach to overcome them. We formulate feature selection as a constrained integer optimization and develop a new algorithm to efficiently and optimally solve this problem. Theoretical analysis and experimental study on both synthetic data and real hippocampus data demonstrate its superior performance over the prevailing SVM-RFE. Our work provides a new efficient feature selection tool for hippocampal shape classification.

Estimation of intracranial stress distribution caused by mass effect is critical to the management of hemorrhagic stroke or brain tumor patients, who may suffer severe secondary brain injury from brain tissue compression. Coupling with physiological parameters that are readily available using MRI, eg, tissue perfusion, a non-invasive, quantitative and regional estimation of intracranial stress distribution could offer a better understanding of brain tissue’s reaction under mass effect. A quantitative and sound measurement serving this particular purpose remains elusive due to multiple challenges associated with biomechanical modeling of the brain. One such challenge for the conventional Lagrangian frame based finite element method (LFEM) is that the mesh distortion resulted from the expansion of the mass effects can terminate the simulation prematurely before the desired pressure loading is achieved. In this work, we adopted an arbitrary Lagrangian and Eulerian FEM method (ALEF) with explicit dynamic solutions to simulate the expansion of brain mass effects caused by a pressure loading. This approach consists of three phases: 1) a Lagrangian phase to deform mesh like LFEM, 2) a mesh smoothing phase to reduce mesh distortion, and 3) an Eulerian phase to map the state variables from the old mesh to the smoothed one. In 2D simulations with simulated geometries, this approach is able to model substantially larger deformations compared to LFEM. We further applied this approach to a simulation with 3D real brain geometry to quantify the distribution of von Mises stress within the brain.

Measures from event-related functional MRI, diffusion tensor imaging tractography and cognitive performance in a language-based task were used to test the hypothesis that both functional and structural connectivity provide independent and complementary information that aids in the identification of network components most related to the neurobiological basis for language and cognitive processing. Structural connectivity was measured by averaging fractional anisotropy (FA) over a geometric fiber bundle model that projects local white matter properties onto a centerline. In the uncinate fasciculus FA was found to predict performance on a measure of decision-making regarding homonym meaning. Functional synchronization of BOLD fMRI signals between frontal and temporal regions connected by the uncinate fasciculus was also found to predict the performance measure. Multiple regression analysis demonstrated that combining equidimensional measures of functional and structural connectivity identified the network components that most significantly predict performance.

Accurate and precise identification of multiple sclerosis (MS) lesions in longitudinal MRI is important for monitoring disease progression and for assessing treatment effects. We present a probabilistic framework to automatically detect new, enlarging and resolving lesions in longitudinal scans of MS patients based on multimodal subtraction magnetic resonance (MR) images. Our Bayesian framework overcomes registration artifact by explicitly modeling the variability in the difference images, the tissue transitions, and the neighbourhood classes in the form of likelihoods, and by embedding a classification of a reference scan as a prior. Our method was evaluated on (a) a scan-rescan data set consisting of 3 MS patients and (b) a multicenter clinical data set consisting of 212 scans from 89 RRMS (relapsing-remitting MS) patients. The proposed method is shown to identify MS lesions in longitudinal MRI with a high degree of precision while remaining sensitive to lesion activity.

The question of how large-scale systems interact with each other is intriguing given the increasingly established network structures of whole brain organization. Commonly used regional interaction approaches, however, cannot address this question. In this paper, we proposed a multivariate network-level framework to directly quantify the interaction pattern between large-scale functional systems. The proposed framework was tested on three different brain states, including resting, finger tapping and movie watching using functional connectivity MRI. The interaction patterns among five predefined networks including dorsal attention (DA), default (DF), frontal-parietal control (FPC), motor-sensory (MS) and visual (V) were delineated during each state. Results show dramatic and expected network-level correlation changes across different states underscoring the importance of network-level interactions for successful transition between different states. In addition, our analysis provides preliminary evidence of the potential regulating role of FPC on the two opposing systems-DA and DF on the network level.

This paper presents a generalized learning based framework for improving both speed and accuracy of the existing deformable registration method. The key of our framework involves the utilization of a support vector regression (SVR) to learn the correlation between brain image appearances and their corresponding shape deformations to a template, for helping significantly cut down the computation cost and improve the robustness to local minima by using the learned correlation to instantly predict a good subject-specific deformation initialization for any given subject under registration. Our framework consists of three major parts: 1)

training

of SVR models based on the statistics of image samples and their shape deformations to capture intrinsic image-deformation correlations, 2)

deformation prediction

for a new subject with the trained SVR models to generate a subject-resemblance intermediate template by warping the original template with the predicted deformations, and 3)

estimating of the residual deformation

from the intermediate template to the subject for refined registration. Any existing deformable registration methods can be easily employed for

training

the SVR models and

estimating the residual deformation

. We have tested in this paper the two widely used deformable registration algorithms, i.e., HAMMER [1] and diffeomorphic demons [2], for demonstration of our proposed frameowrk. Experimental results show that, compared to the registration using the original methods (with no deformation prediction), our framework achieves a significant speedup (6X faster than HAMMER, and 3X faster than diffeomorphic demons), while maintaining comparable (or even slighly better) registration accuracy.

Cellular processes are crucial for cells to survive and function properly. To study their underlying mechanisms quantitatively with fluorescent live cell microscopy, it is necessary to track a large number of particles involved in these processes. In this paper, we present a method to automatically track particles, called clathrin coated pits (CCPs), which are formed in clathrin mediated endocytosis (CME). The tracking method is developed based on a MAP framework, and it consists of particle detection and trajectory estimation. To detect particles in 2D images and take account of Poisson noise, a Gaussian mixture model is fitted to image data, for which initial parameters are provided by a combination of image filtering and histogram based thresholding methods. A multiple hypothesis based algorithm is developed to estimate the trajectories based on detection data. To use the current knowledge about CCPs, their properties of motion and intensity are considered in our models. The tracking method is evaluated on synthetic data and real data, and experimental results show that it has high accuracy and is in good agreement with manual tracking.

We develop a new algorithm to automatically register hippocampal(HP) surfaces with complete geometric matching, avoiding the need to manually label landmark features. A good registration depends on a reasonable choice of shape energy that measures the dissimilarity between surfaces. In our work, we first propose a complete shape index using the Beltrami coefficient and curvatures, which measures subtle local differences. The proposed shape energy is zero if and only if two shapes are identical up to a rigid motion. We then seek the best surface registration by minimizing the shape energy. We propose a simple representation of surface diffeomorphisms using Beltrami coefficients, which simplifies the optimization process. We then iteratively minimize the shape energy using the proposed Beltrami Holomorphic flow (BHF) method. Experimental results on 212 HP of normal and diseased (Alzheimer’s disease) subjects show our proposed algorithm is effective in registering HP surfaces with complete geometric matching. The proposed shape energy can also capture local shape differences between HP for disease analysis.

Due to the complex noise structure of functional magnetic resonance imaging (fMRI) data, methods that rely on information within a single subject often results in unsatisfactory functional segmentation. We thus propose a new graph-theoretic method, “Group Random Walker” (GRW), that integrates group information in detecting single-subject activation. Specifically, we extend each subject’s neighborhood system in such a way that enables the states of both intra- and inter-subject neighbors to be regularized without having to establish a one-to-one voxel correspondence as required in standard fMRI group analysis. Also, the GRW formulation provides an exact, unique closed-form solution for jointly estimating the probabilistic activation maps of all subjects with global optimality guaranteed. Validation is performed on synthetic and real data to demonstrate GRW’s superior detection power over standard analysis methods.

Tensor based morphology (TBM) is a powerful approach to analyze local structural changes in brain anatomy. However, conventional scalar TBM methods are unable to present direction-specific analysis of volume changes required to model complex changes such as those during brain growth. In this paper, we describe novel TBM descriptors for studying direction-specific changes in a subject population which can be used in conjunction with scalar TBM to analyze local patterns in directionality of volume change during brain development. We illustrate the use of these methods by studying brain developmental patterns in fetuses. Results show that this approach detects early changes local growth that are related to the early stages of sulcal and gyral formation.

This paper presents a connectivity-based parcellation of the human post-central gyrus, at the level of the group of subjects. The dimension of the clustering problem is reduced using a set of cortical regions of interest determined at the inter-subject level using a surface-based coordinate system, and representing the regions with a strong connection to the post-central gyrus. This process allows a clustering based on criteria which are more reproducible across subjects than in an intra-subject approach. We obtained parcels relatively stable in localisation across subjects as well as homogenous and well-separated to each other in terms of connectivity profiles. To address the parcellation at the inter-subject level provides a direct matching between parcels across subjects. In addition, this method allows the identification of subject-specific parcels. This property could be useful for the study of pathologies.

Super-resolution techniques provide a route to studying fine scale anatomical detail using multiple lower resolution acquisitions. In particular, techniques that do not depend on regular sampling can be used in medical imaging situations where imaging time and resolution are limited by subject motion. We investigate in this work the use of a super-resolution technique for anisotropic fetal brain MR data reconstruction without modifying the data acquisition protocol. The approach, which consists of iterative motion correction and high resolution image estimation, is compared with a previously used scattered data interpolation-based reconstruction method. To optimize acquisition time, an evaluation of the influence of the number of input images and image noise is also performed. Evaluation on simulated MR images and real data show significant improvements in performance provided by the super-resolution approach.

In this paper we present a new method for spatial regularization of functional connectivity maps based on Markov Random Field (MRF) priors. The high level of noise in fMRI leads to errors in functional connectivity detection algorithms. A common approach to mitigate the effects of noise is to apply spatial Gaussian smoothing, which can lead to blurring of regions beyond their actual boundaries and the loss of small connectivity regions. Recent work has suggested MRFs as an alternative spatial regularization in detection of fMRI activation in task-based paradigms. We propose to apply MRF priors to the computation of functional connectivity in resting-state fMRI. Our Markov priors are in the space of pairwise voxel connections, rather than in the original image space, resulting in a MRF whose dimension is twice that of the original image. The high dimensionality of the MRF estimation problem leads to computational challenges. We present an efficient, highly parallelized algorithm on the Graphics Processing Unit (GPU). We validate our approach on a synthetically generated example as well as real data from a resting state fMRI study.

This paper presents a new modelling technique for the deformation of thin anatomical structures like membranes and hollow organs. We show that the behaviour of this type of surface tissue can be abstracted with a modelling of their elastic resistance using shell theory. In order to apply the shell theory in the context of medical simulation, our method propose to base the geometrical reconstruction of the organ on the shape functions of the shell element. Moreover, we also use these continuous shape functions to handle the contacts and the interactions with other types of deformable tissues. The technique is illustrated using several examples including the simulation of an angioplasty procedure.

Cubic Hermite meshes provide an efficient representation of anatomy, and are useful for simulating soft tissue mechanics. However, their personalization can be a complex, time consuming and labour-intensive process. This paper presents a method based on image registration and using an existing template for deriving a patient-specific cubic Hermite mesh. Its key contribution is a solution to customise a Hermite continuous description of a shape with the use of a discrete warping field. Fitting accuracy is first tested and quantified against an analytical ground truth solution. To then demonstrate its clinical utility, a generic cubic Hermite heart ventricular model is personalized to the anatomy of a patient, and its mechanical stability is successfully tested. The method achieves an easy, fast and accurate personalization of cubic Hermite meshes, constituting a crucial step for the clinical adoption of physiological simulations.

Reduced order modelling, in which a full system response is projected onto a subspace of lower dimensionality, has been used previously to accelerate finite element solution schemes by reducing the size of the involved linear systems. In the present work we take advantage of a secondary effect of such reduction for explicit analyses, namely that the stable integration time step is increased far beyond that of the full system. This phenomenon alleviates one of the principal drawbacks of explicit methods, compared with implicit schemes. We present an explicit finite element scheme in which time integration is performed in a reduced basis. The computational benefits of the procedure within a GPU-based execution framework are examined, and an assessment of the errors introduced is given. Speedups approaching an order of magnitude are feasible, without introduction of prohibitive errors, and without hardware modifications. The procedure may have applications in medical image-guidance problems in which both speed and accuracy are vital.

A method is proposed to simulate nodules and diffuse infiltrates in chest radiographs. This allows creation of large annotated databases for training of both radiologists and computer aided diagnosis systems. Realistic nodules and diffuse infiltrates were generated from three-dimensional templates segmented from CT data. These templates are rescaled, rotated, projected and superimposed on a radiograph. This method was compared, in an observer study, to a previously published method that simulates pulmonary nodules as perfectly spherical objects. Results show that it is hard for human observers to distinguish real and simulated nodules when using templates (AUC-values do not significantly differ from .5,

p

 > .05 for all observers). The method that produced spherical nodules performed slightly worse (AUC of one observer differs significantly from .5,

p

 = .011). Simulation of diffuse infiltrates is challenging but also feasible (AUC=0.67 for one observer).

This paper presents a method for constructing detailed geometric models of tissue microstructure for synthesizing realistic diffusion MRI data. We construct three-dimensional mesh models from confocal microscopy image stacks using the marching cubes algorithm. Random-walk simulations within the resulting meshes provide synthetic diffusion MRI measurements. Experiments optimise simulation parameters and complexity of the meshes to achieve accuracy and reproducibility while minimizing computation time. Finally we assess the quality of the synthesized data from the mesh models by comparison with scanner data as well as synthetic data from simple geometric models and simplified meshes that vary only in two dimensions. The results support the extra complexity of the three-dimensional mesh compared to simpler models although sensitivity to the mesh resolution is quite robust.

The mechanisms of human cerebral cortex folding and their interactions during brain development are largely unknown, partly due to the difficulties in biological experiments and data acquisition for the developing fetus brain. Computational modeling and simulation provide a novel approach to the understanding of cortex folding processes in normal or aberrant neurodevelopment. Based on our recently developed computational model of the cerebral cortex folding using neuronal growth model and mechanical skull constraint, this paper presents a computational dynamic model of the brain skull that regulates the cortical folding simulation. Our simulation results show that the dynamic skull model is more biologically realistic and significantly improves our cortical folding simulation results. This work provides further computational support to the hypothesis that skull is an important regulator of cortical folding.

Despite recent efforts in cardiac electrophysiology modelling, there is still a strong need to make macroscopic models usable in planning and assistance of the clinical procedures. This requires model personalisation i.e. estimation of patient-specific model parameters and computations compatible with clinical constraints. Fast macroscopic models allow a quick estimation of the tissue conductivity, but are often unreliable in prediction of arrhythmias. On the other side, complex biophysical models are quite expensive for the tissue conductivity estimation, but are well suited for arrhythmia predictions. Here we present a coupled personalisation framework, which combines the benefits of the two models. A fast Eikonal (EK) model is used to estimate the conductivity parameters, which are then used to set the parameters of a biophysical model, the Mitchell-Schaeffer (MS) model. Additional parameters related to Action Potential Duration (APD) and APD restitution curves for the tissue are estimated for the MS model. This framework is applied to a clinical dataset provided with an hybrid X-Ray/MR imaging on an ischemic patient. This personalised MS Model is then used for

in silico

simulation of clinical Ventricular Tachycardia (VT) stimulation protocol to predict the induction of VT. This proof of concept opens up possibilities of using VT induction modelling directly in the intervention room, in order to plan the radio-frequency ablation lines.

In this paper, we propose an interpolation-based method for simulating needle images in B-mode ultrasound. We parametrize the needle image as a function of needle position and orientation. We collect needle images under various spatial configurations in a water-tank using a guidance robot. Then we use multi-dimensional tensor-product interpolation to simulate images of needles with arbitrary poses and positions using the collected images. Interpolated needle images are superimposed on top of phantom image backgrounds. The similarity between the simulated and the real images is measured using a correlation metric. A comparison with in-vivo images is also performed. The simulation procedure is demonstrated using transverse needle images and extended to sagittal needle images and brachytherapy seed images. The proposed method could be used in clinical procedure training simulators.

Postprocedural analysis of gastrointestinal (GI) endoscopic videos is a difficult task because the videos often suffer from a large number of poor-quality frames due to the motion or out-of-focus blur, specular highlights and artefacts caused by turbid fluid inside the GI tract. Clinically, each frame of the video is examined individually by the endoscopic expert due to the lack of a suitable visualisation technique. In this work, we introduce a low dimensional representation of endoscopic videos based on a manifold learning approach. The introduced endoscopic video manifolds (EVMs) enable the clustering of poor-quality frames and grouping of different segments of the GI endoscopic video in an unsupervised manner to facilitate subsequent visual assessment. In this paper, we present two novel inter-frame similarity measures for manifold learning to create structured manifolds from complex endoscopic videos. Our experiments demonstrate that the proposed method yields high precision and recall values for uninformative frame detection (90.91% and 82.90%) and results in well-structured manifolds for scene clustering.

Image focus quality is of utmost importance in digital microscopes because the pathologist cannot accurately characterize the tissue state without focused images. We propose to train a classifier to measure the focus quality of microscopy scans based on an extensive set of image features. However, classifiers rely heavily on the quality and quantity of the training data, and collecting annotated data is tedious and expensive. We therefore propose a new method to automatically generate large amounts of training data using image stacks. Our experiments demonstrate that a classifier trained with the image stacks performs comparably with one trained with manually annotated data. The classifier is able to accurately detect out-of-focus regions, provide focus quality feedback to the user, and identify potential problems of the microscopy design.

The current procedure for diagnosis of Crohn’s disease (CD) from Capsule Endoscopy is a tedious manual process which requires the clinician to visually inspect large video sequences for matching and categorization of diseased areas (lesions). Automated methods for matching and classification can help improve this process by reducing diagnosis time and improving consistency of categorization. In this paper, we propose a novel SVM-based similarity learning method for distinguishing between correct and incorrect matches in Capsule Endoscopy (CE). We also show that this can be used in conjunction with a voting scheme to categorize lesion images. Results show that our methods outperform standard classifiers in discriminating similar from dissimilar lesion images, as well as in lesion categorization. We also show that our methods drastically reduce the complexity (training time) by requiring only one half of the data for training, without compromising the accuracy of the classifier.

While there has been substantial progress in segmenting natural images, state-of-the-art methods that perform well in such tasks unfortunately tend to underperform when confronted with the different challenges posed by electron microscope (EM) data. For example, in EM imagery of neural tissue, numerous cells and subcellular structures appear within a single image, they exhibit irregular shapes that cannot be easily modeled by standard techniques, and confusing textures clutter the background. We propose a fully automated approach that handles these challenges by using sophisticated cues that capture global shape and texture information, and by learning the specific appearance of object boundaries. We demonstrate that our approach significantly outperforms state-of-the-art techniques and closely matches the performance of human annotators.

Full reconstruction of neuron morphology is of fundamental interest for the analysis and understanding of neuron function. We have developed a novel method capable of tracing neurons in three-dimensional microscopy data automatically. In contrast to template-based methods, the proposed approach makes no assumptions on the shape or appearance of neuron’s body. Instead, an efficient seeding approach is applied to find significant pixels almost certainly within complex neuronal structures and the tracing problem is solved by computing an graph tree structure connecting these seeds. In addition, an automated neuron comparison method is introduced for performance evaluation and structure analysis. The proposed algorithm is computationally efficient. Experiments on different types of data show promising results.

Learning medical image interpretation is an evolutive process that requires modular training systems, from non-expert to expert users. Our study aims at developing such a system for endomicroscopy diagnosis. It uses a difficulty predictor to try and shorten the physician learning curve. As the understanding of video diagnosis is driven by visual similarities, we propose a content-based video retrieval approach to estimate the level of interpretation difficulty. The performance of our retrieval method is compared with several state of the art methods, and its genericity is demonstrated with two different clinical databases, on the Barrett’s Esophagus and on colonic polyps. From our retrieval results, we learn a difficulty predictor against a ground truth given by the percentage of false diagnoses among several physicians. Our experiments show that, although our datasets are not large enough to test for statistical significance, there is a noticeable relationship between our retrieval-based difficulty estimation and the difficulty experienced by the physicians.

We propose a technique for recovering the position and depth of pronuclei of human zygotes, from Z-stacks acquired using Hoffman Modulation Contrast microscopy. We use Local Binary Pattern features for describing local texture, and integrate information from multiple neighboring areas of the stack, including those where the object to be detected would appear defocused; interestingly, such defocused areas provide very discriminative information for detection. Experimental results confirm the effectiveness of our approach, which allows one to derive new 3D measurements for improved scoring of zygotes during In Vitro Fertilization.

The effectiveness and clinical benefits of image guided surgery are well established for procedures where there is manageable tissue motion. In minimally invasive cardiac, gastrointestinal, or abdominal surgery, large scale tissue deformation prohibits accurate registration and fusion of pre- and intra-operative data. Vision based techniques such as structure from motion and simultaneous localization and mapping are capable of recovering 3D structure and laparoscope motion. Current research in the area generally assumes the environment is static, which is difficult to satisfy in most surgical procedures. In this paper, a novel framework for simultaneous online estimation of laparoscopic camera motion and tissue deformation in a dynamic environment is proposed. The method only relies on images captured by the laparoscope to sequentially and incrementally generate a dynamic 3D map of tissue motion that can be co-registered with pre-operative data. The theoretical contribution of this paper is validated with both simulated and

ex vivo

data. The practical application of the technique is further demonstrated on

in vivo

procedures.

Co-located optical and virtual colonoscopy images provide important clinical information during routine colonoscopy procedures. Tracking algorithms that rely on image features to align virtual and optical images can fail when they encounter blurry image sequences. This is a common occurrence in colonoscopy images, when the endoscope touches a wall or is immersed in fluid. We propose a

region-flow

based matching algorithm to determine the large changes between images that bridge such interruptions in the visual field. The region flow field is used as the means to limit the search space for computing corresponding feature points; a sequence of refining steps is performed to identify the most reliable and accurate feature point pairs. The feature point pairs are then used in a deformation based scheme to compute the final camera parameters. We have successfully tested this algorithm on four clinical colonoscopy image sequences containing anywhere from 9-57 consecutive blurry images. Two additional tabletop experiments were performed to quantitatively validate the algorithm: the endoscope was moved along a slightly curved path by 24 mm and along a straight path by 40 mm. Our method reported errors within 1-5% in these experiments.

Endoscopy guided probe-based optical biopsy is a new method for detecting sites for tissue biopsy. These sites need to be re-targeted precisely and accurately in subsequent images of the same endoscopy session for treatment or for visual guidance of surgical instruments. A new system for re-targeting biopsy sites and for characterising analytically their uncertainty is presented. It makes use of epipolar lines derived from multiple endoscopic views of the same site. It was tested on real patient data acquired during colonoscopy and gastroscopy. Gold standards of the biopsy site were obtained by Argon Plasma Coagulation tattooing. Re-targeting precision and accuracy were better than 0.8mm which is sufficient for most clinical endoscopic applications.

A video recording of an examination by Wireless Capsule Endoscopy (WCE) may typically contain more than 55,000 video frames, which makes the manual visual screening by an experienced gastroenterologist a highly time-consuming task. In this paper, we propose a novel method of epitomized summarization of WCE videos for efficient visualization to a gastroenterologist. For each short sequence of a WCE video, an epitomized frame is generated. New constraints are introduced into the epitome formulation to achieve the necessary visual quality for manual examination, and an EM algorithm for learning the epitome is derived. First, the local context weights are introduced to generate the epitomized frame. The epitomized frame preserves the appearance of all the input patches from the frames of the short sequence. Furthermore, by introducing spatial distributions for semantic interpretation of image patches in our epitome formulation, we show that it also provides a framework to facilitate the semantic description of visual features to generate organized visual summarization of WCE video, where the patches in different positions correspond to different semantic information. Our experiments on real WCE videos show that, using epitomized summarization, the number of frames have to be examined by the gastroenterologist can be reduced to less than one-tenth of the original frames in the video.

In this paper, we study the problem of finding organization patterns of chromosomes inside the cell nucleus from microscopic nucleus images. Emerging evidence from cell biology research suggests that global chromosome organization has a vital role in fundamental cell processes related to gene expression and regulation. To understand how chromosome territories are neighboring (or associated) to each other, in this paper we present a novel technique for computing a common association pattern, represented as a Maximum Association Graph (MAG), from the nucleus images of a population of cells. Our approach is based on an interesting integer linear programming formulation of the problem and utilizes inherent observations of the problem to yield optimal solutions. A two-stage technique is also introduced for producing near optimal approximations for large data sets.

With the ultimate goal to quantify important biological parameters of microtubules, we present a method to estimate the 3D positions of microtubules from multi-angle TIRF data based on the calibrated decay profiles for each angle. Total Internal Reflection Fluorescence (TIRF) Microscopy images are actually projections of 3D volumes and hence cannot alone produce an accurate localization of structures in the z-dimension, however, they provide greatly improved axial resolution for biological samples. Multiple angle-TIRF microscopy allows controlled variation of the incident angle of the illuminating laser beam, thus generating a set of images of different penetration depths with the potential to estimate the 3D volume of the sample. Our approach incorporates prior information about intensity and geometric smoothness. We validate our method using computer simulated phantom data and test its robustness to noise. We apply our method to TIRF images of microtubules in PTK

2

cells and compare the distribution of the microtubule curvatures with electron microscopy (EM) images.

Non-parametric image registration is still among the most challenging problems in both computer vision and medical imaging. Here, one tries to minimize a joint functional that is comprised of a similarity measure and a regularizer in order to obtain a reasonable displacement field that transforms one image to the other. A common way to solve this problem is to formulate a necessary condition for an optimizer, which in turn leads to a system of partial differential equations (PDEs). In general, the most time consuming part of the registration task is to find a numerical solution for such a system. In this paper, we present a generalized and efficient numerical scheme for solving such PDEs simply by applying 1-dimensional recursive filtering to the right hand side of the system based on the Green’s function of the differential operator that corresponds to the chosen regularizer. So in the end we come up with a general linear algorithm. We present the associated Green’s function for the diffusive and curvature regularizers and show how one may efficiently implement the whole process by using recursive filter approximation. Finally, we demonstrate the capability of the proposed method on realistic examples.

Registration uncertainty may be important information to convey to a surgeon when surgical decisions are taken based on registered image data. However, conventional non-rigid registration methods only provide the most likely deformation. In this paper we show how to determine the registration uncertainty, as well as the most likely deformation, by using an elastic Bayesian registration framework that generates a dense posterior distribution on deformations. We model both the likelihood and the elastic prior on deformations with Boltzmann distributions and characterize the posterior with a Markov Chain Monte Carlo algorithm. We introduce methods that summarize the high-dimensional uncertainty information and show how these summaries can be visualized in a meaningful way. Based on a clinical neurosurgical dataset, we demonstrate the importance that uncertainty information could have on neurosurgical decision making.

The acquisition of intra-subject data from multiple images is routinely performed to provide complementary information where a single image is not sufficient. However, these images are not always co-registered since they are acquired with different scanners, affected by subject’s movements during scans, and consist of different image attributes, e.g. image resolution, field of view (FOV) and intensity distributions. In this study, we propose a coupled registration-segmentation framework that simultaneously registers and segments intra-subject images with different image attributes. The proposed coupled framework is demonstrated with the processing of multiple level of detail (LOD) MRI acquisitions of the hip joint structures, which yield efficient and automated approaches to analyze soft tissues (from high-resolution MRI) in conjunction with the entire hip joint structures (from low resolution MRI).

Groupwise registration has received more and more attention in the area of medical image analysis, due to its importance in analysis of population data. One popular way for groupwise registration is to alternatively estimate the group mean image and register all subject images to the estimated group mean. However, for achieving better registration performance, it is important to always keep the sharpness of the group mean image during the registration, which has not been well investigated yet in the literature. To achieve this, we propose to treat each aligned subject, as well as its anatomical regions, differently when constructing the group mean image. Specifically, we propose a new objective function to generalize the conventional groupwise registration method by using a dynamic weighting strategy to weight adaptively across subjects and spatial regions, to construct a sharp group mean image in each stage of registration. By integrating this strategy into diffeomorphic demons algorithm, the performance of our groupwise registration can be significantly improved, compared to the conventional groupwise registration method that starts with a fuzzy group mean image.

We present a registration algorithm that can handle the discontinuity of deformation with an ultimate goal to investigate how pulmonary lobes deform to accommodate chest wall shape changes. We first show that discontinuities can exist in both normal and tangent directions. Such discontinuities are accounted for by a spatially varying diffusive regularization which restricts smoothing inside objects. Meanwhile, a distance term is combined with the sum of squared intensity differences (SSD) to explicitly match corresponding interfaces and intensity patterns. The capability of this new method is demonstrated using two-dimensional (2-D) synthetic examples with complete or incomplete “fissures” and three-dimensional (3-D) computed tomography (CT) lung datasets.

In this work we discuss the generalized treatment of the deformable registration problem in Sobolev spaces. We extend previous approaches in two points: 1) by employing a general energy model which includes a regularization term, and 2) by changing the notion of distance in the Sobolev space by problem-dependent Riemannian metrics. The actual choice of the metric is such that it has a preconditioning effect on the problem, it is applicable to arbitrary similarity measures, and features a simple implementation. The experiments demonstrate an improvement in convergence and runtime by several orders of magnitude in comparison to semi-implicit gradient flows in

L

2

. This translates to increased accuracy in practical scenarios. Furthermore, the proposed generalization establishes a theoretical link between gradient flow in Sobolev spaces and elastic registration methods.

In this paper, we present a new algorithm for non-linear registration of point sets. We estimate both forward and backward deformations fields best superposing the two point sets of interest and we make sure that they are consistent with each other by designing a symmetric cost function where they are coupled. Regularisation terms are included in this cost function to enforce deformation smoothness. Then we present a two-step iterative algorithm to optimise this cost function, where the two fields and the fuzzy matches between the two sets are estimated in turn. Building regularisers using the RKHS theory allows to obtain fast and efficient closed-form solutions for the optimal fields. The resulting algorithm is efficient and can deal with large point sets.

Deformable image registration in the presence of considerable contrast differences and large-scale size and shape changes represents a significant challenge for image registration. A representative driving application is the study of early brain development in neuroimaging, which requires co-registration of images of the same subject across time or building 4-D population atlases. Growth during the first few years of development involves significant changes in size and shape of anatomical structures but also rapid changes in tissue properties due to myelination and structuring that are reflected in the multi-modal Magnetic Resonance (MR) contrast measurements. We propose a new registration method that generates a mapping between brain anatomies represented as a multi-compartment model of tissue class posterior images and geometries. We transform intensity patterns into combined probabilistic and geometric descriptors that drive the matching in a diffeomorphic framework, where distances between geometries are represented using currents which does not require geometric correspondence. We show preliminary results on the registrations of neonatal brain MRIs to two-year old infant MRIs using class posteriors and surface boundaries of structures undergoing major changes. Quantitative validation demonstrates that our proposed method generates registrations that better preserve the consistency of anatomical structures over time.

In this paper, we present a fine and coarse approach for the multiscale registration of 3D medical images using Large Deformation Diffeomorphic Metric Mapping (LDDMM). This approach has particularly interesting properties since it estimates large, smooth and invertible optimal deformations having a rich descriptive power for the quantification of temporal changes in the images. First, we show the importance of the smoothing kernel and its influence on the final solution. We then propose a new strategy for the spatial regularization of the deformations, which uses simultaneously fine and coarse smoothing kernels. We have evaluated the approach on both 2D synthetic images as well as on 3D MR longitudinal images out of the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study. Results highlight the regularizing properties of our approach for the registration of complex shapes. More importantly, the results also demonstrate its ability to measure shape variations at several scales simultaneously while keeping the desirable properties of LDDMM. This opens new perspectives for clinical applications.

Accurate measurement of longitudinal changes of anatomical structure is important and challenging in many clinical studies. Also, for identification of disease-affected regions due to the brain disease, it is extremely necessary to register a population data to the common space simultaneously. In this paper, we propose a new method for simultaneous longitudinal and groupwise registration of a set of longitudinal data acquired from multiple subjects. Our goal is to 1) consistently measure the longitudinal changes from a sequence of longitudinal data acquired from the same subject; and 2) jointly align all image data (acquired from all time points of all subjects) to a hidden common space. To achieve these two goals, we

first

introduce a set of temporal fiber bundles to explore the spatial-temporal behavior of anatomical changes in each longitudinal data of the same subject.

Then

, a probabilistic model is built upon the hidden state of spatial smoothness and temporal continuity on the fibers.

Finally

, the transformation fields that connect each time-point image of each subject to the common space are simultaneously estimated by the expectation maximization (EM) approach, via the maximum a posterior (MAP) estimation of probabilistic models. Promising results are obtained to quantitatively measure the longitudinal changes of hippocampus volume, indicating better performance of our method than the conventional pairwise methods.

A novel registration method between 3D ultrasound and stereoscopic cameras is proposed based on tracking a registration tool featuring both ultrasound fiducials and optical markers. The registration tool is pressed against an air-tissue boundary where it can be seen both in ultrasound and in the camera view. By localizing the fiducials in the ultrasound volume, knowing the registration tool geometry, and tracking the tool with the cameras, a registration is found. This method eliminates the need for external tracking, requires minimal setup, and may be suitable for a range of minimally invasive surgeries. A study of the appearance of ultrasound fiducials on an air-tissue boundary is presented, and an initial assessment of the ability to localize the fiducials in ultrasound with sub-millimeter accuracy is provided. The overall accuracy of registration (1.69 ± 0.60 mm) is a noticeable improvement over other reported methods and warrants patient studies.

We seek to automatically establish dense correspondences across groups of images. Existing non-rigid registration methods usually involve local optimisation and thus require accurate initialisation. It is difficult to obtain such initialisation for images of complex structures, especially those with many self-similar parts. In this paper we show that satisfactory initialisation for such images can be found by a parts+geometry model. We use a population based optimisation strategy to select the best parts from a large pool of candidates. The best matches of the optimal model are used to initialise a groupwise registration algorithm, leading to dense, accurate results. We demonstrate the efficacy of the approach on two challenging datasets, and report on a detailed quantitative evaluation of its performance.

We propose a new, adaptive local measure based on gradient orientation similarity for the purposes of multimodal image registration. We embed this metric into a hierarchical registration framework, where we show that registration robustness and accuracy can be improved by adapting both the similarity metric and the pixel selection strategy to the Gaussian blurring scale and to the modalities being registered. A computationally efficient estimation of gradient orientations is proposed based on patch-wise rigidity. We have applied our method to both rigid and non-rigid multimodal registration tasks with different modalities. Our approach outperforms mutual information (MI) and previously proposed local approximations of MI for multimodal (e.g. CT/MRI) brain image registration tasks. Furthermore, it shows significant improvements in terms of mTRE over standard methods in the highly challenging clinical context of registering pre-operative brain MRI to intra-operative US images.

Non-linear image registration is a standard approach to track soft tissues in medical images. By estimating spatial transformations between images, visible structures can be followed over time. For clinical applications the model of transformation must be consistent with the properties of the biological tissue, such as incompressibility. LogDemons is a fast non-linear registration algorithm that provides diffusion-like diffeomorphic transformations parameterised by stationary velocity fields. Yet, its use for tissue tracking has been limited because of the

ad-hoc

Gaussian regularisation that prevents implementing other transformation models. In this paper, we propose a mathematical formulation of demons regularisation that fits into LogDemons framework. This formulation enables to ensure volume-preserving deformations by minimising the energy functional directly under the linear divergence-free constraint, yielding little computational overhead. Tests on synthetic incompressible fields showed that our approach outperforms the original logDemons in terms of incompressible deformation recovery. The algorithm showed promising results on one patient for the automatic recovery of myocardium strain from cardiac anatomical and 3D tagged MRI.

Intra-operative registration is one of the main challenges related to computer-assisted interventions. One common approach involves matching intra-operatively acquired surfaces (e.g. from a laser range scanner) to pre-operatively acquired planning data. In this paper, we propose a new method for correspondences search between surfaces, which can be used for the computation of an initial alignment. It generates graph representations and establishes correspondences by maximizing a global similarity measure. The method does not rely on landmarks or prominent surface characteristics and is independent on the initial pose of the surfaces relative to each other. According to an evaluation on a set of liver meshes, the method is able to correctly match small submeshes even in this presence of noise.

Bioluminescence imaging (BLI) offers the possibility to study and image biology at molecular scale in small animals with applications in oncology or gene expression studies. Here we present a novel model-based approach to 3D animal tracking from monocular video which allows the quantification of bioluminescence signal on freely moving animals. The 3D animal pose and the illumination are dynamically estimated through minimization of an objective function with constraints on the bioluminescence signal position. Derived from an inverse problem formulation, the objective function enables explicit use of temporal continuity and shading information, while handling important self-occlusions and time-varying illumination. In this model-based framework, we include a constraint on the 3D position of bioluminescence signal to enforce tracking of the biologically produced signal. The minimization is done efficiently using a quasi-Newton method, with a rigorous derivation of the objective function gradient. Promising experimental results demonstrate the potentials of our approach for 3D accurate measurement with freely moving animal.

In this paper, we introduce a novel approach to bridge the gap between the landmark-based and the iconic-based voxel-wise registration methods. The registration problem is formulated with the use of Markov Random Field theory resulting in a discrete objective function consisting of thee parts. The first part of the energy accounts for the iconic-based volumetric registration problem while the second one for establishing geometrically meaningful correspondences by optimizing over a set of automatically generated mutually salient candidate pairs of points. The last part of the energy penalizes locally the difference between the dense deformation field due to the iconic-based registration and the implied displacements due to the obtained correspondences. Promising results in real MR brain data demonstrate the potentials of our approach.

We present a novel feature-based groupwise registration method to simultaneously warp the subjects towards the common space. Due to the complexity of the groupwise registration, we resort to decoupling it into two easy-to-solve tasks, i.e., alternatively establishing the robust correspondences across different subjects and interpolating the dense deformation fields based on the detected sparse correspondences. Specifically, several novel strategies are proposed in the correspondence detection step.

First

, attribute vector, instead of intensity only, is used as a morphological signature to guide the anatomical correspondence detection among all subjects.

Second

, we detect correspondence only on the driving voxels with distinctive attribute vectors for avoiding the ambiguity in detecting correspondences for non-distinctive voxels.

Third

, soft correspondence assignment (allowing for adaptive detection of multiple correspondences in each subject) is also presented to help establish reliable correspondences across all subjects, which is particularly necessary in the beginning of groupwise registration. Based on the sparse correspondences detected on the driving voxels of each subject, thin-plate splines (TPS) are then used to propagate the correspondences on the driving voxels to the entire brain image for estimating the dense transformation for each subject. By iteratively repeating correspondence detection and dense transformation estimation, all the subjects will be aligned onto a common space simultaneously. Our groupwise registration algorithm has been extensively evaluated by 18 elderly brains, 16 NIREP, and 40 LONI data. In all experiments, our algorithm achieves more robust and accurate registration results, compared to a groupwise registration method and a pairwise registration method, respectively.