Medical Image Computing and Computer-Assisted Intervention – MICCAI 2013

Recent advances in computational electrophysiology (EP) models make them attractive for clinical use. We propose a novel data-driven approach to calibrate an EP model from standard 12-lead electrocardiograms (ECG), which are in contrast to invasive or dense body surface measurements widely available in clinical routine. With focus on cardiac depolarization, we first propose an efficient forward model of ECG by coupling a mono-domain, Lattice-Boltzmann model of cardiac EP to a boundary element formulation of body surface potentials. We then estimate a polynomial regression to predict myocardium, left ventricle and right ventricle endocardium electrical diffusion from QRS duration and ECG electrical axis. Training was performed on 4,200 ECG simulations, calculated in ≈3

s

each, using different diffusion parameters on 13 patient geometries. This allowed quantifying diffusion uncertainty for given ECG parameters due to the ill-posed nature of the ECG problem. We show that our method is able to predict myocardium diffusion within the uncertainty range, yielding a prediction error of less than 5

ms

for QRS duration and 2° for electrical axis. Prediction results compared favorably with those obtained with a standard optimization procedure, while being 60 times faster. Our data-driven model can thus constitute an efficient preliminary step prior to more refined EP personalization.

Despite extensive efforts to enhance catheter navigation, limited research has been done to visualize and monitor the tissue lesions created during ablation in the attempt to provide feedback for effective therapy. We propose a technique to visualize the temperature distribution and extent of induced tissue injury via an image-based model that uses physiological tissue parameters and relies on heat transfer principles to characterize lesion progression in near real time. The model was evaluated both numerically and experimentally using

ex vivo

bovine muscle samples while emulating a clinically relevant ablation protocol. Results show agreement to within 5°C between the model-predicted and experimentally measured end-ablation tissue temperatures, as well as comparable predicted and observed lesion characteristics. The model yields temperature and lesion updates in near real-time, thus providing reasonably accurate and sufficiently fast monitoring for effective therapy.

In cranio-maxillofacial surgery, the determination of a proper surgical plan is an important step to attain a desired aesthetic facial profile and a complete denture closure. In the present paper, we propose an efficient modeling approach to predict the surgical planning on the basis of the desired facial appearance and optimal occlusion. To evaluate the proposed planning approach, the predicted osteotomy plan of six clinical cases that underwent CMF surgery were compared to the real clinical plan. Thereafter, simulated soft-tissue outcomes were compared using the predicted and real clinical plan. This preliminary retrospective comparison of both osteotomy planning and facial outlook shows a good agreement and thereby demonstrates the potential application of the proposed approach in cranio-maxillofacial surgical planning prediction.

The growing availability of data from robotic and laparoscopic surgery has created new opportunities to investigate the modeling and assessment of surgical technical performance and skill. However, previously published methods for modeling and assessment have not proven to scale well to large and diverse data sets. In this paper, we describe a new approach for simultaneous detection of gestures and skill that can be generalized to different surgical tasks. It consists of two parts: (1) descriptive curve coding (DCC), which transforms the surgical tool motion trajectory into a coded string using accumulated Frenet frames, and (2) common string model (CSM), a classification model using a similarity metric computed from longest common string motifs. We apply DCC-CSM method to detect surgical gestures and skill levels in two kinematic datasets (collected from the da Vinci surgical robot). DCC-CSM method classifies gestures and skill with 87.81% and 91.12% accuracy, respectively.

Automatic and robust registration of pre-operative magnetic resonance imaging (MRI) and intra-operative ultrasound (US) is essential to neurosurgery. We reformulate and extend an approach which uses a Linear Correlation of Linear Combination (LC

2

)-based similarity metric, yielding a novel algorithm which allows for fully automatic US-MRI registration in the matter of seconds. It is invariant with respect to the unknown and locally varying relationship between US image intensities and both MRI intensity and its gradient. The overall method based on this both recovers global rigid alignment, as well as the parameters of a free-form-deformation (FFD) model. The algorithm is evaluated on 14 clinical neurosurgical cases with tumors, with an average landmark-based error of 2.52

mm

for the rigid transformation. In addition, we systematically study the accuracy, precision, and capture range of the algorithm, as well as its sensitivity to different choices of parameters.

Reconstructing the depth of stereo-endoscopic scenes is an important step in providing accurate guidance in robotic-assisted minimally invasive surgery. Stereo reconstruction has been studied for decades but remains a challenge in endoscopic imaging. Current approaches can easily fail to reconstruct an accurate and smooth 3D model due to textureless tissue appearance in the real surgical scene and occlusion by instruments. To tackle these problems, we propose a dense stereo reconstruction algorithm using convex optimisation with a cost-volume to efficiently and effectively reconstruct a smooth model while maintaining depth discontinuity. The proposed approach has been validated by quantitative evaluation using simulation and real phantom data with known ground truth. We also report qualitative results from real surgical images. The algorithm outperforms state of the art methods and can be easily parallelised to run in real-time on recent graphics hardware.

Despite the fact that several theories link cortical development and function to the development of white matter and its geometrical structure, the relationship between gray and white matter morphology has not been widely researched. In this paper, we propose a novel framework for investigating this relationship. Given a set of fiber tracts which connect to a particular cortical region, the key idea is to compute two scalar fields that represent geometrical characteristics of the white matter and of the surface of the cortical region. The distributions of these scalar values are then linked via Mutual Information, which results in a quantitative marker that can be used in the study of normal and pathological brain structure and development. We apply this framework to a population study on autism spectrum disorder in children.

Accurate and consistent labeling of longitudinal cortical surfaces is essential to understand the early dynamic development of cortical structure and function in both normal and abnormal infant brains. In this paper, we propose a novel method for simultaneous, consistent, and unbiased labeling of longitudinal dynamic cortical surfaces in the infant brain MR images. The proposed method is formulated as minimization of an energy function, which includes the data fitting, spatial smoothness and temporal consistency terms. Specifically, in the spirit of multi-atlas based label fusion, the data fitting term is designed to integrate adaptive contributions from multi-atlas surfaces, according to the similarity of their local cortical folding with that of the subject surface. The spatial smoothness term is designed to adaptively encourage label smoothness based on the local folding geometries, i.e., also allowing label discontinuity at sulcal bottoms, where the cytoarchitecturally and functionally distinct cortical regions are often divided. The temporal consistency term is further designed to encourage the label consistency between temporal corresponding vertices with similar local cortical folding. Finally, the entire energy function is efficiently minimized by a graph cuts method. The proposed method has been successfully applied to the labeling of longitudinal cortical surfaces of 13 infants, each with 6 serial images scanned from birth to 2 years of age. Both qualitative and quantitative evaluation results demonstrate the validity of the proposed method.

Identification of common and corresponding white matter (WM) regions of interest (ROI) across human brains has attracted growing interest because it not only facilitates comparison among individuals and populations, but also enables the assessment of structural/functional connectivity in populations. However, due to the complexity and variability of the WM structure and a lack of effective white matter streamline descriptors, establishing accurate correspondences of WM ROIs across individuals and populations has been a challenging open problem. In this paper, a novel fiber shape descriptor which can facilitate quantitative measurement of fiber bundle profile including connection complexity and similarity has been proposed. A novel framework was then developed using the descriptor to identify group-wise consistent connection hubs in WM regions as landmarks. 12 group-wise consistent WM landmarks have been identified in our experiment. These WM landmarks are found highly reproducible across individuals and accurately predictable on new individual subjects by our fiber shape descriptor. Therefore, these landmarks, as well as proposed fiber shape descriptor has shown great potential to human brain mapping.

In this work we compare parametric diffusion MRI models which explicitly seek to explain fibre dispersion in nervous tissue. These models aim at providing more specific biomarkers of disease by disentangling these structural contributions to the signal. Some models are drawn from recent work in the field; others have been constructed from combinations of existing compartments that aim to capture both intracellular and extracellular diffusion. To test these models we use a rich dataset acquired

in vivo

on the corpus callosum of a human brain, and then compare the models via the Bayesian Information Criteria. We test this ranking via bootstrapping on the data sets, and cross-validate across unseen parts of the protocol. We find that models that capture fibre dispersion are preferred. The results show the importance of modelling dispersion, even in apparently coherent fibres.

Diffusion MRI is a valuable tool for mapping tissue microstructure; however, multi-fiber models present challenges to image analysis operations. In this paper, we present a method for estimating models for such operations by clustering fiber orientations. Our approach is applied to ball-and-stick diffusion models, which include an isotropic tensor and multiple sticks encoding fiber volume and orientation. We consider operations which can be generalized to a weighted combination of fibers and present a method for representing such combinations with a mixture-of-Watsons model, learning its parameters by Expectation Maximization. We evaluate this approach with two experiments. First, we show it is effective for filtering in the presence of synthetic noise. Second, we demonstrate interpolation and averaging by construction of a tractography atlas, showing improved reconstruction of white matter pathways. These experiments indicate that our method is useful in estimating multi-fiber ball-and-stick diffusion volumes resulting from a range of image analysis operations.

Network representation of brain connectivity has provided a novel means of investigating brain changes arising from pathology, development or aging. The high dimensionality of these networks demands methods that are not only able to extract the patterns that highlight these sources of variation, but describe them individually. In this paper, we present a unified framework for learning subnetwork patterns of connectivity by their projective non-negative decomposition into a reconstructive basis set, as well as, additional basis sets representing development and group discrimination. In order to obtain these components, we exploit the geometrical distribution of the population in the connectivity space by using a graph-theoretical scheme that imposes locality-preserving properties. In addition, the projection of the subject networks into the basis set provides a low dimensional representation of it, that teases apart the different sources of variation in the sample, facilitating variation-specific statistical analysis. The proposed framework is applied to a study of diffusion-based connectivity in subjects with autism.

We present a method to detect epileptic regions based on functional connectivity differences between individual epilepsy patients and a healthy population. Our model assumes that the global functional characteristics of these differences are shared across patients, but it allows for the epileptic regions to vary between individuals. We evaluate the detection performance against intracranial EEG observations and compare our approach with two baseline methods that use standard statistics. The baseline techniques are sensitive to the choice of thresholds, whereas our algorithm automatically estimates the appropriate model parameters and compares favorably with the best baseline results. This suggests the promise of our approach for pre-surgical planning in epilepsy.

Whole-body MR receives increasing interest as potential alternative to many conventional diagnostic methods. Typical whole-body MR scans contain multiple data channels and are acquired in a multi-station manner. Quantification of such data typically requires correction of two types of artefacts: different intensity scaling on each acquired image stack, and intensity inhomogeneity (bias) within each stack. In this work, we present an all-in-one method that is able to correct for both mentioned types of acquisition artefacts. The most important properties of our method are: 1) All the processing is performed jointly on all available data channels, which is necessary for preserving the relation between them, and 2) It allows easy incorporation of additional knowledge for estimation of the bias field. Performed validation on two types of whole-body MR data confirmed superior performance of our approach in comparison with state-of-the-art bias removal methods.

Sparse perfusion deconvolution has been recently proposed to effectively improve the image quality and diagnostic accuracy of low-dose perfusion CT by extracting the complementary information from the high-dose perfusion maps to restore the low-dose using a joint spatio-temporal model. However the low-contrast tissue classes where infarct core and ischemic penumbra usually occur in cerebral perfusion CT tend to be over-smoothed, leading to loss of essential biomarkers. In this paper, we extend this line of work by introducing tissue-specific sparse deconvolution to preserve the subtle perfusion information in the low-contrast tissue classes by learning tissue-specific dictionaries for each tissue class, and restore the low-dose perfusion maps by joining the tissue segments reconstructed from the corresponding dictionaries. Extensive validation on clinical datasets of patients with cerebrovascular disease demonstrates the superior performance of our proposed method with the advantage of better differentiation between abnormal and normal tissue in these patients.

Ultrasound acquisition is a challenging task that requires simultaneous adjustment of several acquisition parameters (the depth, the focus, the frequency and its operation mode). If the acquisition parameters are not properly chosen, the resulting image will have a poor quality and will degrade the patient diagnosis and treatment workflow. Several hardware-based systems for autotuning the acquisition parameters have been previously proposed, but these solutions were largely abandoned because they failed to properly account for tissue inhomogeneity and other patient-specific characteristics. Consequently, in routine practice the clinician either uses population-based parameter presets or manually adjusts the acquisition parameters for each patient during the scan. In this paper, we revisit the problem of autotuning the acquisition parameters by taking a completely novel approach and producing a solution based on image analytics. Our solution is inspired by the autofocus capability of conventional digital cameras, but is significantly more challenging because the number of acquisition parameters is large and the determination of “good quality” images is more difficult to assess. Surprisingly, we show that the set of acquisition parameters which produce images that are favored by clinicians comprise a 1D manifold, allowing for a real-time optimization to maximize image quality. We demonstrate our method for acquisition parameter autotuning on several live patients, showing that our system can start with a poor initial set of parameters and automatically optimize the parameters to produce high quality images.

Increasing scan resolution in magnetic resonance imaging is possible with advances in acquisition technology. The increase in resolution, however, comes at the expense of severe image noise. The current approach is to acquire multiple images and average them to restore the lost quality. This approach is expensive as it requires a large number of acquisitions to achieve quality comparable to lower resolution images. We propose an image restoration method for reducing the number of required acquisitions. The method leverages a high-quality lower-resolution image of the same subject and a database of pairs of high-quality low/high-resolution images acquired from different individuals. Experimental results show that the proposed method decreases noise levels and improves contrast differences between fine-scale structures, yielding high signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). Comparisons with the current standard method of averaging approach and state-of-the-art non-local means denoising demonstrate the method’s advantages.

3-D endoscopy is an evolving field of research with the intention to improve safety and efficiency of minimally invasive surgeries.

Time-of-Flight

(ToF) imaging allows to acquire range data in real-time and has been engineered into a 3-D endoscope in combination with an RGB sensor (640×480 px) as a hybrid imaging system, recently. However, the ToF sensor suffers from a low spatial resolution (64×48 px) and a poor signal-to-noise ratio. In this paper, we propose a novel multi-frame super-resolution framework to improve range images in a ToF/RGB multi-sensor setup. Our approach exploits high-resolution RGB data to estimate subpixel motion used as a cue for range super-resolution. The underlying non-parametric motion model based on optical flow makes the method applicable to endoscopic scenes with arbitrary endoscope movements. The proposed method was evaluated on synthetic and real images. Our approach improves the peak-signal-to-noise ratio by 1.6 dB and structural similarity by 0.02 compared to single-sensor super-resolution.

The combination of functional and anatomical imaging technologies such as Positron Emission Tomography (PET) and Computed Tomography (CT) has shown its value in the preclinical and clinical fields. In PET/CT hybrid acquisition systems, CT-derived attenuation maps enable a more accurate PET reconstruction. However, CT provides only very limited soft-tissue contrast and exposes the patient to an additional radiation dose. In comparison, Magnetic Resonance Imaging (MRI) provides good soft-tissue contrast and the ability to study functional activation and tissue microstructures, but does not directly provide patient-specific electron density maps for PET reconstruction.

The aim of the proposed work is to improve PET/MR reconstruction by generating synthetic CTs and attenuation-maps. The synthetic images are generated through a multi-atlas information propagation scheme, locally matching the MRI-derived patient’s morphology to a database of pre-acquired MRI/CT pairs. Results show improvements in CT synthesis and PET reconstruction accuracy when compared to a segmentation method using an Ultrashort-Echo-Time MRI sequence.

Most natural images can be approximated using their low-rank components. This fact has been successfully exploited in recent advancements of matrix completion algorithms for image recovery. However, a major limitation of low-rank matrix completion algorithms is that they cannot recover the case where a whole row or column is missing. The missing row or column will be simply filled as an arbitrary combination of other rows or columns with known values. This precludes the application of matrix completion to problems such as super-resolution (SR) where missing values in many rows and columns need to be recovered in the process of up-sampling a low-resolution image. Moreover, low-rank regularization considers information globally from the whole image and does not take proper consideration of local spatial consistency. Accordingly, we propose in this paper a solution to the SR problem via simultaneous (global) low-rank and (local) total variation (TV) regularization. We solve the respective cost function using the alternating direction method of multipliers (ADMM). Experiments on MR images of adults and pediatric subjects demonstrate that the proposed method enhances the details of the recovered high-resolution images, and outperforms the nearest-neighbor interpolation, cubic interpolation, non-local means, and TV-based up-sampling.

Different types of nuclear imaging systems have been used in the past, starting with pre-operative gantry-based SPECT systems and gamma cameras for 2D imaging of radioactive distributions. The main applications are concentrated on diagnostic imaging, since traditional SPECT systems and gamma cameras are bulky and heavy. With the development of compact gamma cameras with good resolution and high sensitivity, it is now possible to use them without a fixed imaging gantry. Mounting the camera onto a robot arm solves the weight issue, while also providing a highly repeatable and reliable acquisition platform. In this work we introduce a novel robotic setup performing scans with a mini gamma camera, along with the required calibration steps, and show the first SPECT reconstructions. The results are extremely promising, both in terms of image quality as well as reproducibility. In our experiments, the novel setup outperformed a commercial fhSPECT system, reaching accuracies comparable to state-of-the-art SPECT systems.

Classical surgery is being disrupted by minimally invasive and transcatheter procedures. As there is no direct view or access to the affected anatomy, advanced imaging techniques such as 3D C-arm CT and C-arm fluoroscopy are routinely used for intra-operative guidance. However, intra-operative modalities have limited image quality of the soft tissue and a reliable assessment of the cardiac anatomy can only be made by injecting contrast agent, which is harmful to the patient and requires complex acquisition protocols. We propose a novel sparse matching approach for fusing high quality pre-operative CT and non-contrasted, non-gated intra-operative C-arm CT by utilizing robust machine learning and numerical optimization techniques. Thus, high-quality patient-specific models can be extracted from the pre-operative CT and mapped to the intra-operative imaging environment to guide minimally invasive procedures. Extensive quantitative experiments demonstrate that our model-based fusion approach has an average execution time of 2.9 s, while the accuracy lies within expert user confidence intervals.

Treatment coronary arteries endovascular involves catheter navigation through patient vasculature. The projective angiography guidance is limited in the case of chronic total occlusion where occluded vessel can not be seen. Integrating standard preoperative CT angiography information with live fluoroscopic images addresses this limitation but requires alignment of both modalities.

This article proposes a structure-based registration method that intrinsically preserves both the geometrical and topological coherencies of the vascular centrelines to be registered, by the means of a dedicated curve-to-curve distance pairs of closest curves are identified, while pairing their points. Preliminary experiments demonstrate that the proposed approach performs better than the standard Iterative Closest Point method giving a wider attraction basin and improved accuracy.

Image-guided interventions often rely on deformable multi-modal registration to align pre-treatment and intra-operative scans. There are a number of requirements for automated image registration for this task, such as a robust similarity metric for scans of different modalities with different noise distributions and contrast, an efficient optimisation of the cost function to enable fast registration for this time-sensitive application, and an insensitive choice of registration parameters to avoid delays in practical clinical use. In this work, we build upon the concept of structural image representation for multi-modal similarity. Discriminative descriptors are densely extracted for the multi-modal scans based on the “self-similarity context”. An efficient quantised representation is derived that enables very fast computation of point-wise distances between descriptors. A symmetric multi-scale discrete optimisation with diffusion regularisation is used to find smooth transformations. The method is evaluated for the registration of 3D ultrasound and MRI brain scans for neurosurgery and demonstrates a significantly reduced registration error (on average 2.1 mm) compared to commonly used similarity metrics and computation times of less than 30 seconds per 3D registration.

In this study, we propose an efficient non-rigid MR-TRUS deformable registration method to improve the accuracy of targeting suspicious locations during a 3D ultrasound (US) guided prostate biopsy. The proposed deformable registration approach employs the multi-channel modality independent neighbourhood descriptor (MIND) as the local similarity feature across the two modalities of MR and TRUS, and a novel and efficient duality-based convex optimization based algorithmic scheme is introduced to extract the deformations which align the two MIND descriptors. The registration accuracy was evaluated using 10 patient images by measuring the TRE of manually identified corresponding intrinsic fiducials in the whole gland and peripheral zone, and performance metrics (DSC, MAD and MAXD) for the apex, mid-gland and base of the prostate were also calculated by comparing two manually segmented prostate surfaces in the registered 3D MR and TRUS images. Experimental results show that the proposed method yielded an overall mean TRE of 1.74 mm, which is favorably comparable to a clinical requirement for an error of less than 2.5 mm.

We present a new framework for diffeomorphic image registration which supports natural interpretations of spatially-varying metrics. This framework is based on

left-invariant diffeomorphic metrics

(LIDM) and is closely related to the now standard

large deformation diffeomorphic metric mapping

(LDDMM). We discuss the relationship between LIDM and LDDMM and introduce a computationally convenient class of spatially-varying metrics appropriate for both frameworks. Finally, we demonstrate the effectiveness of our method on a 2D toy example and on the 40 3D brain images of the LPBA40 dataset.

This work presents a novel pharmacokinetic model based registration algorithm for the motion correction of dynamic positron emission tomography (PET) images. The algorithm employs a generalised model that derives the input function from the tomographic data itself to model the PET tracer kinetics and thus eliminates the need of arterial blood sampling. Both the temporal constraint from the tracer kinetic behaviour and spatial constraint from the image similarity are integrated in a joint probabilistic model, in which the subject motion and tracer kinetic parameters are iteratively optimised, leading to a groupwise registration framework of motion corrupted dynamic PET data. The algorithm is evaluated with simulated and measured human dopamine D3 receptor imaging data using [

11

C]-(+)-PHNO. The simulation-based validation demonstrates that the new algorithm has a subvoxel registration accuracy on average for noisy data with simulated motion artefacts. The algorithm also shows reductions in motion on initial experiments with measured clinical [

11

C]-(+)-PHNO brain data.

Atlases have a tremendous impact on the study of anatomy and function, such as in neuroimaging, or cardiac analysis. They provide a means to compare corresponding measurements across populations, or model the variability in a population. Current approaches to construct atlases rely on examples that show the same anatomical structure (e.g., the brain). If we study large heterogeneous clinical populations to capture subtle characteristics of diseases, we cannot assume consistent image acquisition any more. Instead we have to build atlases from imaging data that show only parts of the overall anatomical structure. In this paper we propose a method for the automatic contruction of an un-biased whole body atlas from so-called

fragments

. Experimental results indicate that the fragment based atlas improves the representation accuracy of the atlas over an initial whole body template initialization.

A key problem for many medical image segmentation tasks is the combination of different-level knowledge. We propose a novel scheme of embedding detected regions into a superpixel based graphical model, by which we achieve a full leverage on various image cues for ultrasound lesion segmentation. Region features are mapped into a higher-dimensional space via a boosted model to become well controlled. Parameters for regions, superpixels and a new affinity term are learned simultaneously within the framework of structured learning. Experiments on a breast ultrasound image data set confirm the effectiveness of the proposed approach as well as our two novel modules.

We describe a method to capture disease-specific components in organ shapes. A statistical shape model, constructed by the principal component analysis (PCA) of organ shapes, is used to define the subspace representing inter-subject shape variability. The first PCA is applied to the datasets of healthy organ shapes to define the subspace of normal variability. Then, the datasets of diseased shapes are projected onto the orthogonal complement (OC) of the subspace of normal variability, and the second PCA is applied to the projected datasets to derive the subspace representing the disease-specific variability. To calculate the OC of an n-dimensional subspace, a novel closed-form formulation is developed. Experiments were performed to show that the support vector machine classification in the OC subspace better discriminated healthy and diseased liver shapes using 99 CT data. The effects of the number of training data and the difference in segmentation methods on the classification accuracy were evaluated to clarify the characteristics of the proposed method.

In this work, we propose a visual phrase learning scheme to learn an optimal visual composite of anatomical components/parts from CT colonography images for computer-aided detection. The key idea is to utilize the anatomical parts of human body from medical images and associate them with biological targets of interest (organs, cancers, lesions, etc.) for joint detection and recognition. These anatomical parts of the human body are not necessarily near each other regarding their physical locations, and they serve more like a human body navigation system for detection and recognition. To show the effectiveness of the proposed learning scheme, we applied it to two sub-problems in computed tomographic colonography: teniae detection and classification of colorectal polyp candidates. Experimental results showed its efficacy.

This paper presents a framework for the fusion of multiple point distribution models (PDMs) with unknown point correspondences. With this work, models built from distinct patient groups and imaging modalities can be merged, with the aim to obtain a PDM that encodes a wider range of anatomical variability. To achieve this, two technical challenges are addressed in this work. Firstly, the model fusion must be carried out directly on the corresponding means and eigenvectors as the original data is not always available and cannot be freely exchanged across centers for various legal and practical reasons. Secondly, the PDMs need to be normalized before fusion as the point correspondence is unknown. The proposed framework is validated by integrating statistical models of the left and right ventricles of the heart constructed from different imaging modalities (MRI and CT) and with different landmark representations of the data. The results show that the integration is statistically and anatomically meaningful and that the quality of the resulting model is significantly improved.

We propose a robust multimodal dictionary learning method for multimodal images. Joint dictionary learning for both modalities may be impaired by lack of correspondence between image modalities in training data, for example due to areas of low quality in one of the modalities. Dictionaries learned with such non-corresponding data will induce uncertainty about image representation. In this paper, we propose a probabilistic model that accounts for image areas that are poorly corresponding between the image modalities. We cast the problem of learning a dictionary in presence of problematic image patches as a likelihood maximization problem and solve it with a variant of the EM algorithm. Our algorithm iterates identification of poorly corresponding patches and refinements of the dictionary. We tested our method on synthetic and real data. We show improvements in image prediction quality and alignment accuracy when using the method for multimodal image registration.

In this paper we propose a Bayesian framework for multi-object atlas estimation based on the metric of currents which permits to deal with both curves and surfaces without relying on point correspondence. This approach aims to study brain morphometry as a whole and not as a set of different components, focusing mainly on the shape and relative position of different anatomical structures which is fundamental in neuro-anatomical studies. We propose a generic algorithm to estimate templates of sets of curves (fiber bundles) and closed surfaces (sub-cortical structures) which have the same “form” (topology) of the shapes present in the population. This atlas construction method is based on a Bayesian framework which brings to two main improvements with respect to previous shape based methods. First, it allows to estimate from the data set a parameter specific to each object which was previously fixed by the user: the trade-off between data-term and regularity of deformations. In a multi-object analysis these parameters balance the contributions of the different objects and the need for an automatic estimation is even more crucial. Second, the covariance matrix of the deformation parameters is estimated during the atlas construction in a way which is less sensitive to the outliers of the population.

Accurate diagnosis of Alzheimer’s disease (AD), as well as its prodromal stage (i.e., mild cognitive impairment, MCI), is very important for possible delay and early treatment of the disease. Recently, multi-modality methods have been used for fusing information from multiple different and complementary imaging and non-imaging modalities. Although there are a number of existing multi-modality methods, few of them have addressed the problem of joint identification of disease-related brain regions from multi-modality data for classification. In this paper, we proposed a manifold regularized multi-task learning framework to jointly select features from multi-modality data. Specifically, we formulate the multi-modality classification as a multi-task learning framework, where each task focuses on the classification based on each modality. In order to capture the intrinsic relatedness among multiple tasks (i.e., modalities), we adopted a group sparsity regularizer, which ensures only a small number of features to be selected jointly. In addition, we introduced a new manifold based Laplacian regularization term to preserve the geometric distribution of original data from each task, which can lead to the selection of more discriminative features. Furthermore, we extend our method to the semi-supervised setting, which is very important since the acquisition of a large set of labeled data (i.e., diagnosis of disease) is usually expensive and time-consuming, while the collection of unlabeled data is relatively much easier. To validate our method, we have performed extensive evaluations on the baseline Magnetic resonance imaging (MRI) and fluorodeoxyglucose positron emission tomography (FDG-PET) data of Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. Our experimental results demonstrate the effectiveness of the proposed method.

The performance of automatic lesion detection is often affected by the intra- and inter-subject feature variations of lesions and normal anatomical structures. In this work, we propose a similarity-guided sparse representation method for image patch labeling, with three aspects of similarity information modeling, to reduce the chance that the best reconstruction of a feature vector does not provide the correct classification. Based on this classification model, we then design a new approach for detecting lesions in positron emission tomography – computed tomography (PET-CT) images. The approach works well with simple image features, and the proposed sparse representation model is effectively applied for both detection of all lesions and characterization of lung tumors and abnormal lymph nodes. The experiments show promising performance improvement over the state-of-the-art.

Magnetic resonance (MR) imaging has been demonstrated to be very useful for clinical diagnosis of Alzheimer’s disease (AD). A common approach to using MR images for AD detection is to spatially normalize the images by non-rigid image registration, and then perform statistical analysis on the resulting deformation fields. Due to the high nonlinearity of the deformation field, recent studies suggest to use initial momentum instead as it lies in a linear space and fully encodes the deformation field. In this paper we explore the use of initial momentum for image classification by focusing on the problem of AD detection. Experiments on the public ADNI dataset show that the initial momentum, together with a simple sparse coding technique—locality-constrained linear coding (LLC)—can achieve a classification accuracy that is comparable to or even better than the state of the art. We also show that the performance of LLC can be greatly improved by introducing proper weights to the codebook.

We present a novel persistent homological sparse network analysis framework for characterizing white matter abnormalities in tensor-based morphometry (TBM) in magnetic resonance imaging (MRI). Traditionally TBM is used in quantifying tissue volume change in each voxel in a massive univariate fashion. However, this obvious approach cannot be used in testing, for instance, if the change in one voxel is related to other voxels. To address this limitation of univariate-TBM, we propose a new persistent homological approach to testing more complex relational hypotheses across brain regions. The proposed methods are applied to characterize abnormal white matter in maltreated children. The results are further validated using fractional anisotropy (FA) values in diffusion tensor imaging (DTI).

In conventional multi-modality based classification framework, feature selection is typically performed separately for each individual modality, ignoring potential strong inter-modality relationship of the same subject. To extract this inter-modality relationship,

L

2,1

norm-based multi-task learning approach can be used to jointly select common features from different modalities. Unfortunately, this approach overlooks different yet complementary information conveyed by different modalities. To address this issue, we propose a novel multi-task feature selection method to effectively preserve the complementary information between different modalities, improving brain disease classification accuracy. Specifically, a new constraint is introduced to preserve the inter-modality relationship by treating the feature selection procedure of each modality as a task. This constraint preserves distance between feature vectors from different modalities after projection to low dimensional feature space. We evaluated our method on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset and obtained significant improvement on Alzheimer’s Disease (AD) and Mild Cognitive Impairment (MCI) classification compared to state-of-the-art methods.

A comprehensive framework for predicting response to therapy on the basis of heterogeneity in dceMRI parameter maps is presented. A motion-correction method for dceMRI sequences is extended to incorporate uncertainties in the pharmacokinetic parameter maps using a variational Bayes framework. Simple measures of heterogeneity (with and without uncertainty) in parameter maps for colorectal cancer tumours imaged before therapy are computed, and tested for their ability to distinguish between responders and non-responders to therapy. The statistical analysis demonstrates the importance of using the spatial distribution of parameters, and their uncertainties, when computing heterogeneity measures and using them to predict response on the basis of the pre-therapy scan. The results also demonstrate the benefits of using the ratio of

K

trans

with the bolus arrival time as a biomarker.

The diagnosis and treatment of fatty liver disease requires accurate quantification of the amount of fat in the liver. Image-based methods for quantification of liver fat are of increasing interest due to the high sampling error and invasiveness associated with liver biopsy, which despite these difficulties remains the gold standard. Current computed tomography (CT) methods for liverfat quantification are only semi-quantitative and infer the concentration of liver fat heuristically. Furthermore, these techniques are only applicable to images acquired without the use of contrast agent, even though contrast-enhanced CT imaging is more prevalent in clinical practice. In this paper, we introduce a method that allows for direct quantification of liver fat for both contrast-free and contrastenhanced CT images. Phantom and patient data are used for validation, and we conclude that our algorithm allows for highly accurate and repeatable quantification of liver fat for spectral CT.

CT image-based diagnosis of the stomach is developed as a new way of diagnostic method. A virtual unfolded (VU) view is suitable for displaying its wall. In this paper, we propose a semi-automated method for generating VU views of the stomach. Our method requires minimum manual operations. The determination of the unfolding forces and the termination of the unfolding process are automated. The unfolded shape of the stomach is estimated based on its radius. The unfolding forces are determined so that the stomach wall is deformed to the expected shape. The iterative deformation process is terminated if the difference of the shapes between the deformed shape and expected shape is small. Our experiments using 67 CT volumes showed that our proposed method can generate good VU views for 76.1% cases.

Kidney lesions are important extracolonic findings at computed tomographic colonography (CTC). However, kidney lesion detection on non-contrast CTC images poses significant challenges due to low image contrast with surrounding tissues. In this paper, we treat the kidney surface as manifolds in Riemannian space and present an intrinsic manifold diffusion approach to identify lesion-caused protrusion while simultaneously removing geometrical noise on the manifolds. Exophytic lesions (those that deform the kidney surface) are detected by searching for surface points with local maximum diffusion response and using the normalized cut algorithm to extract them. Moreover, multi-scale diffusion response is a discriminative feature descriptor for the subsequent classification to reduce false positives. We validated the proposed method and compared it with a baseline method using shape index on CTC datasets from 49 patients. Free-response receiver operating characteristic analysis showed that at 7 false positives, the proposed method achieved 87% sensitivity while the baseline method achieved only 22% sensitivity. The proposed method showed far fewer false positives compared with the baseline method which makes it feasible for clinical practice.

The use of a passive MRI-visible tracking frame is a common method of localizing devices in MRI space for MRI-guided procedures. One of the most common tracking frame designs found in the literature is the z-frame, as it allows six degree-of-freedom pose estimation using only a single image slice. Despite the popularity of this design, it is susceptible to errors in pose estimation due to various image distortion mechanisms in MRI. In this paper, the absolute error in using a z-frame to localize a tool in MRI is quantified over various positions of the z-frame relative to the MRI isocenter, and for various levels of static magnetic field inhomogeneity. It was found that the error increases rapidly with distance from the isocenter in both the horizontal and vertical directions, but the error is much less sensitive to position when multiple contiguous slices are used with slice-select gradient nonlinearity correction enabled, as opposed to the more common approach of only using a single image slice. In addition, the error is found to increase rapidly with an increasing level of static field inhomogeneity, even with the z-frame placed within 10 cm of the isocenter.

Minimally invasive procedures are of growing importance in modern surgery. Navigation and orientation are major issues during these interventions as conventional endoscopes only cover a limited field of view. We propose the application of a Time-of-Flight (ToF) satellite camera at the zenith of the pneumoperitoneum to survey the operation situs. Due to its limited field of view we propose a fusion of different 3-D views to reconstruct the situs using photometric and geometric information provided by the ToF sensor. We were able to reconstruct the entire abdomen with a mean absolute mesh-to-mesh error of less than 5 mm compared to CT ground truth data, at a frame rate of 3 Hz. The framework was evaluated on real data from a miniature ToF camera in an open surgery pig study and for quantitative evaluation with a realistic human phantom. With the proposed approach to operation situs reconstruction we improve the surgeons’ orientation and navigation and therefore increase safety and speed up surgical interventions.

We propose the development and assessment of a multi-section continuum robot for endoscopic surgical clipping of intracranial aneurysms. The robot has two sections for bending actuated by tendon wires. By actuating the two sections independently, the robot can generate a variety of posture combinations by these sections while maintaining the tip angle. This feature offers more flexibility in positioning of the tip than a conventional endoscope for large viewing angles of up to 180 degrees. To estimate the flexible positioning of the tip, we developed kinematic mapping with friction in tendon wires. In a kinematic-mapping simulation, the two-section robot at the target scale (i.e., an outer diameter of 1.7 mm and a length of 60 mm) had a variety of tip positions within 50-mm ranges at the 180°-angled view. In the experimental validation, the 1:10 scale prototype performed the three salient postures with different tip positions at the 180°-angled view.

The screening of oesophageal adenocarcinoma involves obtaining biopsies at different regions along the oesophagus. The localization and tracking of these biopsy sites inter-operatively poses a significant challenge for providing targeted treatments. This paper presents a novel framework for providing a guided navigation to the gastro-intestinal specialist for accurate re-positioning of the endoscope at previously targeted sites. Firstly, we explain our approach for the application of electromagnetic tracking in acheiving this objective. Then, we show on three

in-vivo

porcine interventions that our system can provide accurate guidance information, which was qualitatively evaluated by five experts.

In this paper, we present an image guidance system for abdominal aortic aneurysm stenting, which brings pre-operative 3-D computed tomography (CT) into the operating room by registering it against intra-operative non-contrast-enhanced cone-beam CT (CBCT). Registration between CT and CBCT volumes is a challenging task due to two factors: the relatively low signal-to-noise ratio of the abdominal aorta in CBCT without contrast enhancement, and the drastically different field of view between the two image modalities. The proposed automatic registration method handles the first issue through a fast quasi-global search utilizing surrogate 2-D images, and solves the second problem by relying on neighboring dominant structures of the abdominal aorta (i.e. the spine) for initial coarse alignment, and using a confined and image-processed volume of interest around the abdominal aorta for fine registration. The proposed method is validated offline using 17 clinical datasets, and achieves 1.48 mm target registration error and 100% success rate in 2.83 s. The prototype system has been installed in hospitals for clinical trial and applied in around 30 clinical cases, with 100% success rate reported qualitatively.

This paper develops a new bronchoscopic ultrasound navigation system that fuses multimodal sensory information including pre-operative images, bronchoscopic video sequences, ultrasound images, and external position sensor measurements. To construct such a system, we must align these information coordinate systems. We use hand-eye calibration to align the video camera and its attached external sensor and introduce a phantom-free method to calibrate the ultrasonic probe and its fixed external sensor. More importantly, we propose a marker-free registration method that uses the bronchoscope and the bronchial tree center information to register the sensor and the pre-operative coordinate systems. We constructed a bronchial phantom to validate our system, whose navigation accuracy was about 2.6 mm. Furthermore, compared to the current navigated bronchoscopy, the main advantage of our system is that it navigates the bronchoscope and the ultrasonic mini probe simultaneously and provides bronchial structures inside and outside the bronchial walls, particularly lymph node structures in ultrasonic images.

Tubular structures are frequently encountered in bio-medical images. The center-lines of these tubules provide an accurate representation of the topology of the structures. We introduce a stochastic Marked Point Process framework for fully automatic extraction of tubular structures requiring no user interaction or seed points for initialization. Our Marked Point Process model enables unsupervised network extraction by fitting a configuration of objects with globally optimal associated energy to the centreline of the arbors. For this purpose we propose special configurations of marked objects and an energy function well adapted for detection of 3D tubular branches. The optimization of the energy function is achieved by a stochastic, discrete-time multiple birth and death dynamics. Our method finds the centreline, local width and orientation of neuronal arbors and identifies critical nodes like bifurcations and terminals. The proposed model is tested on 3D light microscopy images from the DIADEM data set with promising results.

Endomicroscopy allows in vivo and in situ imaging with cellular resolution. One limitation of endomicroscopy is the small field of view which can however be extended using mosaicing techniques. In this paper, we describe a methodological framework aiming to reconstruct a mosaic of endomicroscopic images acquired following a noisy robotized spiral trajectory. First, we infer the topology of the frames, that is the map of neighbors for every frame in the spiral. For this, we use a Viterbi algorithm considering every new acquired frame in the current branch of the spiral as an observation and the index of the best neighboring frame from the previous branch as the underlying state. Second, the estimated transformation between each spatial pair previously found is assessed. Mosaicing is performed based only on the pairs of frames for which the registration is considered successful. We tested our method on 3 spiral endomicroscopy videos each including more than 200 frames: a printed grid, an ex vivo tissue sample and an in vivo animal trial. Results were statistically significantly improved compared to reconstruction where only registration between successive frames was used.

Quantitative measurements of spatial arrangement of nuclei in histopathology images for different cancers has been shown to have prognostic value. Traditionally, graph algorithms (with cell/nuclei as node) have been used to characterize the spatial arrangement of these cells. However, these graphs inherently extract only global features of cell or nuclear architecture and, therefore, important information at the local level may be left unexploited. Additionally, since the graph construction does not draw a distinction between nuclei in the stroma or epithelium, the graph edges often traverse the stromal and epithelial regions. In this paper, we present a new spatially aware cell cluster (SpACCl) graph that can efficiently and accurately model local nuclear interactions, separately within the stromal and epithelial regions alone. SpACCl is built locally on nodes that are defined on groups/clusters of nuclei rather than individual nuclei. Local nodes are connected with edges which have a certain probability of connectedness. The SpACCl graph allows for exploration of (a) contribution of nuclear arrangement within the stromal and epithelial regions separately and (b) combined contribution of stromal and epithelial nuclear architecture in predicting disease aggressiveness and patient outcome. In a cohort of 160 p16+ oropharyngeal tumors (141 non-progressors and 19 progressors), a support vector machine (SVM) classifier in conjunction with 7 graph features extracted from the SpACCl graph yielded a mean accuracy of over 90% with PPV of 89.4% in distinguishing between progressors and non-progressors. Our results suggest that (a) stromal nuclear architecture has a role to play in predicting disease aggressiveness and that (b) combining nuclear architectural contributions from the stromal and epithelial regions yields superior prognostic accuracy compared to individual contributions from stroma and epithelium alone.

This paper proposes a new image restoration method for phase contrast microscopy as a mean to enhance the quality of images prior to image analysis. Compared to state-of-the-art image restoration algorithms, our method has a more solid theoretical foundation and is orders of magnitude more efficient in computation. We validated the proposed method by applying it to automated muscle myotube detection, a challenging problem that has not been tackled without staining images. Results on 300 phase contrast microscopy images from three different culture conditions demonstrate that the proposed restoration scheme improves myotube detection, and that our approach is far more computationally efficient than previous methods.

We proposed a generative probabilistic modeling framework for automated segmentation of retinal layers from Optical Coherence Tomography (OCT) data. The objective is to learn a segmentation protocol from a collection of training images that have been manually labeled. Our model results in a novel OCT retinal layer segmentation approach which integrates algorithms of simultaneous searching of multiple interacting layer interfaces, image registration and machine learning. Different from previous work, our approach combines the benefits of constraining spatial layout of retinal layers, using a set of more robust local image descriptors, employing a mechanism for learning from manual labels and incorporating the inter-subject anatomical similarities of retina. With a set of OCT volumetric images from mutant canine retinas, we experimentally validated that our approach outperforms two state-of-the-art techniques.

The Ki-67 labeling index is a valid and important biomarker to gauge neuroendocrine tumor cell progression. Automatic Ki-67 assessment is very challenging due to complex variations of cell characteristics. In this paper, we propose an integrated learning-based framework for accurate Ki-67 scoring in pancreatic neuroendocrine tumor. The main contributions of our method are: a novel and robust cell detection algorithm is designed to localize both tumor and non-tumor cells; a repulsive deformable model is applied to correct touching cell segmentation; a two stage learning-based scheme combining cellular features and regional structure information is proposed to differentiate tumor from non-tumor cells (such as lymphocytes); an integrated automatic framework is developed to accurately assess the Ki-67 labeling index. The proposed method has been extensively evaluated on 101 tissue microarray (TMA) whole discs, and the cell detection performance is comparable to manual annotations. The automatic Ki-67 score is very accurate compared with pathologists’ estimation.

We address the problem of cell segmentation in confocal microscopy membrane volumes of the ascidian

Ciona

used in the study of morphogenesis. The primary challenges are non-uniform and patchy membrane staining and faint spurious boundaries from other organelles (e.g. nuclei). Traditional segmentation methods incorrectly attach to faint boundaries producing spurious edges. To address this problem, we propose a linear optimization framework for the joint correction of multiple over-segmentations obtained from different methods. The main idea motivating this approach is that multiple over-segmentations, resulting from a pool of methods with various parameters, are likely to agree on the correct segment boundaries, while spurious boundaries are method- or parameter-dependent. The challenge is to make an optimized decision on selecting the correct boundaries while discarding the spurious ones. The proposed unsupervised method achieves better performance than state of the art methods for cell segmentation from membrane images.

In this paper we describe an algorithm for accurately segmenting the individual cytoplasm and nuclei from a clump of overlapping cervical cells. Current methods cannot undertake such a complete segmentation due to the challenges involved in delineating cells with severe overlap and poor contrast. Our approach initially performs a scene segmentation to highlight both free-lying cells, cell clumps and their nuclei. Then cell segmentation is performed using a joint level set optimization on all detected nuclei and cytoplasm pairs. This optimisation is constrained by the length and area of each cell, a prior on cell shape, the amount of cell overlap and the expected gray values within the overlapping regions. We present quantitative nuclei detection and cell segmentation results on a database of synthetically overlapped cell images constructed from real images of free-lying cervical cells. We also perform a qualitative assessment of complete fields of view containing multiple cells and cell clumps.

We propose a method for targeted segmentation that identifies and delineates only those spatially-recurring objects that conform to specific geometrical, topological and appearance priors. By adopting a “tribes”-based, global genetic algorithm, we show how we incorporate such priors into a faithful objective function unconcerned about its convexity. We evaluated our framework on a variety of histology and microscopy images to segment potentially overlapping cells with complex topology. Our experiments confirmed the generality, reproducibility and improved accuracy of our approach compared to competing methods.

The use of human embryonic stem cell cardiomyocytes (hESC-CMs) in tissue transplantation and repair has led to major recent advances in cardiac regenerative medicine. However, to avoid potential arrhythmias, it is critical that hESC-CMs used in replacement therapy be electrophysiologically compatible with the adult atrial, ventricular, and nodal phenotypes. The current method for classifying the electrophysiology of hESC-CMs relies mainly on the shape of the cell’s action potential (AP), which each expert subjectively decides if it is nodal-like, atrial-like or ventricular-like. However, the classification is difficult because the shape of the AP of an hESC-CMs may not coincide with that of a mature cell. In this paper, we propose to use a

metamorphosis distance

for comparing the AP of an hESC-CMs to that of an adult cell model. This involves constructing a family of APs corresponding to different stages of the maturation process, and measuring the amount of deformation between APs. Experiments show that the proposed distance leads to better interpolation and classification results.

We propose a novel two-layer level set approach for segmentation of the left ventricle (LV) from cardiac magnetic resonance (CMR) short-axis images. In our method, endocardium and epicardium are represented by two specified level contours of a level set function. Segmentation of the LV is formulated as a problem of optimizing the level set function such that these two level contours best fit the epicardium and endocardium. More importantly, a distance regularization (DR) constraint on the level contours is introduced to preserve smoothly varying distance between them. This DR constraint leads to a desirable interaction between the level contours that contributes to maintain the anatomical geometry of the endocardium and epicardium. The negative influence of intensity inhomogeneities on image segmentation are overcome by using a data term derived from a local intensity clustering property. Our method is quantitatively validated by experiments on the datasets for the MICCAI grand challenge on left ventricular segmentation, which demonstrates the advantages of our method in terms of segmentation accuracy and consistency with anatomical geometry.

The aortic valve has been described with variable anatomical definitions, and the consistency of 2D manual measurement of valve dimensions in medical image data has been questionable. Given the importance of image-based morphological assessment in the diagnosis and surgical treatment of aortic valve disease, there is considerable need to develop a standardized framework for 3D valve segmentation and shape representation. Towards this goal, this work integrates template-based medial modeling and multi-atlas label fusion techniques to automatically delineate and quantitatively describe aortic leaflet geometry in 3D echocardiographic (3DE) images, a challenging task that has been explored only to a limited extent. The method makes use of expert knowledge of aortic leaflet image appearance, generates segmentations with consistent topology, and establishes a shape-based coordinate system on the aortic leaflets that enables standardized automated measurements. In this study, the algorithm is evaluated on 11 3DE images of normal human aortic leaflets acquired at mid systole. The clinical relevance of the method is its ability to capture leaflet geometry in 3DE image data with minimal user interaction while producing consistent measurements of 3D aortic leaflet geometry.

Quantitative motion analysis from cardiac imaging is important to study the function of heart. Most of existing image-based motion estimation methods model the myocardium as an isotropically elastic continuum. We propose a novel anisotropic regularization method which enforces the transmural homogeneity of the strain along myofiber. The myofiber orientation in the end-diastolic frame is obtained by registering it with a diffusion tensor atlas. Our method is formulated in a diffeomorphic registration framework, and tested on multimodal cardiac image sequences of two subjects using 3D echocardiography and cine and tagged MRI. Results show that the estimated transformations in our method are more smooth and more accurate than those in isotropic regularization.

While tomographic imaging of cardiac structure and kinetics has improved substantially, electrophysiological mapping of the heart is still restricted to the body or heart surface with little or no depth information beneath. The progress in reconstructing transmural action potentials from surface voltage data has been hindered by the challenges of intrinsic ill-posedness and the lack of a unique solution in the absence of prior assumptions. In this work, we propose to exploit the unique spatial property of transmural action potentials that it is often piecewise smooth with a steep boundary (gradient) separating the depolarized and repolarized regions. This steep gradient could reveal normal or disrupted electrical propagation wavefronts, or pinpoint the border between viable and necrotic tissue. In this light, we propose a novel adaption of the total-variation (TV) prior into the reconstruction of transmural action potentials, where a variational TV operator is defined instead of a common discrete operator, and the TV-minimization is solved by a sequence of weighted, first-order L2-norm minimizations. In a large set of phantom experiments performed on image-derived human heart-torso models, the proposed method is shown to outperform existing quadratic methods in preserving the steep gradient of action potentials along the border of infarcts, as well as in capturing the disruption to the normal path of electrical wavefronts. The former is further attested by real-data experiments on two post-infarction human subjects, demonstrating the potential of the proposed method in revealing the location and shape of the underlying infarcts when existing quadratic methods fail to do so.

We propose a fast algorithm for 3D segmentation of the right ventricle (RV) in MRI using shape and appearance constraints based on probability product kernels (PPK). The proposed constraints remove the need for large, manually-segmented training sets and costly pose estimation (or registration) procedures, as is the case of the existing algorithms. We report comprehensive experiments, which demonstrate that the proposed algorithm (i) requires only a single subject for training; and (ii) yields a performance that is not significantly affected by the choice of the training data. Our PPK constraints are non-linear (high-order) functionals, which are not directly amenable to standard optimizers. We split the problem into several surrogate-functional optimizations, each solved via an efficient convex relaxation that is amenable to parallel implementations. We further introduce a scale variable that we optimize with fast fixed-point computations, thereby achieving pose invariance in real-time. Our parallelized implementation on a graphics processing unit (GPU) demonstrates that the proposed algorithm can yield a real-time solution for typical cardiac MRI volumes, with a speed-up of more than 20 times compared to the CPU version. We report a comprehensive experimental validations over 400 volumes acquired from 20 subjects, and demonstrate that the obtained 3D surfaces correlate with independent manual delineations.

We propose a learning-based method for robust tracking in long ultrasound sequences for image guidance applications. The framework is based on a scale-adaptive block-matching and temporal realignment driven by the image appearance learned from an initial training phase. The latter is introduced to avoid error accumulation over long sequences. The vessel tracking performance is assessed on long 2D ultrasound sequences of the liver of 9 volunteers under free breathing. We achieve a mean tracking accuracy of 0.96 mm. Without learning, the error increases significantly (2.19 mm, p<0.001).

We present a novel, fully-discriminative method for curvilinear structure segmentation that simultaneously learns a classifier and the features it relies on. Our approach requires almost no parameter tuning and, in the case of 2D images, removes the requirement for hand-designed features, thus freeing the practitioner from the time-consuming tasks of parameter and feature selection. Our approach relies on the Gradient Boosting framework to learn discriminative convolutional filters in closed form at each stage, and can operate on raw image pixels as well as additional data sources, such as the output of other methods like the Optimally Oriented Flux. We will show that it outperforms state-of-the-art curvilinear segmentation methods on both 2D images and 3D image stacks.

We propose a novel algorithm to jointly delineate the femoral artery lumen and outer wall surfaces from 3D black-blood MR images, while enforcing the spatial consistency of the reoriented MR slices along the medial axis of the femoral artery. We demonstrate that the resulting optimization problem of the proposed segmentation can be solved globally and exactly by means of convex relaxation, for which we introduce a novel

coupled continuous max-flow (CCMF) model

based on an Ishikawa-type flow configuration and show its duality to the studied convex relaxed optimization problem. Using the proposed

CCMF model

, the exactness and globalness of its dual convex relaxation problem is proven. Experiment results demonstrate that the proposed method yielded high accuracy (i.e. Dice similarity coefficient > 85%) for both the lumen and outer wall and high reproducibility (intra-class correlation coefficient of 0.95) for generating vessel wall area. The proposed method outperformed the previous method, in terms of computation time, by a factor of ~20.

Extracting centerlines of coronary arteries is a challenging but important task in clinical applications of cardiac CTA. In this paper, we propose a model-guided approach, the directional minimal path, for the centerline extraction. The proposed method is based on the minimal path algorithm and a prior coronary model is used. The model is first registered to the unseen image. Then, the start point and end point for the minimal path algorithm are provided by the model to automate the centerline extraction process. Also, the direction information of the coronary model is used to guide the path tracking of the minimal path procedure. This directional tracking improves the robustness and accuracy of the centerline extraction. Finally, the proposed method can automatically recognize the branches of the extracted coronary artery using the prior information in the model. We validated the proposed method by extracting the three main coronary branches. The mean accuracy of the 56 cases was 1.32±0.81 mm and the detection ratio was 88.7%.

We introduce a novel fast marching approach with

curvature

regularization for vessel segmentation. Since most vessels have a smooth path, curvature can be used to distinguish desired vessels from short cuts, which usually contain parts with high curvature. However, in previous fast marching approaches, curvature information is not available, so it cannot be used for regularization directly. Instead, usually

length

regularization is used under the assumption that shorter paths should also have a lower curvature. However, for vessel segmentation, this assumption often does not hold and leads to short cuts. We propose an approach, which integrates curvature regularization directly into the fast marching framework, independent of length regularization. Our approach is

globally optimal

, and numerical experiments on synthetic and real retina images show that our approach yields more accurate results than two previous approaches.

Blood flow measurements using 4D Phase Contrast blood flow imaging (PCI) provide an excellent fully non-invasive technique to assess the hemodynamics clinically in-vivo. Iterative reconstruction techniques combined with parallel MRI have been proposed to reduce the data acquisition time, which is the biggest drawback of 4D PCI. The novel LoSDeCoS technique combines these ideas with the separation into a low-rank and a sparse component. The high-dimensionality of the PC data renders it ideally suited for this approach. The proposed method is not limited to a single body region, but can be applied to any 4D flow measurement. The benefits of the new method are twofold: It allows to significantly accelerate the acquisition; and generates additional images highlighting temporal and directional flow changes. Reduction in acquisition time improves patient comfort and can be used to achieve better temporal or spatial resolution, which in turn allows more precise calculations of clinically important quantitative numbers such as flow rates or the wall shear stress. With LoSDeCoS, acceleration factors of 6-8 were achieved for 16 in-vivo datasets of both the carotid artery (6 datasets) and the aorta (10 datasets), while decreasing the Normalized Root Mean Square Error by over 10 % compared to a standard iterative reconstruction and by achieving similarity values of over 0.93. Inflow-Outflow phantom experiments showed good parabolic profiles and an excellent mass conservation.

A new method for anatomically labeling the vasculature is presented and applied to the Circle of Willis. Our method converts the segmented vasculature into a graph that is matched with an annotated graph atlas in a maximum a posteriori (MAP) way. The MAP matching is formulated as a quadratic binary programming problem which can be solved efficiently. Unlike previous methods, our approach can handle non tree-like vasculature and large topological differences. The method is evaluated in a leave-one-out test on MRA of 30 subjects where it achieves a sensitivity of 93% and a specificity of 85% with an average error of 1.5 mm on matching bifurcations in the vascular graph.

Diffusion tensor imaging, tractography and the subsequent derivation of network measures are becoming an established approach in the exploration of brain connectivity. However, no gold standard exists in respect to how the brain should be parcellated and therefore a variety of atlas- and random-based parcellation methods are used. The resulting challenge of comparing graphs with differing numbers of nodes and uncertain node correspondences necessitates the use of normalisation schemes to enable meaningful intra- and inter-subject comparisons. This work proposes methods for normalising brain network measures using random graphs. We show that the normalised measures are locally stable over distinct random parcellations of the same subject and, applying it to a neonatal serial diffusion MRI data set, we demonstrate their potential in characterising changes in brain connectivity during early development.

Fetal MRI is a rapidly emerging diagnostic imaging tool. Its main focus is currently on brain imaging, but there is a huge potential for whole body studies. We propose a method for accurate and robust localisation of the fetal brain in MRI when the image data is acquired as a stack of 2D slices misaligned due to fetal motion. We first detect possible brain locations in 2D images with a Bag-of-Words model using SIFT features aggregated within Maximally Stable Extremal Regions (called bundled SIFT), followed by a robust fitting of an axis-aligned 3D box to the selected regions. We rely on prior knowledge of the fetal brain development to define size and shape constraints. In a cross-validation experiment, we obtained a median error distance of 5.7mm from the ground truth and no missed detection on a database of 59 fetuses. This 2D approach thus allows a robust detection even in the presence of substantial fetal motion.

In this article we propose to investigate the analogy between early cortical folding process and cortical smoothing by mean curvature flow. First, we introduce a one-parameter model that is able to fit a developmental trajectory as represented in a Volume-Area plot and we propose an efficient optimization strategy for parameter estimation. Second, we validate the model on forty cortical surfaces of preterm newborns by comparing global geometrical indices and trajectories of central sulcus along developmental and simulation time.

We present a novel surface parameterization technique using

hyperspherical harmonics

(HSH) in representing compact, multiple, disconnected brain subcortical structures as a single analytic function. The proposed hyperspherical harmonic representation (HyperSPHARM) has many advantages over the widely used spherical harmonic (SPHARM) parameterization technique. SPHARM requires flattening 3D surfaces to 3D sphere which can be time consuming for large surface meshes, and can’t represent multiple disconnected objects with single parameterization. On the other hand, HyperSPHARM treats 3D object, via simple stereographic projection, as a surface of 4D hypersphere with extremely large radius, hence avoiding the computationally demanding flattening process. HyperSPHARM is shown to achieve a better reconstruction with only 5 basis compared to SPHARM that requires more than 441.

We propose a general database-driven framework for coherent synthesis of subject-specific scans of desired modality, which adopts and generalizes the patch-based label propagation (LP) strategy. While modality synthesis has received increased attention lately, current methods are mainly tailored to specific applications. On the other hand, the LP framework has been extremely successful for certain segmentation tasks, however, so far it has not been used for estimation of entities other than categorical segmentation labels. We approach the synthesis task as a

modality propagation

, and demonstrate that with certain modifications the LP framework can be generalized to continuous settings providing coherent synthesis of different modalities, beyond segmentation labels. To achieve high-quality estimates we introduce a new data-driven regularization scheme, in which we integrate intermediate estimates within an iterative search-and-synthesis strategy. To efficiently leverage population data and ensure coherent synthesis, we employ a spatio-population search space restriction. In experiments, we demonstrate the quality of synthesis of different MRI signals (T2 and DTI-FA) from a T1 input, and show a novel application of modality synthesis for abnormality detection in multi-channel MRI of brain tumor patients.

Myelin is an essential component of nerve fibers and monitoring its health is important for studying diseases that attack myelin, such as multiple sclerosis (MS). The amount of water trapped within myelin, which is a surrogate for myelin content and integrity, can be measured

in vivo

using MRI relaxation techniques that acquire a series of images at multiple echo times to produce a T

2

decay curve at each voxel. These curves are then analyzed, most commonly using non-negative least squares (NNLS) fitting, to produce T

2

distributions from which water measurements are made. NNLS is unstable with respect to the noise and variations found in typical T

2

relaxation images, making some form of regularization inevitable. The current methods of NNLS regularization for measuring myelin water have two key limitations: 1) they use strictly local neighborhood information to regularize each voxel, which limits their effectiveness for very noisy images, and 2) the neighbors of each voxel contribute to its regularization equally, which can over-smooth fine details. To overcome these limitations, we propose a new regularization algorithm in which local and non-local information is gathered and used adaptively for each voxel. Our results demonstrate that the proposed method provides more globally consistent myelin water measurements yet preserves fine structures. Our experiment with real patient data also shows that the algorithm improves the ability to distinguish two sample groups, one of MS patients and the other of healthy subjects.

Demyelinating diseases such as multiple sclerosis cause changes in the brain white matter microstructure. Multi-exponential T2 relaxometry is a powerful technology for detecting these changes by generating a myelin water fraction (MWF) map. However, conventional approaches are subject to noise and spatial in-consistence. We proposed a novel approach by imposing spatial consistency and smoothness constraints. We first introduce a two-Gaussian model to approximate the T2 distribution. Then an expectation-maximization framework is introduced with an edge-preserving prior incorporated. Three-dimensional multi-echo MRI data sets were collected from three patients and three healthy volunteers. MWF maps ob-tained using the conventional, Spatially Regularized Non-negative Least Squares (srNNLS) algorithm as well as the proposed algorithm are compared. The proposed method provides MWF maps with improved depiction of brain structures and significantly lower coefficients of variance in various brain regions.

Availability of multi-modal magnetic resonance imaging (MRI) databases opens up the opportunity to synthesize different MRI contrasts without actually acquiring the images. In theory such synthetic images have the potential to reduce the amount of acquisitions to perform certain analyses. However, to what extent they can substitute real acquisitions in the respective analyses is an open question. In this study, we used a synthesis method based on patch matching to test whether synthetic images can be useful in segmentation and inter-modality cross-subject registration of brain MRI. Thirty-nine T1 scans with 36 manually labeled structures of interest were used in the registration and segmentation of eight proton density (PD) scans, for which ground truth T1 data were also available. The results show that synthesized T1 contrast can considerably enhance the quality of non-linear registration compared with using the original PD data, and it is only marginally worse than using the original T1 scans. In segmentation, the relative improvement with respect to using the PD is smaller, but still statistically significant.

Compressed Sensing (CS) takes advantage of signal sparsity or compressibility and allows superb signal reconstruction from relatively few measurements. Based on CS theory, a suitable dictionary for sparse representation of the signal is required. In diffusion MRI (dMRI), CS methods proposed for reconstruction of diffusion-weighted signal and the Ensemble Average Propagator (EAP) utilize two kinds of Dictionary Learning (DL) methods: 1) Discrete Representation DL (DR-DL), and 2) Continuous Representation DL (CR-DL). DR-DL is susceptible to numerical inaccuracy owing to interpolation and regridding errors in a discretized

q

-space. In this paper, we propose a novel CR-DL approach, called Dictionary Learning - Spherical Polar Fourier Imaging (DL-SPFI) for effective compressed-sensing reconstruction of the

q

-space diffusion-weighted signal and the EAP. In DL-SPFI, a dictionary that sparsifies the signal is learned from the space of continuous Gaussian diffusion signals. The learned dictionary is then adaptively applied to different voxels using a weighted LASSO framework for robust signal reconstruction. Compared with the start-of-the-art CR-DL and DR-DL methods proposed by Merlet et al. and Bilgic et al., respectively, our work offers the following advantages. First, the learned dictionary is proved to be optimal for Gaussian diffusion signals. Second, to our knowledge, this is the first work to learn a voxel-adaptive dictionary. The importance of the adaptive dictionary in EAP reconstruction will be demonstrated theoretically and empirically. Third, optimization in DL-SPFI is only performed in a small subspace resided by the SPF coefficients, as opposed to the

q

-space approach utilized by Merlet et al. We experimentally evaluated DL-SPFI with respect to L1-norm regularized SPFI (L1-SPFI), which uses the original SPF basis, and the DR-DL method proposed by Bilgic et al. The experiment results on synthetic and real data indicate that the learned dictionary produces sparser coefficients than the original SPF basis and results in significantly lower reconstruction error than Bilgic et al.’s method.

The main contribution of this work is the careful syntactical definition of major white matter tracts in the human brain based on a neuroanatomist’s expert knowledge. We present a technique to formally describe white matter tracts and to automatically extract them from diffusion MRI data. The framework is based on a novel query language with a near-to-English textual syntax. This query language allows us to construct a dictionary of anatomical definitions describing white matter tracts. The definitions include adjacent gray and white matter regions, and rules for spatial relations. This enables automated coherent labeling of white matter anatomy across subjects. We use our method to encode anatomical knowledge in human white matter describing 10 association and 8 projection tracts per hemisphere and 7 commissural tracts. The technique is shown to be comparable in accuracy to manual labeling. We present results applying this framework to create a white matter atlas from 77 healthy subjects, and we use this atlas in a proof-of-concept study to detect tract changes specific to schizophrenia.

Human brain connectivity can be studied using graph theory. Many connectivity studies parcellate the brain into regions and count fibres extracted between them. The resulting network analyses require validation of the tractography, as well as region and parameter selection. Here we investigate whole brain connectivity from a different perspective. We propose a mathematical formulation based on studying the eigenvalues of the Laplacian matrix of the diffusion tensor field at the voxel level. This voxelwise matrix has over a million parameters, but we derive the Kirchhoff complexity and eigen-spectrum through elegant mathematical theorems, without heavy computation. We use these novel measures to accurately estimate the voxelwise connectivity in multiple biomedical applications such as Alzheimer’s disease and intelligence prediction.

Spherical deconvolution models the diffusion MRI signal as the convolution of a fiber orientation density function (fODF) with a single fiber response. We propose a novel calibration procedure that automatically determines this fiber response. This has three advantages: First, the user no longer needs to provide an estimate of the response. Second, we estimate a per-voxel fiber response, which is more adequate for the analysis of patient data with focal white matter degeneration. Third, parameters of the estimated response reflect diffusion properties of the white matter tissue, and can be used for quantitative analysis.

Our method works by finding a tradeoff between a low fitting error and a sparse fODF. Results on simulated data demonstrate that auto-calibration successfully avoids erroneous fODF peaks that can occur with standard deconvolution, and that it resolves fiber crossings with better angular resolution than FORECAST, an alternative method. Parameter maps and tractography results corroborate applicability to clinical data.

Brain networks are becoming forefront research in neuroscience. Network-based analysis on the functional and structural connectomes can lead to powerful imaging markers for brain diseases. However, constructing the structural connectome can be based upon different acquisition and reconstruction techniques whose information content and mutual differences has not yet been properly studied in a unified framework. The variations of the structural connectome if not properly understood can lead to dangerous conclusions when performing these type of studies. In this work we present evaluation of the structural connectome by analysing and comparing graph-based measures on real data acquired by the three most important Diffusion Weighted Imaging techniques: DTI, HARDI and DSI. We thus come to several important conclusions demonstrating that even though the different techniques demonstrate differences in the anatomy of the reconstructed fibers the respective connectomes show variations of 20%.

This paper extends Jones’ popular electrostatic repulsion based algorithm for distribution of single-shell Q-space samples in two fundamental ways. The first alleviates the single-shell requirement enabling full Q-space sampling. Such an extension is not immediately obvious since it requires distributing samples evenly in 3 dimensions. The extension is as elegant as it is simple: Add a container volume of the desired shape having a constant charge density and a total charge equal to the negative of the sum of the moving point charges. Results for spherical and cubic charge containers are given. The second extension concerns the way distances between sample point are measured. The Q-space samples represent orientation, rather than direction and it would seem appropriate to use a metric that reflects this fact, e.g. a tensor metric. To this end we present a means to employ a generalized metric in the optimization. Minimizing the energy will result in a 3-dimensional distribution of point charges that is uniform in the terms of the specified metric. The radically different distributions generated using different metrics pinpoints a fundamental question: Is there an inherent optimal metric for Q-space sampling? Our work provides a versatile tool to explore the role of different metrics and we believe it will be an important contribution to further the continuing debate and research on the matter.

Several data acquisition schemes for diffusion MRI have been proposed and explored to date for the reconstruction of the 2nd order tensor. Our main contributions in this paper are: (i) the definition of a new class of sampling schemes based on repeated measurements in every sampling point; (ii) two novel schemes belonging to this class; and (iii) a new reconstruction framework for the second scheme. We also present an evaluation, based on Monte Carlo computer simulations, of the performances of these schemes relative to known optimal sampling schemes for both 2nd and 4th order tensors. The results demonstrate that tensor estimation by the proposed sampling schemes and estimation framework is more accurate and robust.

Diffusion tensor imaging cannot represent heterogeneous fascicle orientations in one voxel. Various models propose to overcome this limitation. Among them, multi-fascicle models are of great interest to characterize and compare white matter properties. However, existing methods fail to estimate their parameters from conventional diffusion sequences with the desired accuracy. In this paper, we provide a geometric explanation to this problem. We demonstrate that there is a manifold of indistinguishable multi-fascicle models for single-shell data, and that the manifolds for different b-values intersect tangentially at the true underlying model making the estimation very sensitive to noise. To regularize it, we propose to learn a prior over the model parameters from data acquired at several b-values in an external population of subjects. We show that this population-informed prior enables for the first time accurate estimation of multi-fascicle models from single-shell data as commonly acquired in clinical context. The approach is validated on synthetic and in vivo data of healthy subjects and patients with autism. We apply it in population studies of the white matter microstructure in autism spectrum disorder. This approach enables novel investigations from large existing DWI datasets in normal development and in disease.

Segmentation of infant brain MR images is challenging due to insufficient image quality, severe partial volume effect, and ongoing maturation and myelination process. During the first year of life, the signal contrast between white matter (WM) and gray matter (GM) in MR images undergoes inverse changes. In particular, the inversion of WM/GM signal contrast appears around 6-8 months of age, where brain tissues appear isointense and hence exhibit extremely low tissue contrast, posing significant challenges for automated segmentation. In this paper, we propose a novel segmentation method to address the above-mentioned challenge based on the sparse representation of the complementary tissue distribution information from T1, T2 and diffusion-weighted images. Specifically, we first derive an initial segmentation from a library of aligned multi-modality images with ground-truth segmentations by using sparse representation in a patch-based fashion. The segmentation is further refined by the integration of the geometrical constraint information. The proposed method was evaluated on 22 6-month-old training subjects using leave-one-out cross-validation, as well as 10 additional infant testing subjects, showing superior results in comparison to other state-of-the-art methods.

Groupwise segmentation that simultaneously segments a set of images and ensures that the segmentations for the same structure of interest from different images are consistent usually can achieve better performance than segmenting each image independently. Our main contribution is that we adopt the groupwise segmentation framework to improve the performance of multi-atlas label fusion. We develop a novel statistical model to allow this extension. Comparing to previous atlas propagation and groupwise segmentation work, one key novelty of our method is that the error produced during label propagation is explicitly addressed in the joint label fusion framework. Experiments on hippocampus segmentation in magnetic resonance images show the effectiveness of the new groupwise segmentation technique.

The delineation of tumor boundaries in medical images is an essential task for the early detection, diagnosis and follow-up of cancer. However accurate segmentation remains challenging due to presence of noise, inhomogeneity and high appearance variability of malignant tissue. In this paper, we propose an automatic segmentation approach using fully-connected higher-order conditional random fields (HOCRF) where potentials are computed within a discriminant Grassmannian manifold. First, the framework learns within-class and between-class similarity distributions from a training set of images to discover the optimal manifold discrimination between normal and pathological tissues. Second, the conditional optimization scheme computes non-local pairwise as well as pattern-based higher-order potentials from the manifold subspace to recognize regions with similar labelings and incorporate global consistency in the inference process. Our HOCRF framework is applied in the context of metastatic liver tumor segmentation in CT images. Compared to state of the art methods, our method achieves better performance on a group of 30 liver tumors and can deal with highly pathological cases.

In this paper we propose a method for whole brain parcellation using the type of generative parametric models typically used in tissue classification. Compared to the non-parametric, multi-atlas segmentation techniques that have become popular in recent years, our method obtains state-of-the-art segmentation performance in both cortical and subcortical structures, while retaining all the benefits of generative parametric models, including high computational speed, automatic adaptiveness to changes in image contrast when different scanner platforms and pulse sequences are used, and the ability to handle multi-contrast (vector-valued intensities) MR data. We have validated our method by comparing its segmentations to manual delineations both within and across scanner platforms and pulse sequences, and show preliminary results on multi-contrast test-retest scans, demonstrating the feasibility of the approach.

The segmentation of lesions in the brain during the development of Multiple Sclerosis is part of the diagnostic assessment for this disease and gives information on its current severity. This laborious process is still carried out in a manual or semiautomatic fashion by clinicians because published automatic approaches have not been universal enough to be widely employed in clinical practice. Thus Multiple Sclerosis lesion segmentation remains an open problem. In this paper we present a new unsupervised approach addressing this problem with dictionary learning and sparse coding methods. We show its general applicability to the problem of lesion segmentation by evaluating our approach on synthetic and clinical image data and comparing it to state-of-the-art methods. Furthermore the potential of using dictionary learning and sparse coding for such segmentation tasks is investigated and various possibilities for further experiments are discussed.

We develop a novel deformable atlas method for multi-structure segmentation that seamlessly combines the advantages of image-based and atlas-based methods. The method formulates a probabilistic framework that combines prior anatomical knowledge with image-based cues that are specific to the subject’s anatomy, and solves it using expectation-maximization method. It improves the segmentation over conventional label fusion methods especially around the structure boundaries, and is robust to large anatomical variation. The proposed method was applied to segment multiple structures in both normal and diseased brains and was shown to significantly improve results especially in diseased brains.

In this paper, we present a fully automated hierarchical probabilistic framework for segmenting brain tumours from multispectral human brain magnetic resonance images (MRIs) using multiwindow Gabor filters and an adapted Markov Random Field (MRF) framework. In the first stage, a customised Gabor decomposition is developed, based on the combined-space characteristics of the two classes (tumour and non-tumour) in multispectral brain MRIs in order to optimally separate tumour (including edema) from healthy brain tissues. A Bayesian framework then provides a coarse probabilistic texture-based segmentation of tumours (including edema) whose boundaries are then refined at the voxel level through a modified MRF framework that carefully separates the edema from the main tumour. This customised MRF is not only built on the voxel intensities and class labels as in traditional MRFs, but also models the intensity differences between neighbouring voxels in the likelihood model, along with employing a prior based on local tissue class transition probabilities. The second inference stage is shown to resolve local inhomogeneities and impose a smoothing constraint, while also maintaining the appropriate boundaries as supported by the local intensity difference observations. The method was trained and tested on the publicly available MICCAI 2012 Brain Tumour Segmentation Challenge (BRATS) Database [1] on both synthetic and clinical volumes (low grade and high grade tumours). Our method performs well compared to state-of-the-art techniques, outperforming the results of the top methods in cases of clinical high grade and low grade tumour core segmentation by 40% and 45% respectively.

New magnetic resonance imaging (MRI) sequences are enabling clinical study of the

in vivo

spinal cord’s internal structure. Yet, low contrast-to-noise ratio, artifacts, and imaging distortions have limited the applicability of tissue segmentation techniques pioneered elsewhere in the central nervous system. Recently, methods have been presented for cord/non-cord segmentation on MRI and the feasibility of gray matter/white matter tissue segmentation has been evaluated. To date, no automated algorithms have been presented. Herein, we present a non-local multi-atlas framework that robustly identifies the spinal cord and segments its internal structure with submillimetric accuracy. The proposed algorithm couples non-local fusion with a large number of slice-based atlases (as opposed to typical volumetric ones). To improve performance, the fusion process is interwoven with registration so that segmentation information guides registration and vice versa. We demonstrate statistically significant improvement over state-of-the-art benchmarks in a study of 67 patients. The primary contributions of this work are (1) innovation in non-volumetric atlas information, (2) advancement of label fusion theory to include iterative registration/segmentation, and (3) the first fully automated segmentation algorithm for spinal cord internal structure on MRI.