This PDF file contains the front matter associated with SPIE Proceedings Volume 8314, including the Title Page, Copyright information, Table of Contents, and the Conference Committee listing.

Traumatic brain injury (TBI) is a major cause of death and disability worldwide. Robust, reproducible segmentations of MR images with TBI are crucial for quantitative analysis of recovery and treatment efficacy. However, this is a significant challenge due to severe anatomy changes caused by edema (swelling), bleeding, tissue deformation, skull fracture, and other effects related to head injury. In this paper, we introduce a multi-modal image segmentation framework for longitudinal TBI images. The framework is initialized through manual input of primary lesion sites at each time point, which are then refined by a joint approach composed of Bayesian segmentation and construction of a personalized atlas. The personalized atlas construction estimates the average of the posteriors of the Bayesian segmentation at each time point and warps the average back to each time point to provide the updated priors for Bayesian segmentation. The difference between our approach and segmenting longitudinal images independently is that we use the information from all time points to improve the segmentations. Given a manual initialization, our framework automatically segments healthy structures (white matter, grey matter, cerebrospinal fluid) as well as different lesions such as hemorrhagic lesions and edema. Our framework can handle different sets of modalities at each time point, which provides flexibility in analyzing clinical scans. We show results on three subjects with acute baseline scans and chronic follow-up scans. The results demonstrate that joint analysis of all the points yields improved segmentation compared to independent analysis of the two time points.

Recently introduced combined PET/MR scanners need to handle the specific problem that a limited MR field of view sometimes truncates arm or body contours, which prevents an accurate calculation of PET attenuation correction maps. Such maps of attenuation coefficients over body structures are required for a quantitatively correct PET image reconstruction. This paper addresses this problem by presenting a method that segments a preliminary reconstruction type of PET images, time of flight non-attenuation corrected (ToF-NAC) images, and outlining a processing pipeline that compensates the arm or body truncation with this segmentation. The impact of this truncation compensation is demonstrated together with a comparison of two segmentation methods, simple gray value threshold segmentation and a watershed algorithm on a gradient image. Our results indicate that with truncation compensation a clinically tolerable quantitative SUV error is robustly achievable.

With the recent introduction of combined Magnetic Resonance Imaging (MRI) / Positron Emission Tomography (PET) systems, the generation of attenuation maps for PET based on MR images gained substantial attention. One approach for this problem is the segmentation of structures on the MR images with subsequent filling of the segments with respective attenuation values. Structures of particular interest for the segmentation are the pelvis bones, since those are among the most heavily absorbing structures for many applications, and they can serve at the same time as valuable landmarks for further structure identification. In this work the model-based segmentation of the pelvis bones on gradient-echo MR images is investigated. A processing chain for the detection and segmentation of the pelvic bones is introduced, and the results are evaluated using CT-generated "ground truth" data. The results indicate that a model based segmentation of the pelvis bone is feasible with moderate requirements to the pre- and postprocessing steps of the segmentation.

Segmentation of bone and cartilage from a three dimensional knee magnetic resonance (MR) image is a crucial element in monitoring and understanding of development and progress of osteoarthritis. Until now, various segmentation methods have been proposed to separate the bone from other tissues, but it still remains challenging problem due to different modality of MR images, low contrast between bone and tissues, and shape irregularity. In this paper, we present a new fully-automatic segmentation method of bone compartments using relevant bone atlases from a training set. To find the relevant bone atlases and obtain the segmentation, a coarse-to-fine strategy is proposed. In the coarse step, the best atlas among the training set and an initial segmentation are simultaneously detected using branch and bound tree search. Since the best atlas in the coarse step is not accurately aligned, all atlases from the training set are aligned to the initial segmentation, and the best aligned atlas is selected in the middle step. Finally, in the fine step, segmentation is conducted as adaptively integrating shape of the best aligned atlas and appearance prior based on characteristics of local regions. For experiment, femur and tibia bones of forty test MR images are segmented by the proposed method using sixty training MR images. Experimental results show that a performance of the segmentation and the registration becomes better as going near the fine step, and the proposed method obtain the comparable performance with the state-of-the-art methods.

Manual delineation of the prostate is a challenging task for a clinician due to its complex and irregular shape. Furthermore, the need for precisely targeting the prostate boundary continues to grow. Planning for radiation therapy, MR-ultrasound fusion for image-guided biopsy, multi-parametric MRI tissue characterization, and context-based organ retrieval are examples where accurate prostate delineation can play a critical role in a successful patient outcome. Therefore, a robust automated full prostate segmentation system is desired. In this paper, we present an automated prostate segmentation system for 3D MR images. In this system, the prostate is segmented in two steps: the prostate displacement and size are first detected, and then the boundary is refined by a shape model. The detection approach is based on normalized gradient fields cross-correlation. This approach is fast, robust to intensity variation and provides good accuracy to initialize a prostate mean shape model. The refinement model is based on a graph-search based framework, which contains both shape and topology information during deformation. We generated the graph cost using trained classifiers and used coarse-to-fine search and region-specific classifier training. The proposed algorithm was developed using 261 training images and tested on another 290 cases. The segmentation performance using mean DSC ranging from 0.89 to 0.91 depending on the evaluation subset demonstrates state of the art performance. Running time for the system is about 20 to 40 seconds depending on image size and resolution.

Organ motion during radiotherapy is one of causes of uncertainty in dose delivery. To cope with this, the planned target volume (PTV) has to be larger than needed to guarantee full tumor irradiation. Existing methods deal with the problem by performing tumor tracking using implanted fiducial markers or magnetic sensors. In this work, we investigate the feasibility of using x-ray based real time 2D/3D registration for non-invasive tumor motion tracking during radiotherapy. Our method uses purely intensity based techniques, thus avoiding markers or fiducials. X-rays are acquired during treatment at a rate of 5.4Hz. We iteratively compare each x-ray with a set of digitally reconstructed radiographs (DRR) generated from the planning volume dataset, finding the optimal match between the x-ray and one of the DRRs. The DRRs are generated using a ray-casting algorithm, implemented using general purpose computation on graphics hardware (GPGPU) programming techniques using CUDA for greater performance. Validation is conducted off-line using a phantom and five clinical patient data sets. The registration is performed on a region of interest (ROI) centered around the PTV. The phantom motion is measured with an rms error of 2.1 mm and mean registration time is 220 ms. For the patient data sets, a sinusoidal movement that clearly correlates to the breathing cycle is seen. Mean registration time is always under 105 ms which is well suited for our purposes. These results demonstrate that real-time organ motion monitoring using image based markerless registration is feasible.

We present an algorithm for geometric matching of graphs embedded in 2D or 3D space. It is applicable for registering any graph-like structures appearing in biomedical images, such as blood vessels, pulmonary bronchi, nerve fibers, or dendritic arbors. Our approach does not rely on the similarity of local appearance features, so it is suitable for multimodal registration with a large difference in appearance. Unlike earlier methods, the algorithm uses edge shape, does not require an initial pose estimate, can handle partial matches, and can cope with nonlinear deformations and topological differences. The matching consists of two steps. First, we find an affine transform that roughly aligns the graphs by exploring the set of all consistent correspondences between the nodes. This can be done at an acceptably low computational expense by using parameter uncertainties for pruning, backtracking as needed. Parameter uncertainties are updated in a Kalman-like scheme with each match. In the second step we allow for a nonlinear part of the deformation, modeled as a Gaussian Process. Short sequences of edges are grouped into superedges, which are then matched between graphs. This allows for topological differences. A maximum consistent set of superedge matches is found using a dedicated branch-and-bound solver, which is over 100 times faster than a standard linear programming approach. Geometrical and topological consistency of candidate matches is determined in a fast hierarchical manner. We demonstrate the effectiveness of our technique at registering angiography and retinal fundus images, as well as neural image stacks.

Given two CT thorax images from the same patient taken at two different points in time, a detailed follow-up assessment of pleural thickenings and their growth requires a registration of the regarded image regions. While the spatio-temporal matching of thickenings could be achieved by a rigid registration, the direct visual comparison or the combination of thickening segmentations from different points in time require a more precise registration. We present a new method which provides a non-rigid registration of the 3D image data in the region close to the lung surface, where pleural thickenings are located. A B-spline based approach is used to compensate the non-rigid deformations of the lungs. The control-grid for the B-splines is determined using a non-iterative method, which requires matched feature points from the registered image pair. However, current non-rigid registration methods compensate all changes of the lung surface. This is in our case explicitly undesired for changes caused by pleural thickenings. Therefore, our approach takes the thickenings into account by choosing feature points not directly located on the lung surface. The number of feature points is reduced and only strong features are kept for a 3D block matching.

Dynamic Positron Emission Tomography is a powerful tool for quantitative imaging of in vivo biological processes. The long scan durations necessitate motion correction, to maintain the validity of the dynamic measurements, which can be particularly challenging due to the low signal-to-noise ratio (SNR) and spatial resolution, as well as the complex tracer behaviour in the dynamic PET data. In this paper we develop a novel automated expectation-maximisation image registration framework that incorporates temporal tracer kinetic information to correct for inter-frame subject motion during dynamic PET scans. We employ the Zubal human brain phantom to simulate dynamic PET data using SORTEO (a Monte Carlo-based simulator), in order to validate the proposed method for its ability to recover imposed rigid motion. We have conducted a range of simulations using different noise levels, and corrupted the data with a range of rigid motion artefacts. The performance of our motion correction method is compared with pairwise registration using normalised mutual information as a voxel similarity measure (an approach conventionally used to correct for dynamic PET inter-frame motion based solely on intensity information). To quantify registration accuracy, we calculate the target registration error across the images. The results show that our new dynamic image registration method based on tracer kinetics yields better realignment of the simulated datasets, halving the target registration error when compared to the conventional method at small motion levels, as well as yielding smaller residuals in translation and rotation parameters. We also show that our new method is less affected by the low signal in the first few frames, which the conventional method based on normalised mutual information fails to realign.

We have applied image analysis methods in the assessment of human kidney perfusion based on 3D dynamic contrast-enhanced (DCE) MRI data. This approach consists of 3D non-rigid image registration of the kidneys and fuzzy C-mean classification of kidney tissues. The proposed registration method reduced motion artifacts in the dynamic images and improved the analysis of kidney compartments (cortex, medulla, and cavities). The dynamic intensity curves show the successive transition of the contrast agent through kidney compartments. The proposed method for motion correction and kidney compartment classification may be used to improve the validity and usefulness of further model-based pharmacokinetic analysis of kidney function.

Magnetic resonance (MR) images of the tongue have been used in both clinical medicine and scientific research to reveal tongue structure and motion. In order to see different features of the tongue and its relation to the vocal tract it is beneficial to acquire three orthogonal image stacks-e.g., axial, sagittal and coronal volumes. In order to maintain both low noise and high visual detail, each set of images is typically acquired with in-plane resolution that is much better than the through-plane resolution. As a result, any one data set, by itself, is not ideal for automatic volumetric analyses such as segmentation and registration or even for visualization when oblique slices are required. This paper presents a method of super-resolution reconstruction of the tongue that generates an isotropic image volume using the three orthogonal image stacks. The method uses preprocessing steps that include intensity matching and registration and a data combination approach carried out by Markov random field optimization. The performance of the proposed method was demonstrated on five clinical datasets, yielding superior results when compared with conventional reconstruction methods.

We present a new approach for efficient estimation and storage of tissue deformation in dynamic medical image data like 3-D+t computed tomography reconstructions of human heart acquisitions. Tissue deformation between two points in time can be described by means of a displacement vector field indicating for each voxel of a slice, from which position in the previous slice at a fixed position in the third dimension it has moved to this position. Our deformation model represents the motion in a compact manner using a down-sampled potential function of the displacement vector field. This function is obtained by a Gauss-Newton minimization of the estimation error image, i. e., the difference between the current and the deformed previous slice. For lossless or lossy compression of volume slices, the potential function and the error image can afterwards be coded separately. By assuming deformations instead of translational motion, a subsequent coding algorithm using this method will achieve better compression ratios for medical volume data than with conventional block-based motion compensation known from video coding. Due to the smooth prediction without block artifacts, particularly whole-image transforms like wavelet decomposition as well as intra-slice prediction methods can benefit from this approach. We show that with discrete cosine as well as with Karhunen-Lo`eve transform the method can achieve a better energy compaction of the error image than block-based motion compensation while reaching approximately the same prediction error energy.

Mitral regurgitation is a wide spread problem. For successful surgical treatment quantification of the mitral annulus, especially its diameter, is essential. Time resolved 3D transesophageal echocardiography (TEE) is suitable for this task. Yet, manual measurement in four dimensions is extremely time consuming, which confirms the need for automatic quantification methods. The method we propose is capable of automatically detecting the cardiac cycle (systole or diastole) for each time step and measuring the mitral annulus diameter. This is done using total variation noise filtering, the graph cut segmentation algorithm and morphological operators. An evaluation took place using expert measurements on 4D TEE data of 13 patients. The cardiac cycle was detected correctly on 78% of all images and the mitral annulus diameter was measured with an average error of 3.08 mm. Its full automatic processing makes the method easy to use in the clinical workflow and it provides the surgeon with helpful information.

Transient ischemic dilation (TID) of the left ventricle measured by myocardial perfusion Single Photon Emission Computed Tomography (SPECT) and defined as a the ratio of stress myocardial blood volume to rest myocardial blood volume has been shown to be highly specific for detection of severe coronary artery disease. This work investigates automated quantification of TID from cardiac Computed Tomography (CT) perfusion images. To date, TID has not been computed from CT. Previous studies to compute TID have assumed accurate segmentation of the left ventricle and performed subsequent analysis of volume change mainly on static or less often on gated myocardial perfusion SPECT. This, however, may limit the accuracy of TID due to potential errors from segmentation, perfusion defects or volume measurement from both images. In this study, we propose to use registration methods to determine TID from cardiac CT scans where deformation field within the structure of interest is used to measure the local volume change between stress and rest. Promising results have been demonstrated with 7 datasets, showing the potential of this approach as a comparative method for measuring TID.

Manifold learning is increasingly being used to discover the underlying structure of medical image data. Traditional approaches operate on whole images with a single measure of similarity used to compare entire images. In this way, information on the locality of differences is lost and smaller trends may be masked by dominant global differences. In this paper, we propose the use of multiple local manifolds to analyse regions of images without any prior knowledge of which regions are important. Localised manifolds are created by partitioning images into regular subsections with a manifold constructed for each patch. We propose a framework for incorporating information from the neighbours of each patch to calculate a coherent embedding. This generates a simultaneous dimensionality reduction of all patches and results in the creation of embeddings which are spatially-varying. Additionally, a hierarchical method is presented to enable a multi-scale embedding solution. We use this to extract spatially-varying respiratory and cardiac motions from cardiac MRI. Although there is a complex interplay between these motions, we show how they can be separated on a regional basis. We demonstrate the utility of the localised joint embedding over a global embedding of whole images and over embedding individual patches independently.

Diffusion MRI (dMRI) is a unique imaging modality for in vivo delineation of the anatomical structure of white matter in the brain. In particular, high angular resolution diffusion imaging (HARDI) is a specific instance of dMRI which is known to excel in detection of multiple neural fibers within a single voxel. Unfortunately, the angular resolution of HARDI is known to be inversely proportional to SNR, which makes the problem of denoising of HARDI data be of particular practical importance. Since HARDI signals are effectively band-limited, denoising can be accomplished by means of linear filtering. However, the spatial dependency of diffusivity in brain tissue makes it impossible to find a single set of linear filter parameters which is optimal for all types of diffusion signals. Hence, adaptive filtering is required. In this paper, we propose a new type of non-local means (NLM) filtering which possesses the required adaptivity property. As opposed to similar methods in the field, however, the proposed NLM filtering is applied in the spherical domain of spatial orientations. Moreover, the filter uses an original definition of adaptive weights, which are designed to be invariant to both spatial rotations as well as to a particular sampling scheme in use. As well, we provide a detailed description of the proposed filtering procedure, its efficient implementation, as well as experimental results with synthetic data. We demonstrate that our filter has substantially better adaptivity as compared to a number of alternative methods.

The evaluation of analysis methods for diffusion tensor imaging (DTI) remains challenging due to the lack of gold standards and validation frameworks. Significant work remains in developing metrics for comparing fiber bundles generated from streamline tractography. We propose a set of volumetric and tract oriented measures for evaluating tract differences. The different methodsdeveloped for this assessment work are: an overlap measurement, a point cloud distance and a quantification of the diffusion properties at similar locations between fiber bundles. The application of the measures in this paper is a comparison of atlas generated tractography to tractography generated in individual images. For the validation we used a database of 37 subject DTIs, and applied the measurements on five specific fiber bundles: uncinate, cingulum (left and right for both bundles) and genu. Each measurments is interesting for specific use: the overlap measure presents a simple and comprehensive metric but is sensitive to partial voluming and does not give consistent values depending on the bundle geometry. The point cloud distance associated with a quantile interpretation of the distribution gives a good intuition of how close and similar the bundles are. Finally, the functional difference is useful for a comparison of the diffusion properties since it is the focus of many DTI analysis to compare scalar invariants. The comparison demonstrated reasonable similarity of results. The tract difference measures are also applicable to comparison of tractography algorithms, quality control, reproducibility studies, and other validation problems.

Quality and consistency of clinical and research data collected from Magnetic Resonance Imaging (MRI) scanners may become suspect due to a wide variety of common factors including, experimental changes, hardware degradation, hardware replacement, software updates, personnel changes, and observed imaging artifacts. Standard practice limits quality analysis to visual assessment by a researcher/clinician or a quantitative quality control based upon phantoms which may not be timely, cannot account for differing experimental protocol (e.g. gradient timings and strengths), and may not be pertinent to the data or experimental question at hand. This paper presents a parallel processing pipeline developed towards experiment specific automatic quantitative quality control of MRI data using diffusion tensor imaging (DTI) as an experimental test case. The pipeline consists of automatic identification of DTI scans run on the MRI scanner, calculation of DTI contrasts from the data, implementation of modern statistical methods (wild bootstrap and SIMEX) to assess variance and bias in DTI contrasts, and quality assessment via power calculations and normative values. For this pipeline, a DTI specific power calculation analysis is developed as well as the first incorporation of bias estimates in DTI data to improve statistical analysis.

Extracting information about axon diameter distributions in the brain is a challenging task which provides useful information for medical purposes; for example, the ability to characterize and monitor axon diameters would be useful in diagnosing and investigating diseases like amyotrophic lateral sclerosis (ALS)¹ or autism.² Three families of operators are defined by Ozarslan,³ whose action upon an NMR attenuation signal extracts the moments of the pore size distribution of the ensemble under consideration; also a numerical method is proposed to continuously reconstruct a discretely sampled attenuation profile using the eigenfunctions of the simple harmonic oscillator Hamiltonian: the SHORE basis. The work presented here extends Ozarlan's method to other bases that can offer a better description of attenuation signal behaviour; in particular, we propose the use of the radial Spherical Polar Fourier (SPF) basis. Testing is performed to contrast the efficacy of the radial SPF basis and SHORE basis in practical attenuation signal reconstruction. The robustness of the method to additive noise is tested and analysed. We demonstrate that a low-order attenuation signal reconstruction outperforms a higher-order reconstruction in subsequent moment estimation under noisy conditions. We propose the simulated annealing algorithm for basis function scale parameter estimation. Finally, analytic expressions are derived and presented for the action of the operators on the radial SPF basis (obviating the need for numerical integration, thus avoiding a spectrum of possible sources of error).

Global fiber tracking algorithms have recently been proposed which were able to compute results of unprecedented quality. They account for avoiding accumulation errors by a global optimization process at the cost of a high computation time of several hours or even days. In this paper, we introduce a novel global fiber tracking algorithm which, for the first time, globally optimizes the underlying diffusion direction map obtained from DTI or HARDI data, instead of single fiber segments. As a consequence, the number of iterations in the optimization process can drastically be reduced by about three orders of magnitude. Furthermore, in contrast to all previous algorithms, the density of the tracked fibers can be adjusted after the optimization within a few seconds. We evaluated our method for diffusion-weighted images obtained from software phantoms, healthy volunteers, and tumor patients. We show that difficult fiber bundles, e.g., the visual pathways or tracts for different motor functions can be determined and separated in an excellent quality. Furthermore, crossing and kissing bundles are correctly resolved. On current standard hardware, a dense fiber tracking result of a whole brain can be determined in less than half an hour which is a strong improvement compared to previous work.

Despite extensive studies in the past, the problem of segmenting globally optimal multiple surfaces in 3D volumetric images remains challenging in medical imaging. The problem becomes even harder in highly noisy and edge-weak images. In this paper we present a novel and highly efficient graph-theoretical iterative method based on a volumetric graph representation of the 3D image that incorporates curvature and shape prior information. Compared with the graph-based method, applying the shape prior to construct the graph on a specific preferred shape model allows easy incorporation of a wide spectrum of shape prior information. Furthermore, the key insight that computation of the objective function can be done independently in the x and y directions makes local improvement possible. Thus, instead of using global optimization technique such as maximum flow algorithm, the iteration based method is much faster. Additionally, the utilization of the curvature in the objective function ensures the smoothness. To the best of our knowledge, this is the first paper to combine the shape-prior penalties with utilizing curvature in objective function to ensure the smoothness of the generated surfaces while striving for achieving global optimality. To evaluate the performance of our method, we test it on a set of 14 3D OCT images. Comparing to the best existing approaches, our experiments suggest that the proposed method reduces the unsigned surface positioning errors form 5.44 ± 1.07(μm) to 4.52 ± 0.84(μm). Moreover, our method has a much improved running time, yields almost the same global optimality but with much better smoothness, which makes it especially suitable for segmenting highly noisy images. The proposed method is also suitable for parallel implementation on GPUs, which could potentially allow us to segment highly noisy volumetric images in real time.

Medical image segmentation is generally an ill-posed problem that can only be solved by incorporating prior knowledge. The ambiguities arise due to the presence of noise, weak edges, imaging artifacts, inhomogeneous interior and adjacent anatomical structures having similar intensity profile as the target structure. In this paper we propose a novel approach to segment the masseter muscle using the graph-cut incorporating additional 3D shape priors in CT datasets, which is robust to noise; artifacts; and shape deformations. The main contribution of this paper is in translating the 3D shape knowledge into both unary and pairwise potentials of the Markov Random Field (MRF). The segmentation task is casted as a Maximum-A-Posteriori (MAP) estimation of the MRF. Graph-cut is then used to obtain the global minimum which results in the segmentation of the masseter muscle. The method is tested on 21 CT datasets of the masseter muscle, which are noisy with almost all possessing mild to severe imaging artifacts such as high-density artifacts caused by e.g. the very common dental fillings and dental implants. We show that the proposed technique produces clinically acceptable results to the challenging problem of muscle segmentation, and further provide a quantitative and qualitative comparison with other methods. We statistically show that adding additional shape prior into both unary and pairwise potentials can increase the robustness of the proposed method in noisy datasets.

Automatic segmentation of parotid glands in head and neck CT images for IMRT planning has drawn attention in recent years. Although previous approaches have achieved substantial success by reaching high overall volume-wise accuracy, suboptimal segmentations are observed on the interior boundary of the gland where the contrast is poor against the adjacent muscle groups. Herein we propose to use a constrained active shape model with landmark uncertainty to improve the segmentation in this area. Results obtained using this method are compared with results obtained using a regular active shape model through a leave-one-out experiment.

It is well known that hippocampal atrophy is a marker of the onset of Alzheimer's disease (AD) and as a result hippocampal volumetry has been used in a number of studies to provide early diagnosis of AD and predict conversion of mild cognitive impairment patients to AD. However, rates of atrophy are not uniform across the hippocampus making shape analysis a potentially more accurate biomarker. This study studies the hippocampi from 226 healthy controls, 148 AD patients and 330 MCI patients obtained from T1 weighted structural MRI images from the ADNI database. The hippocampi are anatomically segmented using the MAPS multi-atlas segmentation method, and the resulting binary images are then processed with SPHARM software to decompose their shapes as a weighted sum of spherical harmonic basis functions. The resulting parameterizations are then used as feature vectors in Support Vector Machine (SVM) classification. A wrapper based feature selection method was used as this considers the utility of features in discriminating classes in combination, fully exploiting the multivariate nature of the data and optimizing the selected set of features for the type of classifier that is used. The leave-one-out cross validated accuracy obtained on training data is 88.6% for classifying AD vs controls and 74% for classifying MCI-converters vs MCI-stable with very compact feature sets, showing that this is a highly promising method. There is currently a considerable fall in accuracy on unseen data indicating that the feature selection is sensitive to the data used, however feature ensemble methods may overcome this.

In order to fit an unseen surface using statistical shape model (SSM), a correspondence between the unseen surface and the model needs to be established, before the shape parameters can be estimated based on this correspondence. The correspondence and parameter estimation problem can be modeled probabilistically by a Gaussian mixture model (GMM), and solved by expectation-maximization iterative closest points (EM-ICP) algorithm. In this paper, we propose to exploit the linearity of the principal component analysis (PCA) based SSM, and estimate the parameters for the unseen shape surface under the EM-ICP framework. The symmetric data terms are devised to enforce the mutual consistency between the model reconstruction and the shape surface. The a priori shape information encoded in the SSM is also included as regularization. The estimation method is applied to the shape modeling of the hippocampus using a hippocampal SSM.

Real-time fusion of Magnetic Resonance (MR) and Trans Rectal Ultra Sound (TRUS) images aid in the localization of malignant tissues in TRUS guided prostate biopsy. Registration performed on segmented contours of the prostate reduces computational complexity and improves the multimodal registration accuracy. However, accurate and computationally efficient 3D segmentation of the prostate in MR images could be a challenging task due to inter-patient shape and intensity variability of the prostate gland. In this work, we propose to use multiple statistical shape and appearance models to segment the prostate in 2D and a global registration framework to impose shape restriction in 3D. Multiple mean parametric models of the shape and appearance corresponding to the apex, central and base regions of the prostate gland are derived from principal component analysis (PCA) of prior shape and intensity information of the prostate from the training data. The estimated parameters are then modified with the prior knowledge of the optimization space to achieve segmentation in 2D. The 2D segmented slices are then rigidly registered with the average 3D model produced by affine registration of the ground truth of the training datasets to minimize pose variations and impose 3D shape restriction. The proposed method achieves a mean Dice similarity coefficient (DSC) value of 0.88±0.11, and mean Hausdorff distance (HD) of 3.38±2.81 mm when validated with 15 prostate volumes of a public dataset in leave-one-out validation framework. The results achieved are better compared to some of the works in the literature.

Breast density has become an established risk indicator for developing breast cancer. Current clinical practice reflects this by grading mammograms patient-wise as entirely fat, scattered fibroglandular, heterogeneously dense, or extremely dense based on visual perception. Existing (semi-) automated methods work on a per-image basis and mimic clinical practice by calculating an area fraction of fibroglandular tissue (mammographic percent density). We suggest a method that follows clinical practice more strictly by segmenting the fibroglandular tissue portion directly from the joint data of all four available mammographic views (cranio-caudal and medio-lateral oblique, left and right), and by subsequently calculating a consistently patient-based mammographic percent density estimate. In particular, each mammographic view is first processed separately to determine a region of interest (ROI) for segmentation into fibroglandular and adipose tissue. ROI determination includes breast outline detection via edge-based methods, peripheral tissue suppression via geometric breast height modeling, and - for medio-lateral oblique views only - pectoral muscle outline detection based on optimizing a three-parameter analytic curve with respect to local appearance. Intensity harmonization based on separately acquired calibration data is performed with respect to compression height and tube voltage to facilitate joint segmentation of available mammographic views. A Gaussian mixture model (GMM) on the joint histogram data with a posteriori calibration guided plausibility correction is finally employed for tissue separation. The proposed method was tested on patient data from 82 subjects. Results show excellent correlation (r = 0.86) to radiologist's grading with deviations ranging between -28%, (q = 0.025) and +16%, (q = 0.975).

This document presents a novel method for the problem of image segmentation, based on random-walks. This method shares similarities with the Mean-shift algorithm, as it finds the modes of the intensity histogram of images. However, unlike Mean-shift, our proposed method is stochastic and also provides class membership probabilities. Also, unlike other random-walk based methods, our approach does not require any form of user interaction, and can scale to very large images. To illustrate the usefulness, efficiency and scalability of our method, we test it on the task of segmenting anatomical structures present in cardiac CT and brain MRI images.

Fully automated prostate segmentation helps to address several problems in prostate cancer diagnosis and treatment: it can assist in objective evaluation of multiparametric MR imagery, provides a prostate contour for MR-ultrasound (or CT) image fusion for computer-assisted image-guided biopsy or therapy planning, may facilitate reporting and enables direct prostate volume calculation. Among the challenges in automated analysis of MR images of the prostate are the variations of overall image intensities across scanners, the presence of nonuniform multiplicative bias field within scans and differences in acquisition setup. Furthermore, images acquired with the presence of an endorectal coil suffer from localized high-intensity artifacts at the posterior part of the prostate. In this work, a three-dimensional method for fast automated prostate detection based on normalized gradient fields cross-correlation, insensitive to intensity variations and coil-induced artifacts, is presented and evaluated. The components of the method, offline template learning and the localization algorithm, are described in detail. The method was validated on a dataset of 522 T2-weighted MR images acquired at the National Cancer Institute, USA that was split in two halves for development and testing. In addition, second dataset of 29 MR exams from Centre d'Imagerie Médicale Tourville, France were used to test the algorithm. The 95% confidence intervals for the mean Euclidean distance between automatically and manually identified prostate centroids were 4.06 ± 0.33 mm and 3.10 ± 0.43 mm for the first and second test datasets respectively. Moreover, the algorithm provided the centroid within the true prostate volume in 100% of images from both datasets. Obtained results demonstrate high utility of the detection method for a fully automated prostate segmentation.

We present an automatic lung lobe segmentation algorithm for COPD patients. The method enhances fissures, removes unlikely fissure candidates, after which a B-spline is fitted iteratively through the remaining candidate objects. The iterative fitting approach circumvents the need to classify each object as being part of the fissure or being noise, and allows the fissure to be detected in multiple disconnected parts. This property is beneficial for good performance in patient data, containing incomplete and disease-affected fissures. The proposed algorithm is tested on 22 COPD patients, resulting in accurate lobe-based densitometry, and a median overlap of the fissure (defined 3 voxels wide) with an expert ground truth of 0.65, 0.54 and 0.44 for the three main fissures. This compares to complete lobe overlaps of 0.99, 0.98, 0.98, 0.97 and 0.87 for the five main lobes, showing promise for lobe segmentation on data of patients with moderate to severe COPD.

Accurate segmentation of overlapping nuclei is essential in determining nuclei count and evaluating the sub-cellular localization of protein biomarkers in image Cytometry and Histology. Current cellular segmentation algorithms generally lack fast and reliable methods for disambiguating clumped nuclei. In immuno-fluorescence segmentation, solutions to challenges including nuclei misclassification, irregular boundaries, and under-segmentation require reliable separation of clumped nuclei. This paper presents a fast and accurate algorithm for joint segmentation of cellular cytoplasm and nuclei incorporating procedures for reliably separating overlapping nuclei. The algorithm utilizes a combination of ideas and is a significant improvement on state-of-the-art algorithms for this application. First, an adaptive process that includes top-hat filtering, blob detection and distance transforms estimates the inverse illumination field and corrects for intensity non-uniformity. Minimum-error-thresholding based binarization augmented by statistical stability estimation is applied prior to seed-detection constrained by a distance-map-based scale-selection to identify candidate seeds for nuclei segmentation. The nuclei clustering step also incorporates error estimation based on statistical stability. This enables the algorithm to perform localized error correction. Final steps include artifact removal and reclassification of nuclei objects near the cytoplasm boundary as epithelial or stroma. Evaluation using 48 realistic phantom images with known ground-truth shows overall segmentation accuracy exceeding 96%. It significantly outperformed two state-of-the-art algorithms in clumped nuclei separation. Tests on 926 prostate biopsy images (326 patients) show that the segmentation improvement improves the predictive power of nuclei architecture features based on the minimum spanning tree algorithm. The algorithm has been deployed in a large scale pathology application.

Labeling or segmentation of structures of interest in medical imaging plays an essential role in both clinical and scientific understanding. Two of the common techniques to obtain these labels are through either fully automated segmentation or through multi-atlas based segmentation and label fusion. Fully automated techniques often result in highly accurate segmentations but lack the robustness to be viable in many cases. On the other hand, label fusion techniques are often extremely robust, but lack the accuracy of automated algorithms for specific classes of problems. Herein, we propose to perform simultaneous automated segmentation and statistical label fusion through the reformulation of a generative model to include a linkage structure that explicitly estimates the complex global relationships between labels and intensities. These relationships are inferred from the atlas labels and intensities and applied to the target using a non-parametric approach. The novelty of this approach lies in the combination of previously exclusive techniques and attempts to combine the accuracy benefits of automated segmentation with the robustness of a multi-atlas based approach. The accuracy benefits of this simultaneous approach are assessed using a multi-label multi-atlas whole-brain segmentation experiment and the segmentation of the highly variable thyroid on computed tomography images. The results demonstrate that this technique has major benefits for certain types of problems and has the potential to provide a paradigm shift in which the lines between statistical label fusion and automated segmentation are dramatically blurred.

Alzheimer's disease (AD) severely affects the hippocampus: it loses mass and shrinks as the disease advances. Thus delineation of the hippocampus is an important task in the clinical study of AD. Because of its simplicity and good performance, multi-atlas based segmentation has become a popular approach for medical image segmentation. We propose to use manifold learning for atlas selection in the framework of multi-atlas based segmentation. The framework only benefits when selecting atlases similar to the target image. Since manifold learning assigns each image a coordinate in low-dimensional space by respecting the neighborhood relationship, it is well suited for atlas selection. The key contribution is that we use manifold learning based on a metric derived from non-rigid transformation as the resulting embedding better captures deformations or shape differences between images than similarity measures based on voxel intensity. The proposed method is evaluated in a leave-one-out experiment on a set of 110 hippocampus images; we report mean Dice score of 0.9114 (0.0227). The method was validated against a state-of-the-art method for hippocampus segmentation.

For multi atlas-based segmentation approaches, a segmentation fusion scheme which considers local performance measures may be more accurate than a method which uses a global performance measure. We improve upon an existing segmentation fusion method called SIMPLE and extend it to be localized and suitable for multi-labeled segmentations. We demonstrate the algorithm performance on 23 CT scans of COPD patients using a leave-one- out experiment. Our algorithm performs significantly better (p < 0.01) than majority voting, STAPLE, and SIMPLE, with a median overlap of the fissure of 0.45, 0.48, 0.55 and 0.6 for majority voting, STAPLE, SIMPLE, and the proposed algorithm, respectively.

Segmentation plays a critical role in exposing connections between biological structure and function. The process of label fusion collects and combines multiple observations into a single estimate. Statistically driven techniques provide mechanisms to optimally combine segmentations; yet, optimality hinges upon accurate modeling of rater behavior. Traditional approaches, e.g., Majority Vote and Simultaneous Truth and Performance Level Estimation (STAPLE), have been shown to yield excellent performance in some cases, but do not account for spatial dependences of rater performance (i.e., regional task difficulty). Recently, the COnsensus Level, Labeler Accuracy and Truth Estimation (COLLATE) label fusion technique augmented the seminal STAPLE approach to simultaneously estimate regions of relative consensus versus confusion along with rater performance. Herein, we extend the COLLATE framework to account for multiple consensus levels. Toward this end, we posit a generalized model of rater behavior of which Majority Vote, STAPLE, STAPLE Ignoring Consensus Voxels, and COLLATE are special cases. The new algorithm is evaluated with simulations and shown to yield improved performance in cases with complex region difficulties. Multi-COLLATE achieve these results by capturing different consensus levels. The potential impacts and applications of generative model to label fusion problems are discussed.

One of key topics in fMRI analysis is discovery of task-related brain areas. We focus on predictive accuracy as a better relevance measure than traditional univariate voxel activations that miss important multivariate voxel interactions. We use sparse regression (more specifically, the Elastic Net¹) to learn predictive models simultaneously with selection of predictive voxel subsets, and to explore transition from task-relevant to task-irrelevant areas. Exploring the space of sparse solutions reveals a much wider spread of task-relevant information in the brain than it is typically suggested by univariate correlations. This happens for several tasks we considered, and is most noticeable in case of complex tasks such as pain rating; however, for certain simpler tasks, a clear separation between a small subset of relevant voxels and the rest of the brain is observed even with multivariate approach to measuring relevance.

β-amyloid has been shown to play a crucial role in Alzheimer's disease (AD). In vivo β-amyloid imaging using [11C]Pittsburgh compound Β (PiB) positron emission tomography has made it possible to analyze the relationship between β-amyloid deposition and different pathological markers involved in AD. PiB allows us to stratify the population between subjects which are likely to have prodromal AD, and those who don't. The comparison of the cortical thickness in these different groups is important to better understanding and detect the first symptoms of the disease which may lead to an earlier therapeutic care to reduce neurone loss. Several techniques have been developed to compare the cortical volume and/or thickness between AD and HC groups. However due to the noise introduced by the cortical thickness estimation and by the registration, these methods do not allow to unveil any major different when comparing prodromal AD groups with healthy control subjects group. To improve our understanding of where initial Alzheimer neurodegeneration occurs in the cortex we have developed a surface based technique, and have applied it to the discrimination between PIB-positive and PiB-negative HCs. We first identify the regions where AD patients show high cortical atrophy by using an AD/PiB- HC vertex-wise T-test. In each of these discriminating regions, comparison between PiB+ HC, PiB- HC and AD are performed. We found some significant differences between the two HC groups in the hippocampus and in the temporal lobe for both hemisphere and in the precuneus and occipital regions only for the left hemisphere.

Automatic labeling of the gyri and sulci on the cortical surface is important for studying cortical morphology and brain functions within populations. A method to simultaneously label gyral regions and extract sulcal curves is proposed. Assuming that the gyral regions parcellate the whole cortical surface into contiguous regions with certain fixed topology, the proposed method labels the subject cortical surface by deformably registering a network of curves that form the boundary of gyral regions to the subject cortical surface. In the registration process, the curves are encouraged to follow the fine details of the sulcal geometry and to observe the shape statistics learned from training data. Using the framework of probabilistic point set registration methods, the proposed algorithm finds the sulcal curve network that maximizes the posterior probability by Expectation-Maximization (EM). The automatic labeling method was evaluated on 15 cortical surfaces using a leave-one-out strategy. Quantitative error analysis is carried out on both labeled regions and major sulcal curves.

In functional neuroimaging studies, the inter-subject alignment of functional magnetic resonance imaging (fMRI) data is a necessary precursor to improve functional consistency across subjects. Traditional structural MRI based registration methods cannot achieve accurate inter-subject functional consistency in that functional units are not necessarily consistently located relative to anatomical structures due to functional variability across subjects. Although spatial smoothing commonly used in fMRI data preprocessing can reduce the inter-subject functional variability, it may blur the functional signals and thus lose the fine-grained information. In this paper we propose a novel functional signal based fMRI image registration method which aligns local functional connectivity patterns of different subjects to improve the inter-subject functional consistency. Particularly, the functional connectivity is measured using Pearson correlation. For each voxel of an fMRI image, its functional connectivity to every voxel in its local spatial neighborhood, referred to as its local functional connectivity pattern, is characterized by a rotation and shift invariant representation. Based on this representation, the spatial registration of two fMRI images is achieved by minimizing the difference between their corresponding voxels' local functional connectivity patterns using a deformable image registration model. Experiment results based on simulated fMRI data have demonstrated that the proposed method is more robust and reliable than the existing fMRI image registration methods, including maximizing functional correlations and minimizing difference of global connectivity matrices across different subjects. Experiment results based on real resting-state fMRI data have further demonstrated that the proposed fMRI registration method can statistically significantly improve functional consistency across subjects.

Ultra-high field 7T magnetic resonance imaging (MRI) offers potentially unprecedented spatial resolution of functional activity within the human brain through increased signal and contrast to noise ratios over traditional 1.5T and 3T MRI scanners. However, the effects physiological and imaging artifacts are also greatly increased. Traditional statistical parametric mapping theories based on distributional properties representative of data acquired at lower fields may be inadequate for new 7T data. Herein, we investigate the model fitting residuals based on two 7T and one 3T protocols. We find that model residuals are substantively more non-Gaussian at 7T relative to 3T. Imaging slices that passed through regions with peak inhomogeneity problems (e.g., mid-brain acquisitions for the 7T hippocampus) exhibited visually higher degrees of distortion along with spatially correlated and extreme values of kurtosis (a measure of non- Gaussianity). The impacts of artifacts have been previously addressed for 3T data by estimating the covariance matrix of the regression errors. We further extend the robust estimation approach for autoregressive models and evaluate the qualitative impacts of this technique relative to traditional inference. Clear differences in statistical significance are shown between inferences based on classical versus robust assumptions, which suggest that inferences based on Gaussian assumptions are subject to practical (as well as theoretical) concerns regarding their power and validity. Hence, modern statistical approaches, such as the robust autoregressive model posed herein, are appropriate and suitable for inference with ultra-high field functional magnetic resonance imaging.

A common task in the analysis of digitized histological sections is reconstructing a volumetric representation of the original specimen. Image registration algorithms are used in this task to compensate for translational, rotational, scale, shear, and local geometric differences between slices. Various systems have been developed to perform volumetric reconstruction by registering pairs of successive slices according to rigid, similarity, affine, and/or deformable transformations. To provide a coarse initial volumetric reconstruction, rigid transformations may be too constrained, as they do not allow for scale or shear; but, affine transformations may be too flexible, enabling larger scale or shear factors than physically reflected in the histological sections. One difficulty with these systems is caused by the aperture problem; even if successive slices are registered reasonably well, the composition of transformations over tens or hundreds of slices can yield global twisting and scale and shear changes that yield a volumetric reconstruction that is significantly distorted from the shape of the true specimen. The impact of the aperture problem can be reduced by considering more than two successive images in the registration process. Systems that take this approach use global energy functions, elastic spring models, post hoc filtering/smoothing, or solutions to shortest-path problems on graphs. In this article, we propose a volume reconstruction algorithm that handles the aperture problem and yields nearly rigid transformations (i.e., affine transformations with small scale and shear factors). Our algorithm is based on robust geometric alignment of descriptive feature points (for example, using SIFT16) via constrained optimization. We will illustrate our algorithm on the task of volumetric reconstruction from histological sections of a chicken embryo with an embedded tumor spheroid.

In this paper, we propose an image registration algorithm named statistically-based FFD registration (SFFD). This registration method is a modification of a well-known free-form deformations (FFD) approach. Our framework dramatically reduces the number of parameters to optimise and only needs to perform a single-resolution optimisation to account for coarse and fine local displacements, in contrast to the multi-resolution strategy employed by the FFD-based registration. The proposed registration uses statistical deformation models (SDMs) as a priori knowledge to guide the alignment of a new subject to a common reference template. These SDMs account for the anatomical mean and variability across a population of subjects. We also propose that available anatomical landmark information can be encoded within the proposed SDM framework to enforce the alignment of certain anatomical structures. We present results in terms of fiducial localisation error, which illustrate the ability of the SDMs to encode landmark position information. We also show that our statistical registration algorithm can provide registration results comparable to the standard FFD-based approach at a much lower computational cost.

Many rigid and affine registration methods rely on optimizing an intensity-based similarity criterion between images. Once registered, however, it is difficult to assess the quality of the registration based solely on the value of the similarity measure. Past work in quantitative error analysis relies on the availability of fiducial markers. Little work has been done on developing techniques that would permit assessing the registration quality between images that do not contain fiducial markers without manual intervention. In this paper, we present an automatic technique that permits to do so. We apply our method to estimate the registration quality of 10 MR and CT pairs and 10 MR and MR-contrast pairs. We show that our technique is capable of detecting cases with registration error larger than 2° around one axis. We also show that our method is better able to identify error in MR to CT registrations than popular similarity measures. Work is under way to better determine the sensitivity of the technique.

The introduction of spectral Optical Coherence Tomography (OCT) scanners has enabled acquisition of high resolution, 3D cross-sectional volumetric images of the retina. 3D-OCT is used to detect and manage eye diseases such as glaucoma and age-related macular degeneration. To follow-up patients over time, image registration is a vital tool to enable more precise, quantitative comparison of disease states. In this work we present a 3D registrationmethod based on a two-step approach. In the first step we register both scans in the XY domain using an Iterative Closest Point (ICP) based algorithm. This algorithm is applied to vessel segmentations obtained from the projection image of each scan. The distance minimized in the ICP algorithm includes measurements of the vessel orientation and vessel width to allow for a more robust match. In the second step, a graph-based method is applied to find the optimal translation along the depth axis of the individual A-scans in the volume to match both scans. The cost image used to construct the graph is based on the mean squared error (MSE) between matching A-scans in both images at different translations. We have applied this method to the registration of Optic Nerve Head (ONH) centered 3D-OCT scans of the same patient. First, 10 3D-OCT scans of 5 eyes with glaucoma imaged in vivo were registered for a qualitative evaluation of the algorithm performance. Then, 17 OCT data set pairs of 17 eyes with known deformation were used for quantitative assessment of the method's robustness.

We introduce a new approach for spline-based elastic registration using both point landmarks and intensity information. With this approach, both types of information and a regularization based on the Navier equation are directly integrated in a single energy minimizing functional. For this functional, we have derived an analytic solution, which is based on matrix-valued non-radial basis functions. Our approach can cope with monomodal and multimodal images. For the latter case, we have integrated a computationally efficient analytic similarity measure. We have successfully applied our approach to synthetic images, phantom images, and MR images of the human brain.

Range imaging modalities, such as time-of-flight cameras (ToF), are becoming very popular for the acquisition of intra-operative data, which can be used for registering the patient's anatomy with pre-operative data, such as 3D images generated by computed tomographies (CT) or magnetic resonance imaging (MRI). However, due to the distortions that appear because of the different acquisition principles of the input surfaces, the noise, and the deformations that may occur in the intra-operative environment, we face different surface properties for points lying on the same anatomical locations and unreliable feature points detection, which are crucial for most surface matching algorithms. In order to overcome these issues, we present a method for automatically finding correspondences between surfaces that searches for minimally deformed configurations. For this purpose, an error metric that expresses the reliability of a correspondence set based on its spatial configuration is employed. The registration error is minimized by a combinatorial analysis through search-trees. Our method was evaluated with real and simulated ToF and CT data, and showed to be reliable for the registration of partial multi-modal surfaces with noise and distortions.

Contrast-enhanced ultrasound (CEUS) is a valuable imaging modality in the detection and evaluation of different kinds of lesions. Three-dimensional CEUS acquisitions allow quantitative volumetric assessments and better visualization of lesions, but automatic and robust analysis of such images is very challenging because of their poor quality. In this paper, we propose a method to automatically segment lesions such as cysts in 3D CEUS data. First we use a pre-processing step, based on the guided filtering framework, to improve the visibility of the lesions. The lesion detection is then performed through a multi-scale radial symmetry transform. We compute the likelihood of a pixel to be the center of a dark rounded shape. The local maxima of this likelihood are considered as lesions centers. Finally, we recover the whole lesions volume with multiple front propagation based on image intensity, using a fast marching method. For each lesion, the final segmentation is chosen as the one which maximizes the gradient flux through its boundary. Our method has been tested on several clinical 3D CEUS images of the kidney and provides promising results.

Breast ultrasound (BUS) imaging is an imaging modality used for the detection and diagnosis of breast lesions and it has become a crucial modality nowadays specially for providing a complementary view when other modalities (i.e. mammography) are not conclusive. However, lesion detection in ultrasound images is still a challenging problem due to the presence of artifacts such as low contrast, speckle, inhomogeneities and shadowing. In order to deal with these problems and improve diagnosis accuracy, radiologists tend to complement ultrasound imaging with elastography data. Following the prominent relevance of elastography in clinical environments, it is reasonable to assume that lesion segmentation methods could also benefit from this complementary information. This paper proposes a novel breast ultrasound lesion segmentation framework for B-mode images including elastography information. A distortion field is estimated to restore the ideal image while simultaneously identifying regions of similar intensity inhomogeneity using a Markov Random Field (MRF) and a maximum a posteriori (MAP) formulation. Bivariate Gaussian distributions are used to model both B-mode and elastography information. This paper compares the fused B-mode and elastography framework with B-mode or elastography alone using different cases, including illustrative cases, where B-mode shows a well defined lesion and where elastography provides more meaningful information, showing a significant improvement when B-mode images are not conclusive which is often the case in non cystic lesions. Results show that combining both B-mode and elastography information in an unique framework makes the algorithm more robust and image quality independent.

We present a novel continuous Max-Flow based method to segment the inner left ventricular wall from 3D trans-esophageal echocardiography image sequences, which minimizes an energy functional encoding two Fisher-Tippett distributions and a geometrical constraint in form of a Euclidean distance map in a numerically efficient and accurate way. After initialization the method is fully automatic and is able to perform at up to 10Hz making it available for image-guided interventions. Results are shown on 4D TEE data sets from 18 patients with pathological cardiac conditions and the speed of the algorithm is assessed under a variety of conditions.

While efficient graph-theoretic approaches exist for the optimal (with respect to a cost function) and simultaneous segmentation of multiple surfaces within volumetric medical images, the appropriate design of cost functions remains an important challenge. Previously proposed methods have used simple cost functions or optimized a combination of the same, but little has been done to design cost functions using learned features from a training set, in a less biased fashion. Here, we present a method to design cost functions for the simultaneous segmentation of multiple surfaces using the graph-theoretic approach. Classified texture features were used to create probability maps, which were incorporated into the graph-search approach. The efficiency of such an approach was tested on 10 optic nerve head centered optical coherence tomography (OCT) volumes obtained from 10 subjects that presented with glaucoma. The mean unsigned border position error was computed with respect to the average of manual tracings from two independent observers and compared to our previously reported results. A significant improvement was noted in the overall means which reduced from 9.25 ± 4.03μm to 6.73 ± 2.45μm (p < 0.01) and is also comparable with the inter-observer variability of 8.85 ± 3.85μm.

Image segmentation is of paramount importance for quantitative analysis of medical image data. Recently, a 3-D graph search method which can detect globally optimal interacting surfaces with respect to the cost function of volumetric images has been introduced, and its utility demonstrated in several application areas. Although the method provides excellent segmentation accuracy, its limitation is a slow processing speed when many surfaces are simultaneously segmented in large volumetric datasets. Here, we propose a novel method of parallel graph search, which overcomes the limitation and allows the quick detection of multiple surfaces. To demonstrate the obtained performance with respect to segmentation accuracy and processing speedup, the new approach was applied to retinal optical coherence tomography (OCT) image data and compared with the performance of the former non-parallel method. Our parallel graph search methods for single and double surface detection are approximately 267 and 181 times faster than the original graph search approach in 5 macular OCT volumes (200 x 5 x 1024 voxels) acquired from the right eyes of 5 normal subjects. The resulting segmentation differences were small as demonstrated by the mean unsigned differences between the non-parallel and parallel methods of 0.0 ± 0.0 voxels (0.0 ± 0.0 μm) and 0.27 ± 0.34 voxels (0.53 ± 0.66 μm) for the single- and dual-surface approaches, respectively.

This paper proposes a greedy algorithm for automated reconstruction of neural arbors from light microscopy stacks of images. The algorithm is based on the minimum cost path method. While the minimum cost path, obtained using the Fast Marching Method, results in a trace with the least cumulative cost between the start and the end points, it is not sufficient for the reconstruction of neural trees. This is because sections of the minimum cost path can erroneously travel through the image background with undetectable detriment to the cumulative cost. To circumvent this problem we propose an algorithm that grows a neural tree from a specified root by iteratively re-initializing the Fast Marching fronts. The speed image used in the Fast Marching Method is generated by computing the average outward flux of the gradient vector flow field. Each iteration of the algorithm produces a candidate extension by allowing the front to travel a specified distance and then tracking from the farthest point of the front back to the tree. Robust likelihood ratio test is used to evaluate the quality of the candidate extension by comparing voxel intensities along the extension to those in the foreground and the background. The qualified extensions are appended to the current tree, the front is re-initialized, and Fast Marching is continued until the stopping criterion is met. To evaluate the performance of the algorithm we reconstructed 6 stacks of two-photon microscopy images and compared the results to the ground truth reconstructions by using the DIADEM metric. The average comparison score was 0.82 out of 1.0, which is on par with the performance achieved by expert manual tracers.