Publications

Partial volume effects are often experienced in diffusion-weighted MRI of biologic tissue. This is when the signal attenuation reflects a mixture of diffusion processes, originating from different tissue compartments, residing in the same voxel. Decomposing the mixture requires elaborated models that account for multiple compartments, yet the fitting problem for those models is usually ill posed. We suggest a novel approach for stabilizing the fitting problem of the multiple-tensors model by a variational framework that adds biologically oriented assumption of neighborhood alignments. The framework is designed to address fiber ambiguity caused by a number of neuronal fiber compartments residing in the same voxel. The method requires diffusion data acquired by common, clinically feasible MRI sequences, and is able to derive familiar tensor quantities for each compartment. Neighborhood alignment is performed by adding piece-wise smooth regularization constraints to an energy function. Minimization with the gradient descent method produces a set of diffusion-reaction partial differential equations that describe a tensor-preserving flow towards a best approximation of the data while maintaining the constraints. We analyze fiber compartment separation capabilities on a synthetic model of crossing fibers and on brain areas known to have crossing fibers. We compare the results with diffusion tensor imaging analysis and discuss applications for the framework.

In MRI, macroscopic boundaries lead to a diffusion-related increase in signal intensity near them—an effect commonly referred to as edge-enhancement. In diffusion-weighted imaging protocols where the signal attenuation due to diffusion results predominantly from the application of magnetic field gradients, edge-enhancement will depend on the orientation of these diffusion gradients. The resulting diffusion anisotropy can be exploited to map the direction normal to the macroscopic boundary. Simulations suggest that the hypothesized anisotropy may be within observable limits even when the voxel contains no boundary itself—hence, the name remote-anisotropy. Moreover, for certain experimental parameters there may be significant phase cancellations within the voxel that may lead to an edge detraction effect. When this is avoided, the eigenvector corresponding to the smallest eigenvalue of the diffusion tensor obtained from diffusion-tensor imaging can be used to create surface-normal maps conveniently. Experiments performed on simple geometric constructs as well as real tissue demonstrate the feasibility of using the edge-enhancement mechanism to map orientations orthogonal to macroscopic surfaces, which may be used to assess the integrity of tissue and organ boundaries noninvasively.

Segmentation involves separating an object from the background in a given image. The use of image information alone often leads to poor segmentation results due to the presence of noise, clutter or occlusion. The introduction of shape priors in the geometric active contour (GAC) framework has proved to be an effective way to ameliorate some of these problems. In this work, we propose a novel segmentation method combining image information with prior shape knowledge, using level-sets. Following the work of Leventon et al., we propose to revisit the use of PCA to introduce prior knowledge about shapes in a more robust manner. We utilize kernel PCA (KPCA) and show that this method outperforms linear PCA by allowing only those shapes that are close enough to the training data. In our segmentation framework, shape knowledge and image information are encoded into two energy functionals entirely described in terms of shapes. This consistent description permits to fully take advantage of the Kernel PCA methodology and leads to promising segmentation results. In particular, our shape-driven segmentation technique allows for the simultaneous encoding of multiple types of shapes, and offers a convincing level of robustness with respect to noise, occlusions, or smearing.

This paper proposes a novel method to apply the standard graph cut technique to segmenting multimodal tensor valued images. The Riemannian nature of the tensor space is explicitly taken into account by first mapping the data to a Euclidean space where non-parametric kernel density estimates of the regional distributions may be calculated from user initialized regions. These distributions are then used as regional priors in calculating graph edge weights. Hence this approach utilizes the true variation of the tensor data by respecting its Riemannian structure in calculating distances when forming probability distributions. Further, the non-parametric model generalizes to arbitrary tensor distribution unlike the Gaussian assumption made in previous works. Casting the segmentation problem in a graph cut framework yields a segmentation robust with respect to initialization on the data tested.

Richly labeled images representing several sub-structures of an organ occur quite frequently in medical images. For example, a typical brain image can be labeled into grey matter, white matter or cerebrospinal fluid, each of which may be subdivided further. Many manipulations such as interpolation, transformation, smoothing, or registration need to be performed on these images before they can be used in further analysis. In this work, we present a novel multi-shape representation and compare it with the existing representations to demonstrate certain advantages of using the proposed scheme. Specifically, we propose label space, a representation that is both flexible and well suited for coupled multi-shape analysis. Under this framework, object labels are mapped to vertices of a regular simplex, e.g. the unit interval for two labels, a triangle for three labels, a tetrahedron for four labels, etc. This forms the basis of a convex linear structure with the property that all labels are equally spaced. We will demonstrate that this representation has several desirable properties: algebraic operations may be performed directly, label uncertainty is expressed equivalently as a weighted mixture of labels or in a probabilistic manner, and interpolation is unbiased toward any label or the background. In order to demonstrate these properties, we compare label space to signed distance maps as well as other implicit representations in tasks such as smoothing, interpolation, registration, and principal component analysis.

RATIONALE AND OBJECTIVES: This article presents an initiative for the translation of advances in neuroimage analysis techniques to clinical research scientists. Our objective is to bridge the gap between scientific advances made by the biomedical imaging community and their widespread use in the clinical research community. Through national collaborative effort supported by the National Institutes of Health Roadmap, the integration of the most sophisticated algorithms into usable working open-source systems enables clinical researchers to have access to a broad spectrum of cutting edge analysis techniques. A critical step to maximize the long-term positive impact of this collaborative effort is to translate these techniques into new skills of clinical researchers. To address this challenge, we developed a methodology based on three criteria: a multidisciplinary approach, a balance between theory and common practice, and an immersive collaborative environment. The article illustrates our initiative through the exemplar case of diffusion tensor imaging tractography, and reports on our experience over the past two years of designing and delivering training workshops to more than 300 clinicians and scientists using the developed methodology.

Diffusion tensor imaging (DTI) has become one of the most popular MRI techniques in brain research, as well as in clinical practice. The number of brain studies with DTI is growing steadily and, over the last decade, has produced more than 700 publications. Diffusion tensor imaging enables visualization and characterization of white matter fascicli in two and three dimensions. Since the introduction of this methodology in 1994, it has been used to study the white matter architecture and integrity of the normal and diseased brains (multiple sclerosis, stroke, aging, dementia, schizophrenia, etc.). Although it provided image contrast that was not available with routine MR techniques, unique information on white matter and 3D visualization of neuronal pathways, many questions were raised regarding the origin of the DTI signal. Diffusion tensor imaging is constantly validated, challenged, and developed in terms of acquisition scheme, image processing, analysis, and interpretation. While DTI offers a powerful tool to study and visualize white matter, it suffers from inherent artifacts and limitations. The partial volume effect and the inability of the model to cope with non-Gaussian diffusion are its two main drawbacks. Nevertheless, when combined with functional brain mapping, DTI provides an efficient tool for comprehensive, noninvasive, functional anatomy mapping of the human brain. This review summarizes the development of DTI in the last decade with respect to the specificity and utility of the technique in radiology and anatomy studies.

MR-guided focused ultrasound (MRgFUS) is a non-invasive method by which tissue is ablated using ultrasound energy focused on a point. The procedure has proven effective for stationary targets (e.g. uterine fibroids) but has not yet been used for liver lesion treatment due to organ motion. We describe a method to compensate for organ motion to enable continuous application of ultrasound energy in the presence of target movement in the liver. The method involves tracking several salient features (typically blood vessels) in the vicinity of the target location. The location of the target point(s) themselves are updated using a thin plate spline (TPS) interpolation scheme. We demonstrate sub-pixel tracking accuracy on synthetic sequences and additionally show results on MRI sequences acquired on human subjects. Per-feature tracking times were measured to be 5.7ms with a standard deviation of 1.6ms, sufficient for real-time use.

The level set method is a popular technique used in medical image segmentation; however, the numerics involved make its use cumbersome. This paper proposes an approximate level set scheme that removes much of the computational burden while maintaining accuracy. Abandoning a floating point representation for the signed distance function, we use integral values to represent the signed distance function. For the cases of 2D and 3D, we detail rules governing the evolution and maintenance of these three regions. Arbitrary energies can be implemented in the framework. This scheme has several desirable properties: computations are only performed along the zero level set; the approximate distance function requires only a few simple integer comparisons for maintenance; smoothness regularization involves only a few integer calculations and may be handled apart from the energy itself; the zero level set is represented exactly removing the need for interpolation off the interface; and evolutions proceed on the order of milliseconds per iteration on conventional uniprocessor workstations. To highlight its accuracy, flexibility and speed, we demonstrate the technique on intensity-based segmentations under various statistical metrics. Results for 3D imagery show the technique is fast even for image volumes.

Transcutaneous ultrasound imaging may be used to detect abdominal hemorrhage in the field setting. The Focused Assessment with Sonography for Trauma (FAST) examination was developed to characterize blunt abdominal trauma and has been shown to be effective for assessing penetrating trauma as well. However, it is unlikely that a minimally trained operator could perform a diagnostic examination. In our system, the operator is be supported by real-time 3D volume displays. The operator will be directed through the examination by prompts from a computer system or outside expert, potentially with knowledge of the anatomy of the injured patient. The key elements of the tele-operated FAST exam capability have been demonstrated; the exam is performed with real-time guidance from anatomic images registered to the body. It appears likely that Image Registration will assist hemorrhage detection at the point of injury or in the initial evaluation by a trauma response team.