Search

The accuracy and precision of segmentations of medical images has been difficult to quantify in the absence of a “ground truth” or reference standard segmentation for clinical data. Although physical or digital phantoms can help by providing a reference standard, they do not allow the reproduction of the full range of imaging and anatomical characteristics observed in clinical data. An alternative assessment approach is to compare to segmentations generated by domain experts. Segmentations may be generated by raters who are trained experts or by automated image analysis algorithms. Typically these segmentations differ due to intra-rater and inter-rater variability. The most appropriate way to compare such segmentations has been unclear. We present here a new algorithm to enable the estimation of performance characteristics, and a true labeling, from observations of segmentations of imaging data where segmentation labels may be ordered or continuous measures. This approach may be used with, amongst others, surface, distance transform or level set representations of segmentations, and can be used to assess whether or not a rater consistently over-estimates or under-estimates the position of a boundary.

After defining the medical image registration problem, we provide a short introduction to a select group of multi-modal image alignment approaches. More precisely, we choose four widely-used statistical methods applied in registration scenarios for analysis and comparison. We clarify the implicit and explicit assumptions made by each, aiming to yield a better understanding of their relative strengths and weaknesses. We also introduce a figural representation of the methods in order to provide an intuitive way of illustrating their similarities and differences.

A method is presented for segmentation of anatomical structures that incorporates prior information about the intensity and curvature profile of the structure from a training set of images and boundaries. Specifically the authors model the intensity distribution as a function of signed distance from the object boundary, instead of modeling only the intensity of the object as a whole. A curvature profile acts as a boundary regularization term specific to the shape being extracted, as opposed to simply penalizing high curvature. Using the prior model, the segmentation process estimates a maximum a posteriori higher dimensional surface whose zero level set converges on the boundary of the object to be segmented. Segmentation results are demonstrated on synthetic data and magnetic resonance imagery.

We propose a model-based curve evolution technique for segmentation of images containing known object types. In particular, motivated by the work of Leventon et al. (2000), we derive a parametric model for an implicit representation of the segmenting curve by applying principal component analysis to a collection of signed distance representations of the training data, The parameters of this representation are then calculated to minimize an objective function for segmentation. We found the resulting algorithm to be computationally efficient, able to handle multidimensional data, robust to noise and initial contour placements, while at the same time, avoiding the need for point correspondences during the training phase of the algorithm. We demonstrate this technique by applying it to two medical applications.

The registration of pre-operative volumetric datasets to intra-operative two-dimensional images provides an improved way of verifying patient position and medical instrument location. In applications from orthopedics to neurosurgery, it has great value in maintaining up-to-date information about changes due to intervention. We propose a mutual information-based registration algorithm which establishes the proper alignment via a stochastic gradient ascent strategy. Our main contribution lies in estimating probability density measures of image intensities with a sparse histogramming method which could lead to potential speedup over existing registration procedures and deriving the gradient estimates required by the maximization procedure. Experimental results are presented on fluoroscopy and CT datasets of a real skull, and on a CT-derived dataset of a real skull, a plastic skull and a plastic lumbar spine segment.

In this paper, we propose a 2D-3D vascular registration algorithm based on an efficient rendering method for generating 2D projected images from a segmented 3D MRA volume. At the preprocessing stage, vessels are segmented and represented by a number of spheres with centers on the skeleton points of the vessels and radii equal to the distance from the skeleton points to their closest boundary. To generate projected images for 2D-3D registration, instead of using the conventional ray-casting technique, the spheres are projected to the projection plane. The discrepancy between the projected image and the reference DSA image is measured by the sum of squared differences. Experimental results show that our method is computationally efficient. Moreover, based on manually selected markers, it is shown that the accuracies of our proposed method and the conventional ray-casting based registration method are comparable.

We present a statistical framework that combines the registration of an atlas with the segmentation of magnetic resonance images. We use an Expectation Maximization-based algorithm to find a solution within the model, which simultaneously estimates image inhomogeneities, anatomical labelmap, and a mapping from the atlas to the image space. An example of the approach is given for a brain structure-dependent affine mapping approach. The algorithm produces high quality segmentations for brain tissues as well as their substructures. We demonstrate the approach on a set of 22 magnetic resonance images. In addition, we show that the approach performs better than similar methods which separate the registration and segmentation problems.

In any medical domain, it is common to have more than one test (classifier) to diagnose a disease. In image analysis, for example, there is often more than one reader or more than one algorithm applied to a certain data set. Combining of classifiers is often helpful, but determining the way in which classifiers should be combined is not trivial. Standard strategies are based on learning classifier combination functions from data. We describe a simple strategy to combine results from classifiers that have not been applied to a common data set, and therefore can not undergo this type of joint training. The strategy, which assumes conditional independence of classifiers, is based on the calculation of a combined Receiver Operating Characteristic (ROC) curve, using maximum likelihood analysis to determine a combination rule for each ROC operating point. We offer some insights into the use of ROC analysis in the field of medical imaging.

A hierarchical model based on the Multivariate Autoreges- sive (MAR) process is proposed to jointly model neurological time-series collected from multiple subjects, and to characterize the distribution of MAR coefficients across the population from which those subjects were drawn. Thus, inference about effective connectivity between brain re- gions may be generalized beyond those subjects studied. The posterior on population- and subject-level connectivity parameters are estimated in a Variational Bayesian (VB) framework, and structural model param- eters are chosen by the corresponding evidence criteria. The significance of resulting connectivity statistics are evaluated by permutation-based approximations to the null distribution. The method is demonstrated on simulated data and on actual multi-subject neurological time-series.

We hypothesized that the high-quality images from multidetector-row computed tomography (MDCT) would lead to improved sensitivity and specificity for predicting resectable pancreatic head adenocarcinoma, thus diminishing the value of staging laparoscopy. Forty four consecutive patients underwent thin-section dual-phase MDCT to stage their tumor, followed by an attempted pancreaticoduodenectomy. Four radiologists who were blinded to the operative outcome reviewed the scans and graded the presence of distant and nodal metastases, as well as the degree of arterial and portal involvement. The radiologic criteria for resectability were no distant metastasis, a patent portal vein, and

We present a population registration framework that acts on large collections or populations of data volumes. The data alignment procedure runs in a simultaneous fashion, with every member of the population approaching the central tendency of the collection at the same time. Such a mechanism eliminates the need for selecting a particular reference frame a priori, resulting in a non-biased estimate of a digital atlas. Our algorithm adopts an affine congealing framework with an information theoretic objective function and is optimized via a gradient-based stochastic approximation process embedded in a multi-resolution setting. We present experimental results on both synthetic and real images.

We describe an animated electro-holographic visualization of brain lesions due to the progression of multiple sclerosis. A research case study is used which documents the expression of visible brain lesions in a series of magnetic resonance imaging (MRI) volumes collected over the interval of one year. Some of the salient information resident within this data is described, and the motivation for using a dynamic spatial display to explore its spatial and temporal characteristics is stated. We provide a brief overview of spatial displays in medical imaging applications, and then describe our experimental visualization pipeline, from the processing of MRI datasets, through model construction, computer graphic rendering, and hologram encoding. The utility, strengths and shortcomings of the electro-holographic visualization are described and future improvements are suggested.

This paper extends our prior work on multi-modal image registration based on the a priori knowledge of the joint intensity dis- tribution that we expect to obtain, and Kullback-Leibler distance. This expected joint distribution can be estimated from pre-aligned training images. Experimental results show that, as compared with the Mutual Information and Approximate Maximum Likelihood based registration methods, the new method has longer capture range at different image resolutions, which can lead to a more robust image registration method. Moreover, with a simple interpolation algorithm based on non-grid point random sampling, the proposed method can avoid interpolation artifacts at the low resolution registration. Finally, it is experimentally demon- strated that our method is applicable to a variety of imaging modalities.

Previous work [5] has shown how Ising spatial priors [1] can be incorported into fMRI analysis in a principled manner by using Mutual Information as a statistic for protocol-related activity. The activation image with maximum a posteriori (MAP) probability can then be computed exactly in polynomial time by reduction to a Min-Cut/Max-Flow Problem [4]. In this work, we show that an Ising prior can be applied in the same manner using a standard, linear activation model.

Functional magnetic resonance imaging (fMRI) has significantly contributed to understanding both normal and diseased human brains. Variability often exists in the magnitude, spatial distribution, and statistical significance of the resulting fMRI maps due to differences in equipment and other site-specific differences. In addition, because of costly imaging, demanding tasks, and analytical burden, understanding the effect of these differences may help develop an efficient pooling and comparison mechanism.

The performance of automatic segmentation algorithms often depends critically upon a number of parameters intrinsic to the algorithm. Appropriate setting of these parameters is a pre-requisite for successful segmentation, and yet may be difficult for users to achieve. We propose here a novel algorithm for the automatic selection of optimal parameters for medical image segmentation. Our algorithm makes use of STAPLE (Simultaneous Truth and Performance Level Estimation), a previously described and validated algorithm for automatically identifying a reference standard by which to assess segmentation generators. We execute a set of independent automated segmentation algorithms with initial parameter settings, on a set of images from any clinical application under consideration, estimate a reference standard from the segmentation results using STAPLE, and then identify the parameter settings for each algorithm that maximizes the quality of the segmentation generator result with respect to the reference standard. The process of estimating a reference standard and estimating the optimal parameter settings is iterated to convergence.

We present a new multimodal image registration method based on the a priori knowledge of the class label mappings between two segmented input images. A joint class histogram between the image pairs is estimated by assigning each bin value equal to the total number of occurrences of the corresponding class label pairs. The discrepancy between the observed and expected joint class histograms should be minimized when the transformation is optimal. Kullback-Leibler distance (KLD) is used to measure the difference between these two histograms. Based on the probing experimental results on a synthetic dataset as well as a pair of precisely registered 3D clinical volumes, we show that, with the knowledge of the expected joint class histogram, our method obtained longer capture range and fewer local optimal points as compared with the conventional mutual information (MI) based registration method. We also applied the proposed method to a 2D-3D rigid registration problems between DSA and MRA volumes. Based on manually selected markers, we found that the accuracies of our method and the MI-based method are comparable. Moreover, our method is more computationally efficient than the MI-based method.

Characterizing the performance of image segmentation approaches has been a persistent challenge. Interactive drawing of the desired segmentation by domain experts has often been the only acceptable approach, and yet suffers from intra-expert and inter-expert variability. Algorithms have been sought in order to remove the variability introduced by experts. The accuracy of segmentations of medical images has been difficult to quantify in the absence of a "ground truth" segmentation for clinical data. Although physical or digital phantoms can help, they have so far been unable to reproduce the full range of imaging and anatomical characteristics observed in clinical data. An attractive alternative is comparison to a collection of segmentations by experts, but the most appropriate way to compare segmentations has been unclear. We present here a method for computing a maximum likelihood estimate of the "ground truth" segmentation from a group of expert segmentations, and a simultaneous measure of the quality of each expert. This approach readily enables the assessment of an automated image segmentation algorithm, and direct comparison of expert and algorithm performance.

We propose a shape-based approach to curve evolution for the segmentation of medical images containing known object types. In particular, motivated by the work of Leventon, Grimson, and Faugeras (2000), we derive a parametric model for an implicit representation of the segmenting curve by applying principal component analysis to a collection of signed distance representations of the training data. The parameters of this representation are then manipulated to minimize an objective function for segmentation. The resulting algorithm is able to handle multidimensional data, can deal with topological changes of the curve, is robust to noise and initial contour placements, and is computationally efficient. At the same time, it avoids the need for point correspondences during the training phase of the algorithm. We demonstrate this technique by applying it to two medical applications; two-dimensional segmentation of cardiac magnetic resonance imaging (MRI) and three-dimensional segmentation of prostate MRI.

This paper explores techniques for constructing a 3D computer model of an object from the real world by applying tomographic methods to a sequence of photographic images. While some existing methods can better handle occlusion and concavities, the techniques proposed have the advantageous capability of generating very high-resolution models with attractive speed and simplicity. The application of these methods is presently limited to an appropriate class of mostly convex objects with Lambertian surfaces. The results are volume rendered or surface rendered to produce an interactive display of the object with near life-like realism.

The authors describe an exercise of building a complete robot system, aimed at being as small as possible, but using off-the-shelf components exclusively. The result is an autonomous one (almost) cubic inch robot which incorporates sensing. actuation, onboard computation, and onboard power supplies. Nicknamed Squirt this robot acts as a bug hiding in dark corners and venturing out in the direction of last-heard noises, only moving after the noises are long gone. >

We formulate and interpret several registration methods in the context of a unified statistical and information theoretic framework. A unified interpretation clarifies the implicit assumptions of each method yielding a better understanding of their relative strengths and weaknesses. Additionally, we discuss a generative statistical model from which we derive a novel analysis tool, the auto-information function, as a means of assessing and exploiting the common spatial dependencies inherent in multi-modal imagery. We analytically derive useful properties of the auto-information as well as verify them empirically on multi-modal imagery. Among the useful aspects of the auto-information function is that it can be computed from imaging modalities independently and it allows one to decompose the search space of registration problems.

The introduction of MRI into neurosurgery has opened multiple avenues, but also introduced new challenges.

We propose a novel bias correction method for magnetic resonance (MR) imaging that uses complementary body coil and surface coil images. The former are spatially homogeneous but have low signal intensity; the latter provide excellent signal response but have large bias fields. We present a variational framework where we optimize an energy functional to estimate the bias field and the underlying image using both observed images. The energy functional contains smoothness-enforcing regularization for both the image and the bias field. We present extensions of our basic framework to a variety of imaging protocols. We solve the optimization problem using a computationally efficient numerical algorithm based on coordinate descent, preconditioned conjugate gradient, half-quadratic regularization, and multigrid techniques. We show qualitative and quantitative results demonstrating the effectiveness of the proposed method in producing debiased and denoised MR images.

In order to achieve its main goal of maximal tumor removal while avoiding postoperative neurologic deficits, neuro-oncological surgery is strongly dependent on image guidance. Among all currently available imaging modalities, MRI provides the best anatomic detail and is highly sensitive for intracranial pathology. However, conventional MRI does not detect the exact location of white matter tracts or areas of cortical activation. This essential information can be obtained non-invasively by means of diffusion tensor MRI (DT-MRI) and functional MRI (fMRI) respectively. Here we present our initial experience with fMRI and DT-MRI for surgical planning and guidance in ten brain tumor cases.

This paper presents extensions which improve the performance of the shape-based deformable active contour model presented earlier in [9]. In contrast to that work, the segmentation framework that we present in this paper allows multiple shapes to be segmented simultaneously in a seamless fashion. To achieve this, multiple signed distance functions are employed as the implicit representations of the multiple shape classes within the image. A parametric model for this new representation is derived by applying principal component analysis to the collection of these multiple signed distance functions. By deriving a parametric model in this manner, we obtain a coupling between the multiple shapes within the image and hence effectively capture the co-variations among the different shapes. The parameters of the multi-shape model are then calculated to minimize a single mutual information-based cost functional for image segmentation. The use of a single cost criterion further enhances the coupling between the multiple shapes as the deformation of any given shape depends, at all times, upon every other shape, regardless of their proximity. We demonstrate the utility of this algorithm to the segmentation of the prostate gland, the rectum, and the internal obturator muscles for MR-guided prostate brachytherapy.

In this paper we address the problem of multi-modal co-registration of medical 3D images. Several techniques for the rigid registration of multi-modal images have been developed; in one of those the Kullback-Leibler distance is used to align 2D-3D angiographic images [1]. In this paper we investigate the performance of this technique on the registration of pairs of 3D CT/US images. We study the effects of various image perturbations on the performance of the registration, and obtain promising results.

Establishing image context is the major difficulty of performing laparoscopic ultrasound. The standard techniques used by transabdominal ultrasonographers to understand image orientation are difficult to apply with laparoscopic instruments. In this paper, we describe a navigation system that displays the position and orientation of laparoscopic ultrasound images to the operating surgeon in real time. The display technique we developed for showing the orientation information uses a 3D model of the aorta as the main visual reference. This technique is helpful because it provides surgeons with important spatial cues, which we show improves their ability to interpret the laparoscopic ultrasound.

The authors are developing an automated 3D change detection system which accurately registers medical imagery (e.g., MRI or CT) of the same patient from different times for diagnosing pathologies, monitoring treatment, and tracking tissue changes. The system employs a combination of energy-minimization registration techniques to achieve accurate and robust alignment of 3D data sets. The bases for the registration are 3D surfaces extracted from the 3D imagery. Resultant structural changes in the data are identified by using an adaptive segmentation technique to automatically determine tissue morphology. The novel contributions of this work are its end-to-end automation of the change detection process and its high accuracy in monitoring and highlighting such physiological changes. The authors have applied this system to a multiple sclerosis study in which each patient had been imaged over 20 times for the purpose of tracking lesion evolution. This report describes preliminary registration performance analysis using this data. >

A new information-theoretic approach is presented for finding the pose of an object in an image. The technique does not require information about the surface properties of the object, besides its shape, and is robust with respect to variations of illumination. In our derivation, few assumptions are made about the nature of the imaging process. As a result, the algorithms are quite general and can foreseeably be used in a wide variety of imaging situations. Experiments are presented that demonstrate the approach in registering magnetic resonance images, aligning a complex 3D object model to real scenes including clutter and occlusion, tracking a human head in a video sequence and aligning a view-based 2D object model to real images. The method is based on a formulation of the mutual information between the model and the image. As applied in this paper, the technique is intensity-based, rather than feature-based. It works well in domains where edge or gradient-magnitude based methods have difficulty, yet it is more robust then traditional correlation. Additionally, it has an efficient implementation that is based on stochastic approximation. >

The paper introduces an algorithm which allows the automatic segmentation of multi channel magnetic resonance images. We extended the Expectation Maximization-Mean Field Approximation Segmenter, to include Local Prior Probability Maps. Thereby our algorithm estimates the bias field in the image while simultaneously assigning voxels to different tissue classes under prior probability maps. The probability maps were aligned to the subject using non-rigid registration. This allowed the parcellation of cortical sub-structures including the superior temporal gyrus. To our knowledge this is the first description of an algorithm capable of automatic cortical parcellation incorporating strong noise reduction and image intensity correction.

The validity of segmentation is an important issue in image processing because it has a direct impact on surgical planning. Binary manual segmentation is not only time-consuming but also lacks the ability of differentiating subtle intensity variations among voxels, particularly for those on the border of a tumor and for different tumor types. Previously we have developed an automated segmentation method that yields voxel-wise continuous probabilistic measures, indicating a level of tumor presence. The goal of this work is to examine three accuracy metrics based on two-sample statistical methods, against the estimated composite latent ground truth derived from several experts' manual segmentation by a maximum likelihood algorithm. We estimated the distribution functions of the tumor and control voxel data parametrically by assuming a mixture of two beta distributions with different shape parameters. We derived the resulting receiver operating characteristic curves, Dice similarity coefficients, and mutual information, over all possible decision thresholds. Based on each validation metric, an optimal threshold was then computed via maximization. We illustrated these methods on MR imaging data from nine brain tumor cases, three with meningiomas, three astrocytomas, and three other low-grade gliomas. The automated segmentation yielded satisfactory accuracy, with varied optimal thresholds.

In this paper, we propose a multi-modal image registration method based on the a priori knowledge of the expected joint intensity distribution estimated from aligned training images. The goal of the registration is to find the optimal transformation such that the discrepancy between the expected and the observed joint intensity distributions is minimised. The difference between distributions is measured using the Kullback-Leibler distance (KLD). Experimental results in 3D-3D registration show that the KLD based registration algorithm is less dependent on the size of the sampling region than the Maximum log-Likelihood based registration method. We have also shown that, if manual alignment is unavailable, the expected joint intensity distribution can be estimated based on the segmented and corresponding structures from a pair of novel images. The proposed method has been applied to 2D-3D registration problems between digital subtraction angiograms (DSAs) and magnetic resonance angiographic (MRA) image volumes.

Shape comparisons of two groups of objects often have two goals: to create a classifier to separate the groups and to provide information that shows differences between classes. We examine issues that are important for shape analysis in a study comparing schizophrenic patients to normal subjects. For this study, non-linear classifiers provide large accuracy gains over linear ones. Using volume information directly in the classifier provides gains over a classifier that normalizes the data for volume. We compare two different representations of shape: displacement fields and distance maps. We show that the classifier based on displacement fields outperforms the one based on distance maps. We also show that displacement fields provide more information in visualizing shape differences than distance maps.

Characterizing the performance of image segmentation approaches has been a persistent challenge. Performance analysis is important since segmentation algorithms often have limited accuracy and precision. Interactive drawing of the desired segmentation by domain experts has often been the only acceptable approach, and yet suffers from intra-expert and inter-expert variability. Automated algorithms have been sought in order to remove the variability introduced by experts, but no single methodology for the assessment and validation of such algorithms has yet been widely adopted. The accuracy of segmentations of medical images has been difficult to quantify in the absence of a "ground truth" segmentation for clinical data. Although physical or digital phantoms can help, they have so far been unable to reproduce the full range of imaging and anatomical characteristics observed in clinical data. An attractive alternative is comparison to a collection of segmentations by experts, but the most appropriate way to compare segmentations has been unclear.We present here an Expectation-Maximization algorithm for computing a probabilistic estimate of the "ground truth" segmentation from a group of expert segmentations, and a simultaneous measure of the quality of each expert. This approach readily enables the assessment of an automated image segmentation algorithm, and direct comparison of expert and algorithm performance.

Objective A major shortcoming of image-guided navigational systems is the use of preoperatively acquired image data, which does not account for intraoperative changes in brain morphology. The occurrence of these surgically induced volumetric deformations ("brain shift") has been well established. Maximal measurements for surface and midline shifts have been reported. There has been no detailed analysis, however, of the changes that occur during surgery. The use of intraoperative magnetic resonance imaging provides a unique opportunity to obtain serial image data and characterize the time course of brain deformations during surgery. Methods The vertically open intraoperative magnetic resonance imaging system (SignaSP, 0.5 T; GE Medical Systems, Milwaukee, WI) permits access to the surgical field and allows multiple intraoperative image updates without the need to move the patient. We developed volumetric display software (the 3D Slicer) that allows quantitative analysis of the degree and direction of brain shift. For 25 patients, four or more intraoperative volumetric image acquisitions were extensively evaluated. Results Serial acquisitions allow comprehensive sequential descriptions of the direction and magnitude of intraoperative deformations. Brain shift occurs at various surgical stages and in different regions. Surface shift occurs throughout surgery and is mainly attributable to gravity. Subsurface shift occurs during resection and involves collapse of the resection cavity and intraparenchymal changes that are difficult to model. Conclusion Brain shift is a continuous dynamic process that evolves differently in distinct brain regions. Therefore, only serial imaging or continuous data acquisition can provide consistently accurate image guidance. Furthermore, only serial intraoperative magnetic resonance imaging provides an accurate basis for the computational analysis of brain deformations, which might lead to an understanding and eventual simulation of brain shift for intraoperative guidance.

In previous work [Tsai et al, 1999] we introduced an information theoretic approach for analysis of fMRI time-series data. Subsequently, [Kim et al, 2000] we established a relationship between our information theoretic approach and a simple non-parametric hypothesis test. In this work, we describe an adaptive approach for incorporating the temporal structure that relates the fMRI time-series to both the current and past values of the experimental protocol. This is achieved via an extension of our previous approach using the information-theoretic concept of entropy rate. It can be shown that, despite a differing implementation, our prior method is a special case of the new approach. The entropy rate of a random process quantifies future uncertainty conditioned on the past and side-information (e.g. the experimental protocol, confounding signals, etc.) without making strong assumptions about the nature of that uncertainty (e.g. Gaussianity). Furthermore, we allow the form of the dependency to vary from voxel to voxel in an adaptive fashion. The combination of the information theoretic principles and adaptive estimation of the temporal dependency allows for a more powerful and flexible approach to fMRI analysis. Empirical results are presented on three fMRI datasets measuring motor, auditory, and visual cortex activation comparing the new approach to the previous one as well as a variation on the general linear model. Particular attention is paid to the differences in the type of phenomenology detected by the respective approaches.

A three-dimensional optical flow method to measure volumetric brain deformation from sequential intraoperative MR images and preliminary clinical results from five cases are reported. Intraoperative MR images were scanned before and after dura opening, twice during tumor resection, and immediately after dura closure. The maximum cortical surface shift measured was 11 mm and subsurface shift was 4 mm. The computed deformation field was most satisfactory when the skin was segmented and removed from the images before the optical flow computation.

This chapter focuses on the surgical planning which should include the generation of a 3D volumetric model, target definition, optimization of access routes and the circumvention of functionally important anatomic structures. Surgical planning is performed in a highly computerized interdisciplinary research center where radiologists, surgeons, and computer scientists combine forces to develop and implement various surgical planning methods. Surgical planning should indicate the likely changes in the anatomy during surgery which is only perceptible through simulations actualized from intraoperative data. Proper surgical approaches and repeated rehearsals can be arranged prior to the patient even entering the operating room. This allows complex and little-known surgeries to be carried out not only with more safety but also with the added benefit of training residents and fellows. The preoperative planning procedure involves an image-processing pipeline that filters the image data, classifies or segments the various tissue classes, and renders the image data in 3D. This pipehne eventually results in a 3D model of the anatomy that encompasses the entire operational volume. Three-dimensional modeling provides multiple advantages to both surgeons and radiologists. The most prevalent advantage is that of better visualization, particularly for those who are not accustomed to mental reconstruction of cross-sectional image slices.

The spatial relationship between the corticospinal tracts and a brain tumor is important for planning the surgical strategy. Although the white matter tracts can be manually outlined from structural MRI this is time consuming and impractical. To enhance the spatial understanding of the distorted pathology we have established a method to retrieve this structural information automatically by registration of a standardized normal brain atlas to the individual pathologic anatomy of brain tumor patients. The skin and the brain were segmented in the patient MRI volume. Subsequently a deformable volumetric atlas of a single normal subject was registered to the patient brain using affine and non-linear registration techniques. The estimated spatial correspondence between atlas and patient brain was used to warp the corticospinal tracts from the atlas onto the patient. The accuracy of the method was evaluated in 5 patients with extrinsic tumors of different histopathology and location by comparing selected anatomical landmark structures from the projected atlas to their manually segmented counterpart in the patient. Our method enables the visualization of complex anatomical information with minimal user interaction for routine surgical planning that would otherwise demand the acquisition of additional imaging modalities.

In previous work a novel information-theoretic approach was introduced for calculating the activation map for fMRI analysis [Tsai et al, 1999]. In that work the use of mutual information as a measure of activation resulted in a nonparametric calculation of the activation map. Nonparametric approaches are attractive as the implicit assumptions are milder than the strong assumptions of popular approaches based on the general linear model popularized by Friston et al [1994]. Here we show that, in addition to the intuitive information-theoretic appeal, such an application of mutual information is equivalent to a hypothesis test when the underlying densities are unknown. Furthermore we incorporate local spatial priors using the well-known Ising model thereby dropping the implicit assumption that neighboring voxel time-series are independent. As a consequence of the hypothesis testing equivalence, calculation of the activation map with local spatial priors can be formulated as mincut/maxflow graph-cutting problem. Such problems can be solved in polynomial time by the Ford and Fulkerson method. Empirical results are presented on three fMRI datasets measuring motor, auditory, and visual cortex activation. Comparisons are made illustrating the differences between the proposed technique and one based on the general linear model.

The combined effects of laser speckle and local oscillator shot noise degrade coherent laser radar range measurements. As a result, laser radar range imagery suffers from both uniformly-distributed range anomalies and Gaussian-distributed local range errors. Our goal in this research is to develop a target recognition system capable of recognizing military vehicles in range images provided by airborne laser radars. In particular, we will focus on using laser radar range imagery in a statistical model-based object recognition system. We shall present performance results for our object recognition system using laser radar data from the MIT Lincoln Laboratory Infrared Airborne Radar (IRAR) data release together with 3-D CAD models which account for the possible military targets that may be present on the site imaged by the laser radar.