A grey parrot, Alex, comprehends and uses English labels to label objects, colors, shapes, and materials. He combines labels to identify proficiently, request, and refuse > 100 different objects. He categorizes objects with respect to color, shape or material, understands concepts of same/different, bigger/smaller, absence of information, and uses the phrases 'come here', 'I want X' and 'Wanna go Y' where X and Y are object or location labels. He distinguishes quantities to 6, including collections of novel objects, heterogeneous sets, sets involving random arrays; he labels the number of items uniquely defined by the combination of one color and one object category. Given a 7-member collection, he can provide information about the specific instance of one category of an item uniquely defined by the conjunction of two other categories, e.g., 'What object is color-A and shape-B' These results show that Alex, unlike nonhuman primates, both produces and comprehends phonological distinctions. Simple labeling has been replicated with additional subjects. The problem of mutual interest, therefore, is determining the mechanisms that a nonhuman, nonprimate, nonmammal uses to make these distinctions. Imaging systems have unlocked the secrets of the human vocal tract; we now need to examine nonhumans.

Little is known about mechanisms of speech production in parrots. Recently, however, techniques for correlating vocal tract shape with vowel production in humans have become more sophisticated and we have adapted these techniques for use with parrots. We scanned two grey parrot heads with intact vocal tracts. One specimen, 'Oldbird' was fixed with its beak propped open; the second 'Youngbird' was fixed with its beak closed. Using VIDA software, we (1) established that differences in tongue and larynx positioning resulted from opening or closing the beak; and (2) obtained lengths and area functions for the trachea, glottis, pharynx, mouth, and choana for both specimens and esophageal length and area functions for the first specimen. We entered lengths and area functions into a 1D wave propagation model to determine the natural formant frequencies associated with an open versus closed beak. We also determined how manipulating lengths and area functions could affect formant frequency and relative intensity. Finally, by comparing observed grey parrot vowel formant, we predict how the parrot uses its vocal tract to produce speech.

A collection of 3D vocal tract shapes corresponding to vowels and consonants of American English have been acquired for a 27 year old adult female subject using a magnetic resonance imaging. Each 3D shape was condensed into a set of cross-sectional areas of oblique sections perpendicular to the centerline of the vocal tract's long axis. Such a collection of areas is typically called an 'area function'. This set of images and subsequent area functions for the female subject compliments a previous similar study concerning an adult male subject. It is the purpose of this paper to explore the morphological differences between the male and female subjects for three 'cardinal' vowels. Comparisons have been made of the 3D vocal tract shapes, area functions, and acoustic characteristics of the three vowels. The primary difference between genders is that the female pharynx is approximately 37 percent shorter than the male. Limited acoustic modeling has suggested that this shortened pharynx may play a significant role in defining male versus female voice quality.

In the past we have proposed a system that measures the deformations of the vocal folds from videostroboscopic images of the larynx, in that system: (1) we extract the boundaries of the vocal folds, (2) we register elastically the vocal fold boundaries in successive frames. This yields the displacement vector field (DVF) between adjacent frames, and (3) we fit using a least-squares approach an affine transformation model to succinctly describe the deformations between adjacent frames. In this paper, we present as an example of the capabilities of this system, an initial study of the deformation changes in rat vocal folds pre and post injection with Botulinum toxin. For this application the generated DVF was segmented into right DVF and left DVF and the deformation of each segment is studied separately.

We have developed a new and powerful method to study the movement and function of upper airway muscles. Our method is to use direct electrical stimulation of individual upper airway muscles, while performing state of the art high resolution magnetic resonance imaging (MRI). We have adapted a paralyzed isolated UA cat model so that positive or negative static pressure in the UA can be controlled at specific levels while electrical muscle stimulation is applied during MRI. With these techniques we can assess the effect of muscle stimulation on airway cross-sectional area compliance and soft tissue motion. We are reporting the preliminary results and MRI techniques which have enabled us to examine changes in airway dimensions which result form electrical stimulation of specific upper airway dilator muscles. The results of this study will be relevant to the development of new clinical treatments for obstructive sleep apnea by providing new information as to exactly how upper airway muscles function to dilate the upper airway and the strength of stimulation required to prevent the airway obstruction when overall muscle tone may not be sufficient to maintain regular breathing.

Bronchiolar obstruction is commonly manifested in computed tomographic (CT) images as areas of decreased attenuation relative to adjacent normal lung parenchyma. The certain identification of such areas is difficult in practice, particularly is such areas are poorly marginated. This paper presents a novel approach to the enhancement of feature differences between normal and diseased lung parenchyma so that reliable visual assessment can be made. The method relies on a hybrid structural filtering technique which removes pulmonary vessels appearing in the CT cross- sectional images without affecting intrinsic subtle intensity details of the lung parenchyma. In order to restore possible structural distortions introduced by the hybrid filter, a feature localization process based on wavelet reconstruction of feature extrema is used. After contrast enhancement the resultant images are used to delineate region borders of the diseased areas and quantification is made with regard to the extent of the disease. Clinically, the proposed technique is especially valuable for presymptomatic cases in which direct visual assessment of the unprocessed images by different observers often yields inconsistent results.

High-resolution x-ray CT (HRCT) provides detailed images of the lungs and bronchial tree. HRCT-based imaging and quantitation of peripheral bronchial airway geometry provides a valuable tool for assessing regional airway physiology. Such measurements have been sued to address physiological questions related to the mechanics of airway collapse in sleep apnea, the measurement of airway response to broncho-constriction agents, and to evaluate and track the progression of disease affecting the airways, such as asthma and cystic fibrosis. Significant attention has been paid to the measurements of extra- and intra-thoracic airways in 2D sections from volumetric x-ray CT. A variety of manual and semi-automatic techniques have been proposed for airway geometry measurement, including the use of standardized display window and level settings for caliper measurements, methods based on manual or semi-automatic border tracing, and more objective, quantitative approaches such as the use of the 'half-max' criteria. A recently proposed measurements technique uses a model-based deconvolution to estimate the location of the inner and outer airway walls. Validation using a plexiglass phantom indicates that the model-based method is more accurate than the half-max approach for thin-walled structures. In vivo validation of these airway measurement techniques is difficult because of the problems in identifying a reliable measurement 'gold standard.' In this paper we report on ex vivo validation of the half-max and model-based methods using an excised pig lung. The lung is sliced into thin sections of tissue and scanned using an electron beam CT scanner. Airways of interest are measured from the CT images, and also measured with using a microscope and micrometer to obtain a measurement gold standard. The result show no significant difference between the model-based measurements and the gold standard; while the half-max estimates exhibited a measurement bias and were significantly different than the gold standard.

The aim of the study 'dynamic chest image analysis' is to develop computing analysis and visualization methods for showing focal and general abnormalities of lung ventilation and perfusion based on a sequence of digital chest fluoroscopy frames collected at different phases of the respiratory/cardiac cycles. A multiresolutional method for ventilation study with an explicit ventilation model based on pyramid images is proposed in this paper. The ventilation model is sophisticated enough in coverage of both inhalation and exhalation phases, but also remains simple enough in model realization. This model plays a critical role in extracting accurate, geographic ventilation parameters; while the pyramid helps in understanding ventilation at multiple resolutions and speeding up the convergence process in optimization. A number of patients have been studied with a research prototype produced in MATLAB. The prototype has proven to be useful aid in dynamic pulmonary ventilation study. However, for clinical use, further work must be done in the future.

Measurements of the in vivo bronchial tree can be used to assess regional airway physiology. High-resolution CT (HRCT) provides detailed images of the lungs and has been used to evaluate bronchial airway geometry. Such measurements have been sued to assess diseases affecting the airways, such as asthma and cystic fibrosis, to measure airway response to external stimuli, and to evaluate the mechanics of airway collapse in sleep apnea. To routinely use CT imaging in a clinical setting to evaluate the in vivo airway tree, there is a need for an objective, automatic technique for identifying the airway tree in the CT images and measuring airway geometry parameters. Manual or semi-automatic segmentation and measurement of the airway tree from a 3D data set may require several man-hours of work, and the manual approaches suffer from inter-observer and intra- observer variabilities. This paper describes a method for automatic airway tree analysis that combines accurate airway wall location estimation with a technique for optimal airway border smoothing. A fuzzy logic, rule-based system is used to identify the branches of the 3D airway tree in thin-slice HRCT images. Raycasting is combined with a model-based parameter estimation technique to identify the approximate inner and outer airway wall borders in 2D cross-sections through the image data set. Finally, a 2D graph search is used to optimize the estimated airway wall locations and obtain accurate airway borders. We demonstrate this technique using CT images of a plexiglass tube phantom.

Presently available anatomic atlases provide useful coordinate systems such as the ubiquitous Talairach system but are sorely lacking in both spatial resolution and completeness. An appropriately sampled anatomic specimen can provide the additional detail necessary to accurately localize activation sites as well as provide other structural perspectives such as chemoarchitecture. We collected serial section postmortem anatomic data from several whole human head and brain specimens using a cryosectioning technique. Tissue imaged so that voxel resolution was 200 microns or better at full color. These high resolution datasets along with collections of MR data were placed within a common coordinate system and used to produce a probabilistic representation. This approach represents anatomy within a coordinate system as a probability. Coordinate locations are assigned a confidence limit to describe the likelihood that a given location belongs to an anatomic structure based upon the population of specimens. A variety of warping strategies are discussed to provide statistics on morphometric variability and probability. High dimensional anatomically based warps utilizing sulcal anatomy are described. These data are an important and necessary part of the comprehensive structural and functional analyses that focus on the mapping of the human brain.

The detection of sites of brain activation in functional MRI has been a topic of immense research interest and many technique shave been proposed to this end. Recently, principal component analysis (PCA) has been applied to extract the activated regions and their time course of activation. This method is based on the assumption that the activation is orthogonal to other signal variations such as brain motion, physiological oscillations and other uncorrelated noises. A distinct advantage of this method is that it does not require any knowledge of the time course of the true stimulus paradigm. This technique is well suited to EPI image sequences where the sampling rate is high enough to capture the effects of physiological oscillations. In this work, we propose and apply tow methods that are based on PCA to conventional gradient-echo images and investigate their usefulness as tools to extract reliable information on brain activation. The first method is a conventional technique where a single image sequence with alternating on and off stages is subject to a principal component analysis. The second method is a PCA-based approach called the common spatial factor analysis technique (CSF). As the name suggests, this method relies on common spatial factors between the above fMRI image sequence and a background fMRI. We have applied these methods to identify active brain ares during visual stimulation and motor tasks. The results from these methods are compared to those obtained by using the standard cross-correlation technique. We found good agreement in the areas identified as active across all three techniques. The results suggest that PCA and CSF methods have good potential in detecting the true stimulus correlated changes in the presence of other interfering signals.

For several diseases in the head and neck area different imaging modalities are applied to the same patient.Each of these image data sets has its specific advantages and disadvantages. The combination of different methods allows to make the best use of the advantageous properties of each method while minimizing the impact of its negative aspects. Soft tissue alterations can be judged better in an MRI image while it may be unrecognizable in the relating CT. Bone tissue, on the other hand, is optimally imaged in CT. Inflammatory nuclei of the bone can be detected best by their increased signal in SPECT. Only the combination of all modalities let the physical come to an exact statement on pathological processes that involve multiple tissue structures. Several surfaces and voxel based matching functions we have tested allowed a precise merging by means of numerical optimization methods like e.g. simulated annealing without the complicated assertion of fiducial markers or the localization landmarks in 2D cross sectional slice images. The quality of the registration depends on the choice of the optimization procedure according to the complexity of the matching function landscape. Precise correlation of the multimodal head and neck area images together with its 2D and 3D presentation techniques provides a valuable tool for physicians.

Quantitative magnetic resonance (MR) imaging of the brain is useful in multiple sclerosis (MS) in order to obtain reliable indices of disease progression. The goal of this project was to estimate the total volume of gliotic and non gliotic plaques in chronic progressive multiple sclerosis with the help of a semiautomatic segmentation method developed at the Ragnar Granit Institute. Youth developed program running on a PC based computer provides de displays of the segmented data, in addition to the volumetric analyses. The volumetric accuracy of the program was demonstrated by segmenting MR images of fluid filed syringes. An anatomical atlas is to be incorporated in the segmentation system to estimate the distribution of MS plaques in various neural pathways of the brain. A total package including MS plaque volume estimation, estimation of brain atrophy and ventricular enlargement, distribution of MS plaques in different neural segments of the brain has ben planned for the near future. Our study confirmed that total lesion volumes in chronic MS disease show a poor correlation to EDSS scores but show a positive correlation to neuropsychological scores. Therefore accurate total volume measurements of MS plaques using the developed semiautomatic segmentation technique helped us to evaluate the degree of neuropsychological impairment.

We have developed image-restoration techniques applicable to dynamic positron emission tomography that improve the visual quality and quantitative accuracy of neuroreceptor images. Starting wit data from a study of dopamine D-2 receptors in rhesus monkey striata using selective radioligands such as fallypride, we performed a novel effective 3D smoothing of the dynamic sinogram at a much lower computational cost than a truly 3D, adaptive smoothing. The processed sinogram was then input to a standard filtered back-projection algorithm and the resulting images were sharper and less noisy than images reconstructed from the unprocessed sinogram. Simulations were performed and the radioligand binding curves extracted from the restored images were found to be smoother and more accurate than those extracted form the unprocessed reconstructions. Comparison was also made to reconstructions from sinograms processed by the principal component analysis/projection onto convex sets algorithm.

Electroencephalographic imaging is the estimation of 3D neuronal current sources on the cortical surface from the measured electroencephalogram (EEG). It is a highly under- determined inverse problem as there are many 'feasible' images which are consistent with the scalp potentials. Previous approaches to this problem have primarily concentrated on the weighted minimum norm inverse methods. While these methods ensure a unique solution, they often produce overly smoothed solutions and are sensitive to noise in the data. Our group previously proposed a maximum entropy approach to obtain better solutions to this problem. We incorporated a noise rejection term into the maximum entropy method, thereby making it analogous to a Bayesian maximum a posteriori formulation. Additional information from other modalities, like functional magnetic resonance imaging, could be incorporated into this method in the form of a prior bias function to improve solutions. While this approach gave better results than the minimum norm methods, the solutions were still somewhat smooth and blurry. In this work, we developed and tested an iterative version of the maximum entropy method to obtain more localized solutions. This method starts with a distributed estimate computed by the maximum entropy method. It then recursively performs maximum entropy estimations producing a progressively more focal current distribution. We present the method and test its validity through computer simulations for both noiseless and noisy data. The results suggest that the proposed method is a powerful algorithm with good utility for EEG imaging.

Advances in neuroimaging have enhanced the clinician's ability to localize the epileptogenic zone in focal epilepsy, but 20-50 percent of these cases still remain unlocalized. Many sophisticated modalities have been used to study epilepsy, but scalp electrode recorded electroencephalography is particularly useful due to its noninvasive nature and excellent temporal resolution. This study is aimed at specific locations of scalp electrode EEG information for correlation with anatomical structures in the brain. 3D position localizing devices commonly used in virtual reality systems are used to digitize the coordinates of scalp electrodes in a standard clinical configuration. The electrode coordinates are registered with a high- resolution MRI dataset using a robust surface matching algorithm. Volume rendering can then be used to visualize the electrodes and electrode potentials interpolated over the scalp. The accuracy of the coordinate registration is assessed quantitatively with a realistic head phantom.

Generators of spontaneous human brain activity such as alpha rhythm may be easier and more accurate to localize in frequency-domain than in time-domain since these generators are characterized by a specific frequency range. We carried out a frequency-domain analysis of synchronous alpha sources by generating equivalent potential maps using the Fourier transform of each channel of electro-encephalographic (EEG) recordings. SInce the alpha rhythm recorded by EEG scalp measurements is probably produced by several independent generators, a distributed source imaging approach was considered more appropriate than a model based on a single equivalent current dipole. We used an imaging approach based on a Bayesian maximum entropy technique. Reconstructed sources were superposed on corresponding anatomy form magnetic resonance imaging. Results from human studies suggest that reconstructed sources responsible for alpha rhythm are mainly located in the occipital and parieto- occipital lobes.

Spatial heterogeneity of myocardial perfusion has been recognized for many years. We have previously shown that whole-body CT is a method for providing the simultaneous measurements of heterogeneity of myocardial perfusion and myocardial blood volume. In the present study we found that the spatial distribution of myocardial metabolism, as indicated by the local accumulation of iodinated phenyl pentadecanoic acid, is slightly more heterogeneous than, but not statistically different from, the heterogeneity of perfusion and blood volume. These findings are consistent with the notion that a common factor is likely to play a major role in determining the spatial heterogeneity of myocardial intravascular blood volume, of myocardial perfusion and of myocardial metabolism.

Information about local diameter variations as a response to the pulse flow in the human coronary arteries may indicate the development of atherosclerosis before this can be seen as a stenosis on coronary arteries may indicate the development of atherosclerosis before this can be seen as a stenosis on coronary angiograms. This paper describes the design of an image processing tool to measure this diameter variation from a sequence of digital coronary angiograms. If a blood vessel responds less elastically to the pulse flow, this may be an indication of atherosclerosis in an early stage. We have developed an image analysis and processing algorithm which is able after vessel segment selection by the user, to calculate automatically the vessel diameter variations from a standard sequence of digital angiograms. Several problems are treated. The periodic motion of the vessel segment in the consecutive frames is taken into account by tracking the vessel segment using a 2D logarithmic search to find the minimum in the mean absolute distance. A robust artery tracing algorithm has been implemented using graph searching techniques. The local diameter is determined by first resampling the image perpendicular to the found trace and afterwards performing edge detection using the Laplacian operator. This is repeated for all frames to show the local diameter variation of the artery segment as a function of time.

The site and mechanism of the dispersion of blood transit times within the pulmonary vascular bed can be described using x-ray angiography images of bolus passage through the pulmonary vasculature. Time-absorbance curves from the lobar inlet artery and outlet vein, various locations within the arterial and venous trees, and regions of the microvasculature were acquired from the images. The overall dispersion within the lung lobe was determined from the inlet arterial and outlet venous curves by examining the difference in their first and second moments, mean transit time and variance, respectively. Subsequently, the moments at each location within the arterial tree were calculated and compared to those of the lobar inlet artery curve. The transit time variance imparted on the bolus as it traveled through the pulmonary arterial tree upstream from the smallest measured arteries was < 5 percent of the variance attributable to transit through the total lung lobe vascular bed. Similar results were obtained for the venous pathways using reverse-flow conditions. Regional capillary mean transit time and variance were obtained from the measured microvascular residue curves using a mass-balance model. These results suggest that most of the bolus dispersion occurs within the pulmonary capillary bed rather than in large feeding arteries or draining veins.

Our previous studies have shown that renal perfusion can be visualized by imaging the transit of a contrast agent through the parenchyma of the organ using gray scale (GS) and power Doppler (PD) ultrasound.However, the relative merits and the sensitivities of the two imaging methods are not known. This study compares the effectiveness of the two modes in visualizing kidney perfusion at the clinical dose of contrast agents. GS and PD images of the dog kidneys were recorded using a clinical ultrasound scanner at 4-7 MHz. A fixed longitudinal plane of the kidney was imaged by mounting the transducer on the animal with a specially designed holder. A dose of 0.1 m1/kg of Echogen was injected intravenously and GS and PD images were recorded simultaneously on two separate time-encoded video tapes during the passage of the contrast agent through the kidneys. The enhancement of GS and PD images was assessed qualitatively by three radiologists. The quantitative assessment was made by measuring the regional and global enhancements of digitized B-scan and PS images. Regional measurements were made by comparing brightness of the post contrast images with that of a pre-contrast reference image pixel by pixel. Student t-test was used to determine the statistical significance of the change. The regions representing statistically significant differences were encoded on the image in color with brightness proportional to the magnitude of change. The regions with no significant change were represented in GS. This generated a series of new images, referred to as StatMap, with color representing regions of perfusion. Changes in power Doppler images were visually detectable with high confidence in all five dogs by al three radiologists. There was no perceptible changes in B-scans. Computer analysis of PD images yielded characteristic indicator dilution curves in all five dogs with an initial rise time of 2-5 sec and a peak at 7-20 sec. The enhancement in PD lasted for 97-400 seconds. The peak to pre-injection Doppler power ratio was 2.41 +/- 0.85. There were not detectable changes in gray scale images except in one dog which exhibited a small change. The StatMap images of PD exhibited perfusion over the entire kidney, whereas the GS images showed perfusion to be sparsely distributed.

The purpose of the present study is to improve the quantification of peripheral arterial stenosis using 3D power Doppler angiography and investigate the potential of this technique for generating the arterial tree of the lower limb for surgery planning. Stenotic wall-less agar arteries were created to simulate the femoral and carotid arteries. 3D power Doppler angiograms of those arteries were generated under different hemodynamic conditions using a 3D ultrasound imaging system developed by the Life Imaging System Inc. The effect of multiple stenoses on the 3D power Doppler angiograms was investigated using the femoral arterial phantoms. Using the carotid arterial phantoms, 3D power Doppler angiograms of the carotid arteries were generated and compared with the known geometry. To image a whole lower limb arterial tree for lower limb salvage surgery planning, multiple scans are required to cover the entire field-of- view interested by using a water-coupled scanner. Preliminary in vivo test was performed using water-coupled scanning.

At present, 3D reconstructions of coronary vessels are generated from intravascular ultrasound (IVUS) pullback sequences by stacking up ECG-gated segmented IVUS frames. Since this approach results in straight vessel reconstructions, it clearly fails in tortuous coronary arteries. This shortcoming may be overcome by data fusion with biplane angiography. The 3D course of the vessel is first derived from angiograms and then combined with segmented IVUS images in order to obtain a spatially correct and anatomically complete vessel reconstruction. In this paper, different approaches to two problems associated with the data fusion method are discussed: the definition of the pullback path and the estimation of IVUS catheter twist during pullback. A procedure for image acquisition, segmentation, mapping, and interpolation is proposed that has been designed with respect to its clinical applicability. In vitro validations of our previously reported algorithm for analytic catheter twist quantification is presented along with data fusion results of coronary arteries in cadaveric pig hearts.

Shear rate has been linked to endothelial and smooth muscle cell function, neointimal hyperplasia, poststenotic dilation and progression of atherosclerotic plaque. In vivo studies of shear rate have been limited in humans due to the lack of a truly accurate noninvasive method of measuring blood flow. In clinical vascular laboratories, the primary method of wall shear rate estimation is the scaled ratio between the center line systolic velocity and the local arterial radius. The present study compares this method with the shear rate calculated directly from data collected using a Doppler ultrasound scanner. Blood flow in the superficial femoral artery of 20 subjects was measured during three stages of distal resistance. Analysis and display programs were written for use with the MATLAB image processing software package. The experimental values of shear rate were calculated using the formal definition and then compared to the standard estimate. In all three states of distal resistance, the experimental values were significantly higher than the estimated values by a factor of approximately 1.57. These results led to the conclusion that the direct method of measuring shear rate is more precise and should replace the estimation model in the clinical laboratory.

Purpose: To develop a valid, reliable and accurate system of measurement of intracranial aneurysm geometry using volumetric data obtained by CT angiography. Materials and methods: A simple model of lateral saccular aneurysm was created. Three models were prepared with different size of aneurysm sac and neck. Volumetric data was acquired using a Toshiba Xpress SX helical scanner. Geometry of an aneurysm model obtained by workstation linked to the scanner applying volume rendering display and dedicated UNIX based computer applying MPR based method. These results were compared with actual caliper measurements of the model. A clinical case of lateral aneurysm arising from the supraclinoid internal carotid artery was also studied. Results: Both the volume rendered image based method and MPR based method provided accurate geometric information of an aneurysm sac and its neck. Conclusions: Volume rendering technique requires editing by a well-informed operator and subjective, while the MPR based method is more objective and better suited for quantitative analysis. Using these mutually complimentary tools, critical geometric information of an aneurysm can be extracted from volumetric data provided by CTA.

A convenient method for assessing cerebral perfusion and related functional parameters has been developed using a third generation slip-ring CT scanner. Dynamic contrast- enhanced scanning at the same level was employed to image the cerebral circulation at the rate of 1 image per second. Using data acquired with this non-helical mode of scanning, we have developed a method for the simultaneous in-vivo determination of cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT). These measurements are given in the same physiological units as positron emission tomography. In order to obtain accurate measurements of these parameters, methods were also developed to correct for recirculation and partial volume averaging in imaging small blood vessels. We have used 6 New Zealand white rabbits in our studies. For each rabbit, up to 3 CT measurements of CBF, CBV, and MTT were made at normocapnia under isoflurane anesthesia. Coronal sections through the brain were imaged while simultaneously imaging either a brain artery or the ear artery. Images were acquired for 1 minute as Isovue 300 was injected intravenously. In the acquired CT images, regions of interest in brain parenchyma and an artery were drawn. For each region of interest, the mean CT number in pre-contrast images was subtracted from the mean in post-contrast images to calculate the contrast concentration curves for the brain regions Q(t) and the arterial region Ca(t). Using a robust deconvolution method, the MTT was determined. CBV was then determined from the ratio of the areas of Q(t) and Ca(t). Finally, CBF was calculated from the Central Volume Principle. The mean regional CBF, CBV and MTT values were 73.3 +/- 5.1 ml/min/100g, 1.93 +/- 0.12 ml/100g and 1.80 +/- 0.18 s respectively. IN order to validate our CT CBF measurements, we also measured CBF using the well- established technique of microspheres with each CT study. The feasibility of our CT method to measure CBF accurately was demonstrated by an almost one-to-one correspondence between CT CBF and microspheres CBF measurements. The slope of the linear regression line between the two sets of CBF measurements was 0.97 +/- 0.03 with a correlation coefficient of 0.837.

Detailed blood flow patterns appear to have a substantial effect upon arterial disease and the outcome of interventional endovascular therapeutic procedures. First, a demonstration of the effect of a stent on eliminating vortex flow in an aneurysm model is given. Next, a method is presented to measure the detailed 3D flow during pulsatile flow in a phantom. Finally, the method is applied to arteriovenous malformations (AVMs). The method consists of the real-time radiographic tracking of contrast droplet paths with images from a biplane pulsed radiographic system used to derive the 3D particle location as a function of time. An example of droplet tracking in a sidewall aneurysm phantom is presented. The application of the method is already impacting the delineation of the detailed morphology and of flow characteristics of AVMs and the selection of transit times for endovascular occlusion of AVMs using cyanoacrylate in patients.

Initial hyperplasia is major cause of restenosis after vascular interventions for arterial occlusive disease. We reported that a fluorescent probe, Texas Red, conjugated to the scavenger-receptor ligand, maleylated bovine serum albumin accumulated almost exclusively in the injured, hyperplastic sites. The purpose of this study is to test the feasibility of enhanced drug delivery to the hyperplastic lesion by targeting the scavenger-receptors.

The accurate and clinically useful estimation of the shape, motion and deformation of the heart's left ventricle (LV) is an important research problem. Recently, computer vision techniques for reconstructing the 3D shape and motion of the LV have been developed. However, the main drawback of these techniques is that the models are formulated in terms of either too many local parameters that require non-trivial processing to be useful for close-to-real-time diagnosis, or too few parameters to offer an adequate approximation to the LV motion. Method: To address the problem of a compact and accurate LV shape representation, we developed a new class of volumetric deformable models, whose parameters are functions rather than constants. These new models were based on the adaptation of the parametric definition of a superquadric whose parameters capture the radial and longitudinal contraction, the axial twisting, and the long-axis deformation. Using a physics-based approach and Lagrangian dynamics, we converted these volumetric geometric models into models that deform due to forces exerted from the datapoints. A novel algorithm to combine two orthogonal sets of planar motion was implemented to estimate the full 3D motion. We also developed a comprehensive method for visualizing the motion of the LV and for comparing normal and abnormal hearts in a way that is readily understood by a clinician. The methods included plotting parameter graphs, coloring techniques, tracking the motion paths of points on the LV, and video sequences displaying the nonrigid motion of the LV using standard SGI (Mountain View, CA) hardware. We applied our models to datasets obtained by a MR tagging method known as SPAMM (Spatial Modulation of Magnetization). In order to estimate the full 3D motion, datasets from short-axis and long-axis views of a heart were used during the model fitting procedure to compute 3D forces from the datapoints to the deformable model. The forces were then used to estimate the parameters which describe the 3D nonrigid motion of the LV. The extracted parameter functions were then displayed using our visualization methods. We also applied our models to datasets of patients with hypertrophic cardiomyopathy for a comparison with data from healthy volunteers. Results: By plotting the variation over time of the extracted LV model parameters, we were able to quantitatively analyze, localize, and compare the epicardial and endocardial motion. Based on the small number of parameters which vary from region to region, we were able to describe the complex volumetric shape and motion of the LV. Since each parameter function describes a clinically useful type of deformation in an intuitive way, it did not require any complex post-processing for a meaningful interpretation. The comparison study showed that the patients with hypertrophic cardiomyopathy had distinct and consistent differences in the extracted parameters as compared to the normal subjects. Conclusion: The results demonstrate that the proposed technique for estimating the 3D left ventricular wall motion provides useful visualization and quantitative motion analysis of the LV throughout its volume. Keywords: heart wall motion, left ventricle, MRI tagging, SPAMM, regional cardiac function, heart visualization, deformable model, parameter functions

The study of cardiac activation dynamics is an important factor in the characterization of the cardiac function. One such example is the localization of WPW-pathways inside the myocardium. Accurate localization of these pathways can be used to determine if the patient should be treated with catheter techniques or surgical techniques. This paper analyzes the temporal information in tissue velocity imaging with both qualitative and quantitative methods. The clinical experiments indicate that echocardiography can become an alternative technique for non-invasive electrophysiology in these kinds of applications.

We have previously reported an automated approach to detection of endocardial and epicardial borders in individual intracardiac ultrasound (ICUS) images. Here, we report the method's extension to 3D ICUS image data sets. Our method is based on fully automated detection of epicardial and endocardial borders inside a single interactively identified region of interest. BOrder detection is based on an optimal graph-searching approach that utilizes a priori knowledge about left ventricular (LV) anatomy and ultrasound imaging physics. Eight cadaveric pig hearts were used for validation. Two ICUS sequences were obtained from each heart, with a 10 MHz CVIS 10F catheter positioned in the LV across (1) the aortic valve and (2) the mitral valve. Performance of the 3D automated border detection method was assessed by comparing the observer- defined and computer-determined quantitative indices of LV volume and by border positioning errors. The 3D reconstruction of the lV was performed from the sequences of the detected epicardial and endocardial borders using shape- based interpolation and surface rendering.

The purpose of this work is to study the displacement of myocardial walls. We have used 2D echographic images acquired in different apical views. 'Fluctuations' between extracted contours complicate accurate measurement and analytical interpretation. Too smooth these contours, we use snakes because of the elastic movement of the left ventricle (LV). The study of the walls' displacement is done in each incidence. The intersection between ventricle's contours and parallel planes to cross sections generates segments. The tracking of these segments' length, during the cardiac cycle, allows us to analyze the movement. This study allows to compute volumes, to determined pathologies related to the dysfunctioning of the ventricle's walls and to control the excitation's propagation intended to ensure the heart's contraction. Classical methods consist in modeling the ventricle from reconstructions with surfaces or parametric models. Nevertheless, it is difficult to exploit data and to determine parameters indicating precisely the present of a pathology. Our method allows to determine pathological zones more precisely. The first results obtained with few incidences are correct. Yet, more important the number o incidences is more complete diagnosis will be.

In this paper we present a local force model and its integration in a hierarchical analysis of the estimation of the left ventricle motion over a cardiac cycle. The local force model is derived from the dynamics of interconnected point masses driven by local constant forces over a short time. The forces drive the interconnected point masses within a regional patch of the left ventricle surface from one time instant to another. The trajectory that minimizes the total energy required to move the masses from one surface to another is considered as the local displacement vector. This estimated trajectory takes into account surface constraints and previous estimations derived from the volumetric images sequences so that the point masses travel along smooth trajectories resembling the realistic left ventricle surface dynamics. This proposed model is able to recover the point correspondences of the nonrigid motions between consecutive frames when the surfaces and the initial conditions of the left ventricle at consecutive time frames are given. The local force model is incorporated into a hierarchical analysis scheme providing us with the complete dynamics of the left ventricle as compared to the local kinematic analysis of previous approaches. Experimental results based on synthetic and real left ventricle CT volumetric images show that the proposed scheme is very promising for cardiac analysis.

We present a new method for detecting and tracking tissue motion in 3D. The method is based on the concept of tracking speckle patterns in 3D using the sum absolute difference (SAD) technique. One potential application of this method is to study the 3D motion of various regions of interest in myocardial tissue using volumetric ultrasound scans of the heart. This could be of great value in assessing the viability of the myocardium. Simulations of 3D speckle patterns were obtained for the real-time ultrasound volumetric scanner developed at Duke University. Volumes of data were studied in pairs. Motion was simulated as whole voxel translations in 3D. A kernel volume was selected and a larger surrounding search volume was then defined. The kernel volume was compared to al possible matching sub- volumes in the search volume using the SAD technique. After the best match was found, the 3D components of motion were calculated by measuring the relative shift of the best match sub-volume from the location of the kernel volume the process was repeated until a 3D map of motion for the first frame was obtained. The performance of the proposed tracking method as a function of the SNR was the plotted for each direction. Experiments were performed to evaluate the performance of the method in vitro. A tissue mimicking materials was imaged using a 5MHz piston transducer translated in 2D to obtain multiple volumes with known shifts. The tracking method was applied and its performance with different shift values was evaluated. The calculated shift values highly matched the true shift values within a small jitter error.

Traumatic mechanical loading of the head-neck complex results cervical spinal cord injury when the distortion of the cord is sufficient to produce functional or structural failure of the cord's neural and/or vascular components. Characterizing cervical spinal cord deformation during physiological loading conditions is an important step to defining a comprehensive injury threshold associated with acute spinal cord injury. In this study, in vivo quasi- static deformation of the cervical spinal cord during flexion of the neck in human volunteers was measured using magnetic resonance (MR) imaging of motion with spatial modulation of magnetization (SPAMM). A custom-designed device was built to guide the motion of the neck and enhance more reproducibility. the SPAMM pulse sequence labeled the tissue with a series of parallel tagging lines. A single- shot gradient-recalled-echo sequence was used to acquire the mid-sagittal image of the cervical spine. A comparison of the tagged line pattern in each MR reference and deformed image pair revealed the distortion of the spinal cord. The results showed the cervical spinal cord elongates during head flexion. The elongation experienced by the spinal cord varies linearly with head flexion, with the posterior surface of the cord stretching more than the anterior surface. The maximal elongation of the cord is about 12 percent of its original length.

It is very difficult to study the biomechanics of individuals tarsal joints in living people because the usual method for in vivo analysis, tracking skin markers related to deeper joints, works poorly in this region. Other available methods for analyzing individual joints are too invasive for practical use. We have developed a method which utilizes 3D reconstructions of bones from magnetic resonance image data. The foot is moved in a controlled manner in a jig which mimics normal pronation and supination, with data gathered at 10 degree intervals across the range of motion. At each position of the foot a 3D reconstruction of each bone is made, and the motions which occur as the bone goes from one position of the foot to the next are determined. From this information, animations are created and the position data are analyzed. We have studied twenty clinically normal individuals, plus several people being treated for various pathologies. Kinematic information on the peritalar joints include such factors as the nature of their rotations and the relative amount of pronation- supination occurring at the subtalar and transverse tarsal joints.

An imaging-microscopy methodology to measure contraction movement in chemically stimulated crustacean skeletal muscle, whose movement speed is about 0.02 mm/s is presented. For this, a CCD camera coupled to a microscope and a high speed digital image acquisition system, allowing us to capture 960 images per second are used. The images are digitally processed in a PC and displayed in a video monitor. A maximal field of 0.198 X 0.198 mm² and a spatial resolution of 3.5 micrometers are obtained.

Recent studies suggest that visual evaluation of ultrasound images could decrease negative biopsies of breast cancer diagnosis. However, visual evaluation requires highly experienced breast sonographers. The objective of this study is to develop computerized radiologist assistant to reduce breast biopsies needed for evaluating suspected breast cancer. The approach of this study utilizes a neural network and tissue features extracted from digital sonographic breast images. The features include texture parameters of breast images: characteristics of echoes within and around breast lesions, and geometrical information of breast tumors. Clusters containing only benign lesions in the feature space are then identified by a modified self- organizing map. This newly developed neural network objectively segments population distributions of lesions and accurately establishes benign and equivocal regions.t eh method was applied to high quality breast sonograms of a large number of patients collected with a controlled procedure at Mayo Clinic. The study showed that the number of biopsies in this group of women could be decreased by 40 percent to 59 percent with high confidence and that no malignancies would have been included in the nonbiopsied group. The advantages of this approach are that it is robust, simple, and effective and does not require highly experienced sonographers.

For the purpose of locating breast lesions, we have utilized holographic interferometry (HI) to generate a fringe contour distortion superficial to an indwelling tumor. HI has been successfully utilized to examine material defects or dynamic characteristics of laminated structure. Application of HI technique in medical imaging has been reported. HI requires appropriate stressing mechanism to produce fringe patterns of reasonable density and contrast. Based on our research, it was found that pressure stressing has the advantages of providing both adequate density and patterns of fringe lines and an easily controlled stressing capacity. In our setup, both breasts of a subject hang within a Lucite pressure chamber while the intramural pressure of the chamber is raised by reducing volume of air inside. Our preliminary results found distorted fringe lines over breast lesions. The accuracy of the developed system in localizing breast lesions was determined by correlating results to those of x- ray mammography.

There has been an increased interest in automatic segmentation of volumetric medical image data. One of the reasons is that automated segmentation takes away the variability which exists when data is segmented manually. It also reduces processing time significantly. However, because of the stochastic nature of biological structures and the fact that no two data sets and scanner models are alike, it is very important to develop automated methods which process images in an adaptive manner and use a priori information to simplify the process. The method which we present here adaptively determines thresholds in order to segment out the primary human airway tree and uses some a priori information about the manner in which branching occurs, specifically the order in which the upward and downward branches arise from the right and left bronchi. We present preliminary results from this method, which automatically segments out the first four generations of the airway tree reliably, in data sets from both normal and airway comprised subjects and present comparisons with the current 'gold standard' of manual segmentation.

The advent of spiral computed tomography (CT) has created the potential to image continuous anatomical volumes during a single breath-hold. The ability to reconstruct overlapping spiral CT images has improved through-plane resolution and contributed to improved diagnostic accuracy. When spiral CT is used to image organ systems such as the colon or airways, it is common to generate up to 500 CT images. We have developed a virtual endoscopy (VE) software system that couples computer-assisted diagnosis capabilities with volume visualization techniques to aid in the analysis of these large datasets. Despite its potential to assist in disease diagnosis, VE faces several important technical and nontechnical challenges that must be addressed before it becomes a clinical reality.

The emerging techniques of 3D spiral CT for 'virtual colonoscopy' show promise as a noninvasive screening modality for the detection of polyps. Our purpose was to evaluate three key post-processing parameters required for depiction of colonic polyps using perspective volume rendering (PVR): image reconstruction, window setting, and opacity map assignment of the attenuation histogram. Spiral CT scans of two different patients with known polyps confirmed by colonoscopy were performed. First, image quality was compared between images generated after interpolation of raw projection data and interpolation of reconstructed image data for longitudinal voxel dimensions of 1x, 2x, 4x, 6x and 8x in-plane voxel dimensions. Second, the dimensions of colonic polyps relative to haustral folds were measured on PVR images for various window settings and compared to similar measures performed on photography performed at colonoscopy. Third, a double sigmoidal and a stair-step opacity function were each applied to a 3D PVR image of a polyp, and quantitative differences in image smoothness were compared using a texture analysis method. In conclusion, spiral CT images reconstructed with 50 percent overlap and displayed with a standard display window permit accurate depiction of polyp dimensions relative to surrounding structures on PVR windows. Image artifacts may be suppressed with use of a double sigmoidal opacity map.

An image processing system developed to support evaluation of computed tomographic colonography (CTC) was presented at this conference in 1996. There is growing acceptance of the merits of interactive, intraluminal volume rendering for CTC, however, most current CTC systems contain technical problems that prohibit their routine clinical application. These problems include: the large amount the data that requires processing, the need for extensive preprocessing, limitations of rendered scene parameters imposed during preprocessing, and long interpretation times.

Clinical signs of radiotherapy failure are often not present until well after treatment has been completed. Methods which could predict the response of tumors either before or early into the radiotherapy schedule would have important implications for patient management. Recent studies performed at our institution suggest that MR perfusion imaging maya be useful in distinguishing between individuals who are likely to benefit from radiation therapy and those who are not. Because MR perfusion imaging reflects tissue vascularity as well as perfusion, quantitative positron emission tomographic (PET) blood flow studies were performed to obtain an independent assessment of tumor perfusion. MR perfusion and PET quantitative blood flow studies were acquired on four women diagnosed with advanced cervical cancer. The MR perfusion studies were acquired on a 1 cm sagittal slice through the epicenter of the tumor mass. Quantitative PET blood flow studies were performed using an autoradiographic technique. The PET and MRI were registered using a manual interactive routine and the mean blood flow in the tumor was compared to the relative signal intensity in a corresponding region on the MR image. The mean blood flow in the cervical tumors ranged form 30-48 ml/min/100 grams. The observed blood flow values are consistent with the assumed relationship between MR contrast enhancement and the distribution of tissue perfusion. The information offered by these studies provides an additional window into the evaluation of the response of cervical tumors to radiation therapy.

Bone autografts are routinely employed in the reconstruction of facial deformities resulting from trauma, tumor ablation or congenital malformations. The combined use of post- operative 3D CT and SPECT imaging provides a means for quantitative in vivo evaluation of bone graft volume and osteoblastic activity. The specific objectives of this study were: (1) Determine the reliability and accuracy of interactive computer-assisted analysis of bone graft volumes based on 3D CT scans; (2) Determine the error in CT/SPECT multimodality image registration; (3) Determine the error in SPECT/SPECT image registration; and (4) Determine the reliability and accuracy of CT-guided SPECT uptake measurements in cranial bone grafts. Five human cadaver heads served as anthropomorphic models for all experiments. Four cranial defects were created in each specimen with inlay and onlay split skull bone grafts and reconstructed to skull and malar recipient sites. To acquire all images, each specimen was CT scanned and coated with Technetium doped paint. For purposes of validation, skulls were landmarked with 1/16-inch ball-bearings and Indium. This study provides a new technique relating anatomy and physiology for the analysis of cranial bone graft survival.

A quantitative characterization of the shape of the right ventricle (RV) of the heart is needed for accurate modeling of the mechanics of the ventricle as well as for better measuring the volume of the ventricle from technologies such as 2D ultrasound, bi-planar ventriculography, and sonomicrometry. A technique was thus developed for modeling RV shape. First, a high-resolution MR image set was obtained of the freshly excised lamb heart under various passive pressurizations of both ventricles ranging from 5 to 30 cmH₂O simulating end-diastole in the beating heart. Typically, 2-3 full images were obtained for each heart. Images were obtained with a multislice spin-echo T1-weighted sequence with the slice plane orientation early equal to the short-axis view of the heart. A 3D characterization of shape was obtained by first characterizing inter-slice changes in shape and orientation and then characterizing the shape of a single representative slice. The slice chosen to represent the RV was in the region directly below the tricuspid valve since it is both near to the apex-base center of the RV and has the greatest size. Intuitive deformations were applied to an initial circular arc anchored at the endpoints of the freewall and initially passing through a point near the center of the freewall contour, so as to best match the true freewall contour. These include a leaning of the circular arc parallel to the septal axis, a flattening perpendicular to the septal axis, a tucking-in or sharpening of the curvature near the junction with the septum, and a pinching- in at a point or points near its center towards the septum, all, in an attempt to account for the asymmetry and non- circularity deformed circular arc which effectively produces tow independent arcs. For all but one of the anterior and posterior arcs in 13 heart shapes, pinch-deformed arcs could be obtained whose average radial distance from the true RV chamber contour was less than 0.9 mm and averaged 0.5 mm for the anterior arc and 0.66 mm for the posterior arc. Worst- case deviation in parameters of the pinched-arc model of cross-sectional shape, and radii of curvature and the RV freewall-septum junction angle due to worst-case deviation in landmark location are 19 percent +/- 11 percent, 10 percent +/- 6 percent, 10 degrees +/- 4 degrees, and 12 degrees +/- 6 degrees. If landmark localization variability is minimized with a rigid translate offset scale model of the landmark region, average measurement error as determined in an adjacent slice comparison was 11 percent +/- 5 percent, -3 percent +/- 2 percent, 7 degrees +/- 2 degrees, and -7 degrees +/- 2 degrees.