The aim of this study is to obtain voxel-by-voxel images of binding parameters between [¹¹C]-flumazenil and benzodiazepine receptors using positron emission tomography (PET). We estimate five local parameters (k₁, k₂, B'_max, k_on/V_R, k_off) by fitting a three- compartment ligand-receptor model for each voxel of a PET time series. It proves difficult to fit the ligand-receptor model to the data. We trade noise and spatial resolution to get better results. Our strategy is based on the use of a multiresolution pyramid. It is much easier to solve the problem at coarse resolution because there are fewer data to process. To increase resolution, we expand the parameter maps to the next finer level and use them as initial solution to further optimization, which then proceeds at a fast pace and is more likely to escape false local minima. For this approach to work optimally, the residue between data at a given pyramid level and data at the next level must be as small as possible. We satisfy this constraint by working with spline-based least- squares pyramids. To achieve speed, the optimizer must be efficient, particularly when it is nearing the solution. To that effect, we have developed a Marquardt-Levenberg algorithm that exhibits superlinear convergence properties.

Physiologic systems can be represented by compartmental models which describe the uptake of radio-labeled tracers from blood to tissue and their subsequent washout. Arterial and venous time-activity curves from isolated heart experiments are analyzed using spectral analysis, in which the impulse response function is represented by a sum of decaying exponentials. Resolution and uniqueness tests are conducted by synthesizing isolated heart data with predefined compartmental models, adding noise, and applying the spectral analysis technique. Venous time-activity curves are generated by convolving a typical arterial input function with the predefined spectrum. The coefficients of a set of decaying exponential basis functions are determined using a non- negative least squares algorithm, and results are compared with the predefined spectrum. The uniqueness of spectral method solutions is investigated by computing model covariance matrices, using error propagation and prior knowledge of noise distributions. Coupling between model parameters is illustrated with correlation matrices.

Virtual bronchoscopy reconstructions of the airway noninvasively provide useful morphologic information of structural abnormalities such as stenoses and masses. In this paper, we show how virtual bronchoscopy can be used to perform aerodynamic calculations in anatomically realistic models. Pressure and flow patterns in a human airway were computed noninvasively. These showed decreased pressure and increased shear stress in the region of a stenosis.

We are presenting a comprehensive system for fusion of intravascular ultrasound (IVUS) data and x-ray angiography, aiming to create a geometrically accurate 3-D or 4-D (3-D plus time) model of the coronary vasculature. For hemodynamic analyses, methods of computational fluid dynamics (CFD) are applied to the reconstructed data, resulting in quantitative estimates of the wall shear stress. Visualization is performed using the Virtual Reality Modeling Language (VRML). Lumen and adventitia borders are modeled as surfaces using indexed face sets; quantitative results are encoded as color per vertex. The endoscopic mode (virtual angioscopy) allows an interactive fly-through animation with variable speed along with arbitrary positioning within the vessel. Since this functionality exceeds those of the standard VRML animation nodes, an external prototype library containing VRML and JavaScript definitions has been developed that provides a 3-D graphical user interface to navigate within the endoscopic mode. The control panel is available on demand, but does neither obstruct any vessel features when not needed, nor does it limit the viewport for the scene. Preliminary results showed a good feasibility of the overall procedure, and a high reliability of the fusion and CFD methods as well as the visualization with the virtual endoscopy VRML library.

Kawasaki Disease is an inflammatory illness of young children that can seriously affect the cardiovascular system. The disease may cause coronary artery aneurysms, a thinning and dilation of the arterial wall when the wall is weakened by disease. Such aneurysms significantly increase the risk of rupture of the arterial wall, an event from which few patients survive. Due to the largely asymptotic nature of coronary aneurysms, diagnosis must be timely and accurate in order for treatment to be effective. Currently, aneurysms are detected primarily using X-ray angiography, MRI, and CT images. Increased insight into the disease and its effects on the arterial wall can be gained by multi-dimensional computerized visualization and quantitative analysis of diagnostic images made possible by the techniques of intravascular imaging and virtual endoscopy. Intravascular ultrasound images (IVUS) of a coronary artery exhibiting aneurysms were acquired from a patient with Kawasaki Disease. The disease is characterized by low luminescent in the IVUS images. Image segmentation of the abnormal, prominent anechoic regions branching from the lumen and originating within other layers of the arterial wall was performed and each region defined as a separate object. An object segmentation map was generated and used in perspective rendering of the original image volume set at successive locations along the length of the arterial segment, producing a 'fly-through' of the interior of the artery. The diseased region (aneurysm) of the wall was well defined by the differences in luminal size and by differences in appearance of the arterial wall shape observed during virtual angioscopic fly-throughs. Erosions of the endovascular surface caused pronounced horizontal and vertical ballooning of the lumen. Minute cracks within the unaffected luminal areas revealed possible early development of an aneurysm on the contralateral wall, originating in the medial section of the artery and spreading outward toward the lumen.

CT colonography is a promising technique with a long-term goal to provide mass screening for colorectal carcinoma. Colorectal screening by CT colonography requires that the examination be cost-effective. The correct interpretation time is excessive for a screening test. Therefore, a computerized detection method capable of indicating regions of suspicion is attractive as a diagnostic aid for radiologists. We have developed a new CAD scheme for automated detection of polyps based on CT colonographic data sets. Our method characterizes polyps by geometric features of volumetric data including the volumetric shape index and curvedness. Polyps were detected by fuzzy clustering in a feature space generated by the feature values and spatial coordinates, followed by a rule-based test in the feature space. In an analysis of 41 patients, 9 of whom had at least one biopsy-proved polyp, our CAD scheme detected 100% of polyps with 2.5 false positives per patient. Our preliminary result indicates that the CAD scheme is potentially useful for highlighting areas of suspicion in the colon and, therefore, facilitates widespread screening by reducing the reading time substantially.

We propose a new approach of patient-specific simulation and planning of vascular interventions using standard CT data. It is based on the virtual environment developed for the exploratory navigation within 3D volume images. Tools actions involved in the considered minimal invasive procedure have been analyzed and we present the basis of a methodological framework for the simulation of particular endovascular procedure, i.e. the artery wall expansion using a non- compliant balloon. We especially describe the characterization step, which concerns the accurate and automatic detection of the internal wall of the vessel, the description of the lumen surface, the extraction of a region of interest and the analysis of the parietal quality including the lesion. A preliminary surface model of the tool, interacting with a volume representation of the vessel wall, has been introduced in order to predict the plaque behavior during the surgical procedure. Thus, the ideal surface of the lumen and the final shape of the vessel as well as the optimal tool parameters (size, position) taking into account the lesion characteristics could be computed in the simulation process to perform actually a patient-specific planning of intervention.

The purpose of the current study is to measure the carbon dioxide reactivity of blood flow in VX2 tumor in the rabbit thigh. The carbon dioxide reactivity of the functional parameters was investigated in eight rabbits by changing the ventilation rate in order to manipulate the arterial carbon dioxide tension (P_aCO₂). In each experiment, functional maps were generated at four P_aCO₂ levels: normocapnia (P_aCO₂ equals 40.7 +/- 1.4 mm Hg), hypocapnia (27.1 +/- 2.5 and 33.7 +/- 2.2) and hypercapnia (53.8 +/- 5.2). The carbon dioxide reactivity of tumor blood flow showed significant differences between normocapnia and the two levels of hypocapnia, but not between normocapnia and hypercapnia. The average fractional change of blood flow from normocapnia for the two hypocapnic level was -0.41 +/- 0.06 and -0.29 +/- 0.08, respectively. The ability to reduce blood flow through hypocapnia has significant implications in thermal therapy, as heat dissipation represents a major obstacle which limits the effectiveness of treatment.

Blood flow temporal waveforms change with position along an artery. The change in the flow waveforms can be accounted for by a transmission line model of flow. According to this model, pulse waves propagate at a finite velocity in both directions along the artery. In principle, given flow waveforms measured at three locations along an artery, the pulse-wave velocity, (c) can be determined from the wave equation (d²Q/dt² equals c²d²Q/dz², Q is flow, t is time, z is position). Given the vessel diameter, the vessel-wall compliance can be derived from pulse-wave velocity. However, direct solution of the wave equation for pulse-wave velocity is highly susceptible to flow-measurement error. Thus, we propose a new method for estimating pulse-wave velocity from arterial flow waveforms. In our method, ideal flow waveforms are reconstructed from three measured flow waveforms. The ideal waveforms are reconstructed by minimization of the total error between the ideal and measured waveforms subject to constraints of the wave equation. Ideal flow waveforms are reconstructed for a range of assumed pulse-wave velocities. The true pulse-wave velocity is considered to be that which produces the minimum total error. The method applied to blood flow measurements made with phase-contrast magnetic resonance imaging.

The visualization of the left ventricle (LV) motion in gated SPECT studies is complicated by the fact that 3D density images cannot be directly presented using common display devices. A number of techniques, most of them concerned with visualization, have been developed to aid in the classification of the images. However, it has been shown that interpretation of LV images by strictly visual techniques is subject to errors and inconsistency. For this reason, assistance in diagnosis can be improved only through the development of automatic or semi-automatic methods to analyze and quantify LV parameters. In this work, we propose an automatic method to estimate the myocardial kinetic energy directly from gated SPECT sequences. It is based on our previous work improved by a multi-resolution technique. Specifically, the method quantifies the 3D LV motion by a series of 3D velocity vector fields computed automatically for each voxel on the sequence of images. Based on this vector analysis it is possible to estimate a new physiological parameter, the kinetic energy index (kef), that may be an indication of the cardiac condition. The proposed method was applied in a group of 30 volunteers. The ke index measured for this population was 0.8582 +/- 0.0365.

Regional heart wall dynamics has been shown to be a sensitive indicator of LV wall ischemia. Rates of local LV wall thickening during a cardiac cycle can be measured and illustrated using functional parametric mappings. This display conveys the spatial distribution of dynamic strain in the myocardium and thereby provides a rapid qualitative appreciation of the severity and extent of the ischemic region. 3D reconstructions were obtained in an anesthetized pig from 8 adjacent, shortaxis, slices of the left ventricle imaged with an Electron Beam Computer Tomograph at 11 time points through one complete cardiac cycle. The 3D reconstructions were obtained before and after injection of 100 micrometer microspheres into the Left Anterior Descending (LAD) coronary artery. This injection causes microembolization of LAD artery branches within the heart wall. The image processing involved radially dividing the tomographic images of the myocardium into small subdivisions with color encoding of the local magnitude of regional thickness or regional velocities of LV wall thickening throughout the cardiac cycle. We compared the effectiveness of animation of wall thickness encoded in color versus a static image of computed rate of wall thickness change in color. The location, extent and severity of regional wall akinesis or dyskinesis, as determined from these displays, can then be compared to the region of embolization as indicated by the distribution of altered LV wall perfusion.

Fifteen patients underwent resting echocardiography (EC), ECG gated cardiac MR ventriculography (MRV) and blood pool planar and SPECT ventriculography (SPV) sequentially on the same day. In addition, 36 patients had sequential ECG gated blood pool and SPV and 20 normal volunteers, age > 18 years, had sequential ECG gated cardiac MRI performed on both Siemens closed, 1.5T, and open, 0.2T, magnets. Echocardiography was performed using a HP 5500 system equipped with an S4 transducer in 2D mode. MRV at 0.2T and 1.5T used a circular polarized body coil. Nuclear Medicine studies used 25 mCi Tc- 99m labeled red blood cells. Gated planar and SPV were acquired on a dual head Siemens E-Cam system. We have found that MRV affords the most accurate measurement of ventricular function. SPV and MRV provide similar estimations of left ventricular function (LVEF). Further, SPV consistently provides higher LVEF, as compared to the planar data simultaneously acquired. Observed significant differences in intermodality measurements indicate that follow up studies in patients, especially in patients whose management is critically dependent on functional measurement changes, should be monitored by one modality only.

Transbronchial needle biopsy is a common procedure for early detection of lung cancer. In practice, accurate results are difficult to obtain, since the bronchoscopy procedure requires a blind puncture into a region hidden behind the airway walls. This paper presents an image-guided endoscopy system for procedure preplanning and for guidance during bronchoscopy. Before the bronchoscopy, a 3D CT scan is analyzed to define guidance paths through the major airways to suspect biopsy sites. During subsequent bronchoscopy, the paths give the physician step-by-step guidance to each suspect site location. At a suspect site, a virtual CT image is registered to the bronchoscopic video. Then, the predefined biopsy site, from the prior CT analysis, is rendered onto the registered video. This gives the physician a reference for performing the needle biopsy. This paper focuses on our recent experiments with this system. These experiments involve a rubber phantom model of the human airway tree and in vivo animal tests. The experiments demonstrate the promise of our approach.

This paper proposes an improved method for camera motion tracking of a real endoscope based on registration endoscopic images and Computerized Tomography (CT) images. Camera motion estimation is a fundamental function of an endoscope navigation system, which provides useful navigation information to medical doctors during endoscopic examinations. Our previous method consists of two steps: (1) rough estimation of camera motion using optical flow information and (2) precise estimation performed by the image based registration technique. The problem of the previous method was that only the forward and the backward motion of the camera were estimated by the optical flow information. To solve this problem, the proposed method effectively uses the change in texture information on real endoscopic video images and estimates the camera motion by mapping the texture to the organ's 3-D shape generated from the CT images. We roughly estimate the all motion of the camera using texture information of the organ's wall, then the precise estimation is performed by using image based registration method. We applied the proposed method to the real bronchoscopic video images and 3-D X-ray CT images. The results sowed that the proposed method was superior to our previous method in many cases.

This paper shows a method for detecting unobserved regions (oversight regions) during fly-through on Virtual Endoscopy System (VES). When a VES is used as a diagnostic tool, e.g., to find colonopolyps, it is very important for doctors to observe the whole target organ without unobserved regions. The proposed method consists of three parts for detecting unobserved regions: (a) recording of observed triangle patches or voxels, (b) calculation of unobserved regions, and (c) display of unobserved regions. First, the system marks triangle patches that are displayed on the screen at each frame as 'observed patches' in the case of the surface rendering. Presented voxels are calculated in the case of the volume rendering. We execute this process for all frames of the fly-through. Then, the method labels those triangle patches or voxels that do not have 'observed' marks as 'unobserved triangle patches or voxels.' We calculate connected 'unobserved' patches or voxels and consider them to be 'unobserved regions.' In the presentation step, the system displays the target organ while coloring unobserved regions. The method also automatically presents each unobserved region by setting the viewpoint and the view direction around the region. We implemented the proposed method in a VES and applied the VES to colon regions. The experimental results showed that the proposed method can detect unobserved regions and display them effectively.

The purpose of this study is the evaluation of a virtual endoscopy system, which provides new possibilities for the exploration of vascular structures from standard 3D patient images. The proposed exploratory navigation approach is based on the local image computation without preprocessing for 3D volume data sets. A statistic analysis has been used to evaluate both the quality of the computed virtual images and the influence of the associated parameters according to clinical criteria. An indirect evaluation of the considered virtual endoscopy technique is performed: the virtual images are qualified subjectively by several observers in order to estimate a degree of stenosis. The different values of the image parameters can lead to significant differences of interpretation. We have shown the influence of the different parameters on the image interpretation and how to fix them. The identification of significant image parameters obtained by a limited number of cases allowed us to define a reliable evaluation protocol. Its application to a set of twenty patients has been used to validate the image computation process and has shown that it was possible to characterize stenosis, in an operator independent way, from CT volumes acquired under clinical routine conditions.

To improve diagnosis and therapy planning with additional information in an easy to use and fast way a virtual endoscopy system was developed. From a technical viewpoint, virtual endoscopy can be generated using image sequencies acquired with CT or MRI. It requires appropriate software for image processing and endoluminal visualization and hardware capabilities for immersive virtual reality. This includes that firstly the intuitive user interaction is supported by data gloves, position tracking systems and stereo display devices. Secondly the virtual environment requires real time visualization supported by high end graphic engines to enable the continuous operation and interaction. To enable the endoluminal view, the precise segmentation of the inner lumina like tracheobronchial tree, inner ear or vessels is necessary. In addition to this pathological findings must be defined. We use automatic segmentation techniques like volume growing as well as semiautomatic techniques like deformable models in a virtual environment. After that the surfaces of the segmented volume are reconstructed. This is the basis for our multidimensional display system which visualizes volumes, surfaces and computation results simultaneously. Our developed method of virtual endoscopy enables the interactive, immersive and endoluminal inspection of complex anatomical structures. It is based on intensive image processing like 3D-segmentation and a so called hybrid technique which displays all the information by volume and surface rendering. The system was applied on virtual bronchoscopy, colonoscopy, angioscopy as well as endoluminal representation of the inner ear.

Monitoring the effect of novel cancer chemotherapeutic agents in nude mice is now commonly done by external direct measurement and by autopsy. The development of small animal imaging has focused on micro-MRI, micro-CT and micro-PET -- each a highly expensive and highly valuable method. Far less work has been done with ultrasound imaging. We wish to demonstrate a new method of ultrasound imaging of living mice named Sono-CT^R, Sono-CT^R provides a compound image by combining the images obtained by electronically directing the transducer to scan from multiple angles.

Severely premature infants are often at high risk of cerebral hemorrhage or ischemic injury due to their inability to properly regulate blood flow to the brain. If blood flow is too high, the infant is at risk of cerebral hemorrhage, while too little blood flow can result in ischemic injury. The purpose of this research is to design and develop a means of non-invasively measuring cerebral blood flow (CBF) with near infrared spectroscopy (NIRS). Such a device would greatly aid the diagnosis and monitoring of afflicted infants. Previous attempts to measure CBF with NIRS have achieved limited success. In this study we acquired high signal-to-noise NIR spectrum from 600 to 980 nm with a cooled CCD spectrometer. This spectrometer enables the differential path length factor (DPF) to be estimated with accuracy using a second derivative technique described by Matcher et al. The validity of our new approach is determined via direct comparison with a previously validated computed tomography (CT) method. Three newborn piglets were studied. CBF measurements were performed at various partial arterial CO₂ tensions (PaCO₂) using both the NIRS and CT methods. The results of the two methods correlate well with a relationship of CBF_CT equals -4.30 + 1.05 CBF_NIRS (r² equals 0.96).

Computational fluid dynamics (CFD) models of the carotid artery are constructed from contrast-enhanced magnetic resonance angiography (MRA) using a deformable model and a surface-merging algorithm. Physiologic flow conditions are obtained from cine phase-contrast MRA at two slice locations below and above the carotid bifurcation. The methodology was tested on image data from a rigid flow-through phantom of a carotid artery with 65% degree stenosis. Predicted flow patterns are in good agreement with MR flow measurements at intermediate slice locations. Our results show that flow in a rigid flow-through phantom of the carotid bifurcation with stenosis can be simulated accurately with CFD. The methodology was then tested on flow and anatomical data from a normal human subject. The sum of the instantaneous flows measured at the internal and external carotids differs from that at the common carotid, indicating that wall compliance must be modeled. Coupled fluid-structure calculations were able to reproduce the significant dampening of the velocity waveform observed between different slices along the common carotid artery. Visualizations of the blood flow in a compliant model of the carotid bifurcation were produced. A comparison between compliant and rigid models shows significant differences in the time-dependent wall shear stress at selected locations. Our results confirm that image-based CFD techniques can be applied to the modeling of hemodynamics in compliant carotid arteries. These capabilities may eventually allow physicians to enhance current image-based diagnosis, and to predict and evaluate the outcome of interventional procedures non- invasively.

Current functional MRI techniques are essentially based on detection of hemodynamic changes induced by neuronal activation. This study is an exploration of the feasibility of direct detection of neuronal current induced NMR phase and/or magnitude changes. Our analysis was based on the approximation that neurons exist in an infinite homogeneous conducting medium, and are represented by an infinitely long cylindrical conductor, carrying uniform current. Neuronal activation was modeled by a current dipole. Simulations were performed to evaluate the effects of the local neuronal magnetic fields on the MRI signal at 3T. The magnetic field changes and the corresponding phase changes induced from a current range of 2 nA to 100 (mu) A were estimated. The conductor diameter was varied from 10 micrometer to 1 mm corresponding to the sizes ranging from that of a single axon to that of a patch of functionally similar cortex. Current induced magnetic field effects were assessed as resolution and neuronal orientation with respect to Bo were modulated. Simulation results were compared with measurements obtained from a current phantom at 3T, for a current range of 10 - 100 (mu) A. Based on the dipole model formulated we found that neuronal currents can produce magnetic fields on the order of 1.3 pT; 0.0006 degree(s) to 9 nT; 4 degree(s) depending on the neuronal bundle diameter, orientation, and current capacity. The mechanism for signal changes is that of Bo shifts. This is fundamentally similar to that of BOLD contrast, but caused by current changes rather than susceptibility changes from changes in blood oxygenation. If the bundles are of random orientations within a voxel, the estimated fields and NMR magnitude changes (decrease in T2*) are on the order of present system detection levels of approximately 1nT; 0.5 degree(s) or 0.01% signal change. Conversely, if the bundle orientation is homogeneous, then neuronal current effects are above system detection levels. The spatial scale of the current distribution also determines the net phase shift and magnitude change. The simulation and phantom experiment results demonstrated the feasibility of using MRI to directly detect local magnetic field perturbations that can result from neuronal currents on the order of a few (mu) A. This study provides a simple model for the evaluation of the feasibility to directly measure neuronal currents with MRI. Additionally it gives a starting point for the design of appropriate imaging methods towards detecting low signal level neuronal currents. A more extensive modeling of cortical and neuronal geometry, tissue inhomogeneity, timing mechanisms, and current distributions, will provide further insight in the development of MRI experimental techniques.

A new method is proposed for imaging the elasticity of tissue using magnetic resonance (MR). Using an external actuator to deform the tissue, local strains way according to local elasticity. The method produces images shows the distribution of strains with different grades of intensity; thus, revealing the underlying elasticity distribution. The advantages of the method are that imaging is much faster than other magnetic resonance elastography (MRE) techniques, there is no need for complicated computations, and it produces high-resolution images.

The lung lobes are natural units for reporting image-based measurements of the respiratory system. Lobar segmentation can also be used in pulmonary image processing to guide registration and drive additional segmentation. We have developed a 3D shape-constrained lobar segmentation technique for volumetric pulmonary CT images. The method consists of a search engine and shape constraints that work together to detect lobar fissures using gray level information and anatomic shape characteristics in two steps: (1) a coarse localization step, (2) a fine tuning step. An error detecting mechanism using shape constraints is used in our method to correct erroneous search results. Our method has been tested in four subjects, and the results are compared to manually traced results. The average RMS difference between the manual results and shape-constrained segmentation results is 2.23 mm. We further validated our method by evaluating the repeatability of lobar volumes measured from repeat scans of the same subject. We compared lobar air and tissue volume variations to show that most of the lobar volume variations are due to difference in air volume scan to scan.

Segmentation of thoracic X-Ray Computed Tomography images is a mandatory pre-processing step in many automated or semi- automated analysis tasks such us region identification, densitometric analysis, or even for 3D visualization purposes when a stack of slices has to be prepared for surface or volume rendering. In this work, we present a fully automated and fast method for pulmonary contour extraction and region identification. Our method combines adaptive intensity discrimination, geometrical feature estimation and morphological processing resulting into a fast and flexible algorithm. A complementary but not less important objective of this work consisted on a quality assessment study of the developed contour detection technique. The automatically extracted contours were statistically compared to manually drawn pulmonary outlines provided by two radiologists. Exploratory data analysis and non-parametric statistical tests were performed on the results obtained using several figures of merit. Results indicate that, besides a strong consistence among all the quality indexes, there is a wider inter-observer variability concerning both radiologists than the variability of our algorithm when compared to each one of the radiologists. As an overall conclusion we claim that the consistence and accuracy of our detection method is more than acceptable for most of the quantitative requirements mentioned by the radiologists.

The purpose was to evaluate changes of the air-tissue ratio (ATR) in previously defined regions of interest after cardiopulmonary resuscitation (CPR) in porcine model. Eight anesthetized and ventilated pigs we scanned in supine position before and 30 minutes after CPR at two different constant PEEP levels (5 cm H₂O, 15 cm H₂O). Volume scans were obtained using 6 mm slices. The gray values of the lung were divided into steps of 100 HU in order to get access to the changes of ATR. ATR was evaluated in ventral, intermediate and dorsal regions of the lung. CPR for 9 minutes led to an uneven distribution of ventilation. In the ventral region, areas with high ATR increased. Areas with normal ATR decreased. In contrast the dorsal regions with low ATR increased. ATR in the intermediate regions remained almost unchanged. Using the higher PEEP level, areas with normal ATR showed a marked increase accompanied by a decrease of areas with low ATR. After CPR, an uneven distribution of lung aeration was detected. According to the impaired hemodynamics, areas with normal ATR decreased and areas with high and low ATR increased. Using higher PEEP levels improved lung aeration.

Image-based study of structure-function relationships is a challenging problem in that the structure or region of interest may vary in position and shape on images captured over time. Such variation may be caused by the change in body posture or the motion of breathing and heart beating. Therefore, the structure or region of interest should be registered before any further regional study can be carried out. In this paper, we propose a novel approach to study the structure-function relationship of ventilation using a previously developed 3D lung warping and registration model. First, we evaluate the effectiveness of the lung warping and registration model using a set of criteria, including apparent lung motion patterns and ground truths. Then, we study the ventilation by integrating the warping model with air content calibration. The warping model is applied to three CT lung data sets, obtained under volume control of FRC, 40% and 75% vital capacity (VC). Dense displacement fields are obtained to represent deformation between different lung volume steps. For any specific region of interest, we first register it between images over time using the dense displacement, and then estimate the corresponding regional inspired air content. Assessments include change of regional volume during inspiration, change of regional air content, and the distribution of regional ventilation. This is the first time that 3D warping of lung images is applied to assess clinically significant pulmonary functions.

Suspected pulmonary embolism (PE) is a common indication for CT scanning of the thorax. Usually, intravenous contrast agent is administered utilizing a power-injector and the vascular structures are examined for the presence of pulmonary emboli. Current Multi-Slice CT-technology allows extending this morphological analysis by adding a more functional visualization of the actual parenchymal perfusion disturbance. We have developed a new image processing technique which allows selective color encoded display of parenchymal enhancement of the lung, which will be reduced in the presence of PE. Based on thin slice reconstructions an automatic 3D segmentation of the lung is performed followed by threshold based extraction of the major airways and vascular structures. This allows applying an adaptive 3D low-pass filter to the parenchymal volume only. The filtered volume data are then color encoded and overlaid onto the original CT-images. This combination of low-resolution perfusion-weighted color maps and high-resolution gray scale structural data from the same data set greatly enhances visualization of spatial relationships. The resulting images can be displayed in axial, sagittal and coronal orientation. Initial experience indicates that this new technique provides relevant additional information for the clinical management of patients with proven PE. A larger controlled patient study is under way.

Emphysema is characterized by destruction of lung tissue with development of small or large holes within the lung. These areas will have Hounsfield values (HU) approaching -1000. It is possible to detect and quantificate such areas using simple density mask technique. The edge enhancement reconstruction algorithm, gravity and motion of the heart and vessels during scanning causes artefacts, however. The purpose of our work was to construct an algorithm that detects such image artefacts and corrects them. The first step is to apply inverse filtering to the image removing much of the effect of the edge enhancement reconstruction algorithm. The next step implies computation of the antero-posterior density gradient caused by gravity and correction for that. Motion artefacts are in a third step corrected for by use of normalized averaging, thresholding and region growing. Twenty healthy volunteers were investigated, 10 with slight emphysema and 10 without. Using simple density mask technique it was not possible to separate persons with disease from those without. Our algorithm improved separation of the two groups considerably. Our algorithm needs further refinement, but may form a basis for further development of methods for computerized diagnosis and quantification of emphysema by HRCT.

This paper presents an image processing method for information extraction from 3D images of vasculature. It automates the study of vascular structures by extracting quantitative information such as skeleton, length, diameter, and vessel-to- tissue ratio for different vessels as well as their branches. Furthermore, it generates 3D visualization of vessels based on desired anatomical characteristics such as vessel diameter or 3D connectivity. Steps of the proposed approach are as follows. (1) Preprocessing, in which intensity adjustment, optimal thresholding, and median filtering are done. (2) 3D thinning, in which medial axis and skeleton of the vessels are found. (3) Branch labeling, in which different branches are identified and each voxel is assigned to the corresponding branch. (4) Quantitation, in which length of each branch is estimated, based on the number of voxels assigned to it, and its diameter is calculated using the medial axis direction. (5) Visualization, in which vascular structure is shown in 3D, using color coding and surface rendering methods. We have tested and evaluated the proposed algorithms using simulated images of multi-branch vessels and real confocal microscopic images of the vessels in rat brains. Experimental results illustrate performance of the methods and usefulness of the results for medical image analysis applications.

In the field of liver surgery 3D, visualizations of the intrahepatic vessel systems and their relationship to tumors are of major interest for the preoperative planning. To achieve a fast and robust computer assistance with optimal quantitative and visual information, we present methods for a geometrical and structural analysis of vessel systems. Based on CT and MR scans the following steps are performed: (1) The volume data is preprocessed and the vessels are segmented. (2) The skeleton of the vessels is determined. (3) The skeleton is transformed into a graph enabling a geometrical (radii, length of branches) and structural (ramifications, hierarchy) shape analysis. Using this information the different intrahepatic vessel systems are identified automatically. Interactive measuring of length, volume and diameter of vessels is supported. Also the devascularized territories of the liver due to tumor resections can be estimated. The methods have been evaluated in the clinical environment in more than 120 cases so far.

Animal models and micro-CT imaging are useful for understanding the functional consequences of, and identifying the genes involved in, the remodeling of vascular structures that accompanies pulmonary vascular disease. Using a micro-CT scanner to image contrast-enhanced arteries in excised lungs from fawn hooded rats (a strain genetically susceptible to hypoxia induced pulmonary hypertension), we found that portions of the pulmonary arterial tree downstream from a given diameter were morphometrically indistinguishable. This 'self-consistency' property provided a means for summarizing the pulmonary arterial tree architecture and mechanical properties using a parameter vector obtained from measurements of the contiguous set of vessel segments comprising the longest (principal) pathway and its branches over a range of vascular pressures. This parameter vector was used to characterize the pulmonary vascular remodeling that occurred in rats exposed to a hypoxic (11.5% oxygen) environment and provided the input to a hemodynamic model relating structure to function. The major effect of the remodeling was a longitudinally (pulmonary artery to arterioles) uniform decrease in vessel distensibility that resulted in a 90% increase in arterial resistance. Despite the almost uniform change in vessel distensibility, over 50% of the resistance increase was attributable to vessels with unstressed diameters less than 125 microns.

A realistic numerical three-dimensional (3D) model of the dynamics of human coronary arteries has been developed. High- resolution 3D images of the coronary arteries of an excised human heart were obtained using a C-arm based computed tomography (CT) system. Cine bi-plane coronary angiograms were then acquired from a patient with similar coronary anatomy. These angiograms were used to determine the vessel motion, which was applied to the static 3D coronary tree. Corresponding arterial bifurcations were identified in the 3D CT image and in the 2D angiograms. The 3D positions of the angiographic landmarks, which were known throughout the cardiac cycle, were used to warp the 3D image via a non-linear thin-plate spline algorithm. The result was a set or 30 dynamic volumetric images sampling a complete cardiac cycle. To the best of our knowledge, the model presented here is the first dynamic 3D model that provides a true representation of both the geometry and motion of a human coronary artery tree. In the future, similar models can be generated to represent different coronary anatomy and motion. Such models are expected to become an invaluable tool during the development of dynamic imaging techniques such as MRI, multi-slice CT and 3D angiography.

X-ray CT measurement of regional blood flow distribution in the lungs is potentially biased because the contrast medium used to track the flow is denser than blood. To evaluate this gravity effect, cross-sectionally uniform boluses (Reno 60, density 1.3) were delivered at the upstream end of a horizontal tube connected to a downstream axisymmetric bifurcation. When the plane of the bifurcation was vertical and actual flow through the two branches was equal, the fraction of contrast medium passing through the downward- directed branch increased with decreasing Reynolds number, increasing length-to-diameter ratio of the horizontal tube, and decreasing bolus volume. In the lungs, Reynolds number decreases and pathway length increases with decreasing vessel diameter. Thus, the results suggest that the spatial resolution of CT flow measurement within the lungs may be limited by density differences between contrast medium and blood.

A novel method for a fully automated determination of maximum blood velocity curves in Doppler ultrasound flow diagrams is reported. The method bases on VCR-recorded image sequences and hence uses an image processing scheme. Time-sequences of flow diagrams are evaluated and a chronological sequence of cardiac cycles is extracted. The cardiac cycles are numerically evaluated and the results in form of the peak velocity and the velocity-time-integral are reported.

The major goal of BSS and ICA (Independent Component Analysis) is to recover the source signals from the sensor observations under the following assumptions: (1) the source signals are statistically independent, and (2) the sensor observations are linear mixtures of source signals. Typically, BSS and ICA are based on higher-order statistics, e.g., 'parallel slices' of the fourth-order cumulant tensor (JADE algorithm) or Kurtosis (FastICA algorithm) for non-Gaussian source signals. A few techniques are based on lower-order statistics, e.g., Temporal Decorrelation SEParation (TDSEP) and Extended Spatial Decorrelation (ESD). Our approach reported in this paper is based on second-order statistics only. The spatial prototype patterns (i.e., the image-wise expanded versions of the region images) are considered as source signals and the images (or the sampled images) are considered as sensor observations. The cross-correlations between independent source signals as well as their spatially shifted versions vanish. The outcomes validate the basic requirements of BSS and ICA and lead to a simultaneous diagnolization of two symmetric correlation matrices of the observations. Then a demixing matrix is generated by standard techniques of numerical linear algebra. We have validated this method on simulated images and fMRI images. The result obtained by applying this method to simulated images indicates that the source signals (i.e., the region images) are correctly separated. The result from fMRI analysis by using this method, compared with the results obtained by using the SPM, demonstrates consistency. These results show the theoretical correctness and computational simplicity of this method.

In fMRI memory study, the temporal behavior of BOLD fMRI signals were consistently observed from various brain processing areas at 1.5 Tesla and consistent with the expected functions. Also, all the activations generally exhibit three types of temporal characteristics: short, sustained and delayed responses in relation to the primary stimuli. To address these cerebral multiphasic responses, a suitable functional data analysis scheme has been used, in which the neural response of a specific brain area to a pre-determined stimulation input of some sort was assumed to be linear. The visual memory study was performed on 6 normal subjects on a clinical MR scanner using a 5 min long rapid dynamical whole brain imaging using EPI acquisition during a single memory task, which involved a 45 sec visual presentation of three simple abstract geometric figures to the subject via LCD projector. The results showed that the activations in visual cortex were tightly correlated with the visual stimulus, while the activations detected in interior temporal, entorhinal cortex and inferior temporal area were delayed. Using the new technique, the brian activations were further characterized quantitatively in terms of delay and prolonged response. The resulting effective impulse response functions corresponding to these brain activations revealed much clearly all the temporal components.

Functional magnetic resonance imaging (fMRI) becomes a common method to study task induced brain activation. Using rapid Echo Planar Imaging (EPI) sequences one can obtain a higher MR-Signal under a task condition close by activated areas as a result of susceptibility changes in blood oxygenation (BOLD effect). Beside the commonly used blocked task designs, event- related paradigms gain more importance for activation of higher cognitive functions enabling more sophisticated and complex paradigms. For the analysis of event-related fMRI data one can use statistical tests, in example t-test used by SPM Software. The introduced analysis method based on an artificial neural network algorithm, a self-organizing map (SOM), is capable to distinguish between task related activation, deactivation and baseline patterns from the time series. This is achieved by temporal sorting and projection of all events from one condition into one combined hemodynamic response sampling for each voxel. These responses, having individual patterns can be separated by their pattern features and is done by training of the neural network. After training the SOM consists of a pattern-to-voxel mapping which is superimposed onto either an anatomical or EPI image of the subject for the task evaluation.

The structure of an fMRI time series coregistration algorithm can be divided into modules (preprocessing, minimization procedure, interpolation method, cost function), for each of which there are many different approaches. In our study we implemented some of the most recent techniques and compared their combinations with regard to both registration accuracy and runtime performance. Bidirectional inconsistency and difference image analysis served as quality measures. The result shows that with an appropriate choice of methods realignment results can be improved by far compared with standard solutions. Finally, an automatic parameter adaptation method was incorporated. Additionally, the algorithm was implemented to run on a distributed 48 processor PC cluster, surpassing the performance of conventional applications running on high end workstations.

To take advantage of MR-guided surgical procedures performed in a combined MR-OR, we have implemented and validated a practical fMRI scheme for localizing primary motor area and assessing its proximity to a lesion volume. The fMRI scheme consists of a dynamical blood oxygenation level dependent (BOLD) imaging technique and a motor task paradigm. The functional imaging was based on gradient-echo (GE) echo planar imaging (EPI) (TR/TE equals 3000/50 m sec). The task paradigm involves a periodic finger movement in which subject was instructed to tap his/her thumb on the other four fingers sequentially as well as to alternate between the right and left sides. During the task, a dynamical fMRI was performed concurrently covering the volume of interest. By using the fMRI scheme, we have successfully performed ten fMRI examinations immediately prior to surgery in the combined MR- OR on the same surgical table top to localize the eloquent functional area of interests. Also we have developed methods of detrending and presenting fMRI results to neurosurgeons in an intuitive 3-dimensional surface-rendered display format that closely matches the patient head position under intervention. Representative cases showed that fMRI results helped neurosurgeons making the optimal surgical decisions prior to craniotomy.

The promising results, recently obtained in phantom experiments employing the MR-based proton resonance frequency (PRF) method as a non-invasive tool for the temperature monitoring of hyperthermia therapy, are not easily reproduced in vivo. One of the reasons is the impact of perfusion changes on the PRF-measured temperature. In our experiments in vivo, heat was supplied on one side of the volunteers knee or pelvis by a rubber hose with circulating warm water (50_iC). The PRF method was calibrated by the constant temperature sensitivity of pure water of 0.011 ppm/_iC. MR mapping of perfusion changes was based on T2*-weighted tracking of the first-pass kinetics of contrast agent. The hemodynamic parameters of regional blood volume (rBV) and mean transit time (MTT) were extracted by fitting pixel-by-pixel the first- pass kinetics to the gamma-variate model. Special attention was directed to improve a quality of the automatic non-linear fit at low signal-to-noise values. The distributions of PRF- based temperature changes show large areas of apparently high temperature elevations (exceeding 10_iC) in regions close to the heat source, and others with just as large temperature decays in more distant regions. Areas of apparently high temperature elevations correlate with areas of blood flow increase and vice versa. In conclusion, the visible heat- induced PRF changes in vivo are primarily perfusion changes, which mask the much smaller true temperature changes.

This paper presents a multimodality approach for the localization of epileptic foci using PET, MRI and EEG combined without the need of external markers. Mutual Information algorithm is used for MRI-PET registration. Dipole coordinates (provided by BESA software) are projected onto the MRI using a specifically developed algorithm. The four anatomical references used for electrode positioning (nasion, inion and two preauricular points) are located on the MRI using a triplanar viewer combined with a surface-rendering tool. Geometric transformation using deformation of the ideal sphere used for dipole calculations is then applied to match the patient's brain size and shape. Eight treatment-refractory epileptic patients have been studied. The combination of the anatomical information from the MRI, hipoperfusion areas in PET and dipole position and orientation helped the physician in the diagnosis of epileptic focus location. Neurosurgery was not indicated for patients where PET and dipole results were inconsistent; in two cases it was clinically indicated despite the mismatch, showing a negative follow up. The multimodality approach presented does not require external markers for dipole projection onto the MRI, this being the main difference with previous methods. The proposed method may play an important role in the indication of surgery for treatment- refractory epileptic patients.

Immediate detector of any surgically induced hemorrhage prior to the closure is important for minimizing the unnecessary post surgical complications. In the case of hemorrhage, the surgical site of interests often involves hemorrhagic blood in the presence of CSF as well as air pockets. It is known that the hemorrhagic blood or air has a different magnetic susceptibility from its surrounding tissue, and CSF has long T1 and T2. Based on these differences, a set of complimentary imaging techniques (T2, FLAIR, and GE) were optimized to reveal the existence of surgically induced acute hemorrhage. Among 330 neurosurgical cases, one relatively severe hemorrhage has been successfully found intra-operatively using the concept. During the case, a new hyperintense area close to the primary motor cortex was initially noticed on T2 weighted HASTE images. As soon as it was found to increase in size rapidly, the patient was treated immediately via craniotomy for aspiration of the intra-parenchymal blood. Owing to early detection and treatment, the patient was completely free of motor deficits. Besides, there were ten much less severe hemorrhages have been noticed using the method. The proper post-surgical care was planned to closely follow-up the patient for any sign of hemorrhage.

A method for quantitative analysis of multiple sclerosis (MS) was presented. An automatic self-adaptive image segmentation algorithm was first employed to classify voxels in multi- spectral magnetic resonance (MR) images. The segmentation results from multi-spectral MR images were then combined to obtain reliable results. The volumes of brain tissues and cerebral spinal fluid (CSF) were finally extracted. Since it is fully automated, the results of the segmentation algorithm are completely reproducible. The repeatability of the presented method was evaluated on volunteer data sets. The variation is less than 0.2% for the intra-cranial volume, the whole brain volume, the central CSF, the white matter (WM) and the gray matter (GM). The variation of 3% for the entire CSF is mainly due to the peripheral CSF part, which has more partial volume effect and is less important than the central one. Methods for minimizing this variation are under investigation. These measurements demonstrate the potential for study on whole brain atrophy and cerebral atrophy. Feasibility studies on 14 MS patients were performed. The results are promising.

AD is a neurode generative disease that involves both neurochemistry and functional alterations in the CNS in addition to the morphological changes. To better characterize the disease and to assess the feasibility of achieving the early detection, a streamlined MR protocol was developed for studying the AD patient in terms of morphology, neurochemistry and brain functional activities. The protocol consists of a set of high resolution morphological imaging scans (coronal T2-weighted and FLAIR imaging acquisitions), a MR proton spectroscopy acquisition and a memory functional MRI studies. The overall study time is about 45 min, which is tolerable for AD patient. To collect the brain metabolic information, we used a PRESS sequence (TE equals 31 msec). During memory study, a 5 min long rapid dynamical whole brain imaging using single shot gradient-echo EPI acquisition was performed during a single memory task, which consisted of a 45 sec visual presentation of three geometrical figures. Six subjects from each normal and AD group were studied. The multidimensional data sets obtained contain sufficient information for volumetric measurement of hippocampus and entorhinal cortex, brain metabolite quantification, and assessment for the short- term memory function. Preliminary results suggests that the scheme can help to further classify various memory dementias into different categories.

After pre-processing and segmentation, the proposed method scores tissue regions between 1 and N. Score 1 is assigned to normal white matter and score N to CSF. Lesion zones are assigned a score based on their relative levels of similarities to white matter and CSF. To evaluate the method, 15 rats were imaged by a 7T MRI system at one of the three time points (acute, sub-acute, chronic) after MCA occlusion. Then, they were sacrificed and their brains were sliced and prepared for histological studies. MRI of 2 or 3 slices of each rat brain, using 2 DWI (b equals 400, b equals 800), 1 PDWI, 1 T2WI, and 1 T1WI, was used and an MRI score between 1 and 100 (N equals 100) was found for each region. Segmented regions were mapped onto the histology images and were scored by an experienced pathologist, from 1 to 10. MRI scores were validated using histology scores. To this end, correlation coefficients between the two scores (MRI and histology) were found. The results showed excellent correlations between MRI and histology scores at different time points.

Brain imaging methods used in experimental brain research such as Positron Emission Tomography (PET) and Functional Magnetic Resonance (fMRI) require the analysis of large amounts of data. Statistical methods are necessary to obtain a reliable measure of a given effect. Typically, researchers report their findings by listing those regions which show significant statistical activity in a group of subjects under some experimental condition or task. A number of methods create statistical parametric maps (SPMs) of the brain on a voxel- basis. However, a major limitation of the voxel-based technique is the inaccuracy of the transformation into a stereotaxic space (e.g., Talairach-Tournoux) given the wide variability in human brain structure. In order to account for this, researchers have turned to computing the statistics not on individual voxels but on predefined anatomical regions-of- interest (ROIs). A correlation coefficient is used to quantify similarity in response for various regions during an experimental setting. Since the functional inter-relationships can become rather complex, they are best understood in the context of the underlying 3-D brain anatomy. In this paper, we present a novel 3-D interface that allows the interactive exploration of the correlation datasets within a common stereotaxic atlas.

We currently evaluate MRI as add-on to dissection. Cases can only build on high evidential values of morphological findings as estimated using Bayesian likelihood-ratios. These values may vary among different cases depending on the quality of the morphology and the discrete hypotheses to be discerned. After scanning 20 bodies using MRI admitted to our institute for autopsy, we reconstructed selected imaging findings from a couple of illustrative cases according to a geometrical model ('Pink Box') designed as an object oriented bridging protocol to enable comparison of autopsy and MRI data. Although it appears obvious that 'three-dimensional imaging yields relevant diagnoses,' comparison of selected findings suggests, that the real evidential value of a postmortem scan depends on basic geometrical features of tissue structures examined. (1) Tissue surfaces are difficult to examine in MRI, including surface features of contact wounds in firearm injuries, lacerations of the pleura, or skin needle marks. (2) Specificity and sensitivity of solid tissue block data depend on contrast and resolution. (3) Tunnels or tubes, such as coronary arteries, linear wound tracks or the aorta offer more degrees of freedom for reconstruction, including spatial reconstruction or cross sectioning in different directions. (4) Three-dimensional rendering of complex objects results in spectacular images. Their evidential value is dependent on the way thresholding of 2D slices is validated. We present illustrative examples which suggest that a possible integration of non-invasive imaging methods into Forensic Pathology in fact need to take basic geometry into consideration when discussing evidential value.

Correction of PET images for partial volume effects (PVE) is of particular utility in studies of metabolism in brain aging and brain disorders. PVE is commonly corrected using voxel-by- voxel factors obtained from a high resolution brain mask (obtained from the coregistered MR scan), after convolution with the point spread function (PSF) of the imaging system. In a recently proposed regional deconvolution (RD) method, the observed regional activity is expressed as linear combinations of the true metabolic activity. The weights are obtained by integrating the PSF over the geometric extent of the brain regions. We have analyzed the accuracy of RD and two other PVE correction algorithms under a variety of conditions using simulated PET scans. Each of the brain regions was assigned a distribution of metabolic activity, with gray matter/white matter contrast representative of subjects in several age categories. Simulations were performed over a wide range of PET resolutions. The influence of PET/MR misregistration and heterogeneity of brain metabolism were also evaluated. Our results demonstrate the importance of correcting PET metabolic images for PVE. Without such correction, the regional brain activity values are contaminated with 30 - 40% errors. Under most conditions studied, the accuracy of RD and of the three- compartmental method were superior to the accuracy of the two- compartmental method. Our study provides the first demonstration of the feasibility of RD algorithm to provide accurate correction for a large number (n equals 109) of brain compartments. PVE correction methods appear to be promising tools in studies of metabolism in normal brain, brain aging, and brain disorders.

In the context of functional positron emission tomographic (PET) images analysis, the segmentation method can not only entails the separation of the image into regions of similar attribute but also presents clearer understanding about the features embedded in the original image to improve the quantitative analysis. However, for completely recording, clinical instruments often collect subject signal as well as signals from background environment, which are regarded as noises of various levels. High noise often makes the original PET image unrecognizable and difficult to analyze. Thus, manual or semiautomatic methods have been utilized to overcome the difficulty of high noise image segmentation. Furthermore, the success of image segmentation is one of the important key factors in the accompanying automated system, and there has been no general segmentation method that can be applied to the high noise PET images of different feature characteristics. However, the PET image is high noisy causing by the imaging procedure, and the image quality of PET image is affected inherently. To improve this issue, a novel nonlinear anisotropic diffusion technique based on the diffusion theorem with multi-scale and edge detection scheme to inhibit the noise level and hold the boundary characteristics of the high noise PET image was provided in this paper.

Accurate estimation of ventricular volume and motion is very important for cardiac diagnosis and treatment planning. Physicians typically calculate ventricular volume by using a few slice images and simplified equations. Such methods are generally limited by assumptions about ventricular shape particularly when the ventricle is distorted by ischemia or infarction. They also estimate ventricular motion by sequentially examining slice and phase images. However, it doesn't always give the same results. In this paper, we present an efficient double time-varying deformable model to estimate left ventricular volume and mass more accurately and to analyze endocardial and epicardial wall motions separately. At each time step, the model first finds global rigid motions and then tracks local non-rigid motions based on 3-D point sets extracted from the surface of myocardium, which are classified into endocardial or epicardial walls. The reconstructed endocardial and epicardial walls are visualized at the same time or separately and their motions are quantitatively analyzed. Results of application to gated SPECT images are given. Using the presented model, physicians can estimate the volume change over the cardiac cycle more easily and accurately, and evaluate ventricular motions reproducibly and objectively.

Echocardiography is the most convenient means for both physicians and patients for heart disease diagnosis. The 3D + 1D echocardiogram provides important information for evaluation of the 3D heart function such as the ejection fraction or wall motion. The most basic task to evaluate such functions of a heart is to segment left ventricles and reconstruct the 3D geometric model of left ventricle from a set of echocardiographic images. Since there are many images involved, the method should not need too many user assists. In this work, we design a method for reconstructing the left ventricles with very few user assists.

In our research program that aims to quantify the functional relevance of partly occluded coronary vessels, we need in one of the approaches to the problem, the 3D structure of the pertinent vessels. The use of standard biplane projection angiograms is limited by the ambiguity about the orientation not resolved by the two projections. In this paper we study to solve the orientation ambiguity based upon the geometrical unsharpness due to the focal spot of the X-ray tube. We describe the influence of the focal spot on the imaging MTF. We present the analysis of the biplane projection geometry based upon fan beam and focal spot. We derive the analytical equation of the MTF due to focal spot and geometrical magnification. We also analyze and indicate practical situations of coronaries from real angiograms.

The objective of this work is to estimate correspondence of points between central curves of coronary in successive frames. The main difficulty is related to tracking points of open curves without fixing any point. A direct application is in the estimation of coronary blood flow using cineangiographic images. Given two consecutive frames and assuming that the central curves (skeletons) of the same artery have been determined, the problem is to find a mapping of the curve points. The proposed approach is based on global minimization of a cost function that involves mapping points of a reference curve to a target one. The cost function encompasses local displacement vector and curvature. The minimization is carried out using dynamic programming and is applied on simulated data and conventional X-ray cineangiographic images. For numerical simulations of coronary skeletons we used a reference curve and several distorted curves as target curves. The transformations were: translation, rotation, shear and a combination of these distortions. The evaluation took into account the root means squared error of matching. For all above mentioned cases, the error was below 0.7 pixel. For the case of actual X-ray angiographic images, the technique presented very good qualitative results.

Pijls and De Bruyne (1993) developed a method employing intravascular blood pressure gradients to calculate the Myocardial Fractional Flow Reserve (FFR). This flow reserve is a better indication of the functional severity of a coronary stenosis than percentage diameter or luminal area reduction as provided by traditional Quantitative Coronary Angiography (QCA). However, to use this method, all of the relevant artery segments have to be select intra-operatively. After the procedure, only the segments for which a pressure reading is available can be graded. We previously introduced another way to assess the functional severity of stenosis using angiographic projections: the Relative Coronary Flow Reserve (RCFR). It is based on standard densitometric blood velocity and flow reserve methods, but without the need to estimate the geometry of the artery. This paper demonstrates that this RCFR method yields -- in theory -- the same results as the FFR, and can be given an almost identical interpretation. This provides the opportunity to use the RCFR retrospectively, when pressure gradients are not available for the segment(s) of interest.

An automated correlation method is introduced to estimate erythrocyte velocity component of erythrocyte flux within the cerebral capillary network. Erythrocyte flux, defined as the number of red blood cells passing through a plane orthogonal to the axis of erythrocyte flow in a vessel per unit time, is considered to be the closest index of capillary flow. Introduced previously is the two-point cross-correlation method, a method whereby a video photometric analyzer captures the voltage produced from two electronic windows placed over a vessel of interest. In our new method, instead of using electronic windows, we use a CCD array, focused on a two- dimensional projection of the three-dimensional capillary structure. Simulations of this method yields accurate velocity measurements at a measured cell intensity of .2 standard deviations above mean noise values or cell counts fewer than 30 cells per minute for image sequences of 180 frames captured over a time interval of three seconds. We conclude that with proper reduction in the measured standard deviation of noise and by increasing the percentage of fluorscently labeled erythrocytes injected into the rat, the correlation timing method of estimating erythrocyte velocity is an accurate substitute for hand-measured velocity calculation.

In this paper we propose a hybrid technique for detection of left ventricle motion and its segmentation from image sequence of a beating human heart. Each point on the surface of the cardiac wall moves at specific trajectory within 3D space over time. Accurate estimation of the cardiac wall motion has been shown to be very important in studying coronary diseases, such as ischemia. The proposed technique has three steps. Optical flow is computed from the sequence of images using gradient based method, then information on movement is introduced to the segmentation algorithm, and finally characteristic points along segmented boundary are detected by matching shape properties in two consecutive time frames. The estimates of optical flow for those points are used as additional constraints in the second run of the optical flow algorithm. We present experimental results obtained by the proposed algorithm and compare the optical flow fields after both runs. The image sequence of segmented LV is presented with estimates of motion at the boundaries. Experiments have shown encouraging results.

We propose an interactive electronic biopsy technique for more accurate colon cancer diagnoses by using advanced volume rendering technologies. The volume rendering technique defines a transfer function to map different ranges of sample values of the original volume data to different colors and opacities, so that the interior structure of the polyps can be clearly recognized by human eyes. Specifically, we provide a user- friendly interface for physicians to modify various parameters in the transfer function, so that the physician can interactively change the transfer function to observe the interior structures inside the abnormalities. Furthermore, to speed up the volume rendering procedure, we propose an efficient space-leaping technique by observing that the virtual camera parameters are often fixed when the physician modifies the transfer function. In addition, we provide an important tool to display the original 2D CT image at the current 3D camera position, so that the physician is able to double check the interior structure of a polyp with the density variation in the corresponding CT image for confirmation. Compared with the traditional biopsy in the procedure of optical colonoscopy, our method is more flexible, noninvasive, and therefore without risk.

Objective: To evaluate the efficacy of Ct colonoscopy for polyp detection using volumetric rendering and comparison to actual colonoscopy. Materials and Methods: 100 patients had CT colonoscopy performed by the same blinded experienced radiologist just prior to their actual colonoscopy. The number of polyps found by both techniques was compared as a function of their pathologically measured size. Complications of both procedures were assessed and ancillary findings found on noncontrast CT were also evaluated for significance. Results: 61 polyps were found by colonoscopy and resected, including 23 which were hyperplastic polyps, 35 tubular adenomas, and 3 villous adenomas. 34 were < 5 mm, 18 5 - 9 mm, and 9 were >= 10 mm in diameter. The sensitivity of CT colonoscopy for < 5 mm polyps was 11.8%, for 5 - 9 mm polyps, 61.1%, and for >= 10 mm, 100%, including the 3 villous adenomas. Evaluating by patient, of the 19 patients who one or more >= 5 mm polyps found by colonoscopy, 17 had polyps identified by CT colonoscopy. The remaining 2 patients with a normal CT colonoscopy had polyps 5 - 9 mm in size. There were no significant complications from CT colonoscopy, but 3 serious complications from actual colonoscopy. CT colonoscopy discovered significant noncolonic pathology in 10 patients. Conclusion: CT colonoscopy performed with volumetric rendering has moderate sensitivity for detecting polyps 5 - 9 mm in size and is highly sensitive for clinically significant polyps >= 10 mm in size.

CT virtual reality using volumetric rendering can tag such structures as the nasofrontal ducts, osteomeatal complexes, the middle turbinates, as well as the planned surgical sites in patients undergoing endoscopic surgery for inflammatory disease. Frequently, anatomical landmarks are obscured by overlying disease, making the endoscopic surgeon's job difficult. We have evaluated the use of CT virtual reality of the paranasal sinuses in assisting the surgeon in these types of cases. This paper reviews 25 patients with 40 sites with significant paranasal sinus disease in whom endoscopic surgery was planned. The ability of volumetric virtual reality with the various surgical sites chosen from the preoperative 2D CT's dramatically improved the accuracy of the endoscopic surgeon in localizing their surgical window. In the sphenoid sinus, the addition of CT endoscopy would have allowed the endoscopist to operate on the correct sinus an additional 28% of the time and help them miss vital structures in 25%. In the frontal sinus, CT endoscopy correctly directed the endoscopist to the correct sinus in an additional 44%. The results of this study indicate CT endoscopy may significantly improve the accuracy of endoscopic surgery into the frontal and sphenoid sinuses.

Virtual colonoscopy on powerful workstations has the distinct advantage of interactive navigation, as opposed to passive viewing of cine loops or pre-computed movies. Because of the prohibitive cost of hardware, only passive displays have been feasible for the wide-scale deployment required for mass screening. The purpose of our work is to compare low-cost commodity hardware as an effective tool for interactive colonographic navigation versus the expensive workstations.

CT virtual reality is an exciting new technique for use in the colon, airways, and paranasal sinuses. Most work in this exciting new modality, however, has used surface rendering where a single threshold value is chosen to delineate a 100% opaque wall from an air-filled lumen. We have found that there is a transition layer 3 - 4 voxels thick between the wall and the lumen corresponding to the mucosa. When this transition layer is reconstructed separately using volumetric rendering, the mucosal detail is dramatically improved. We performed a study to prove this point using 27 anatomical sites and pathological lesions evaluated by 4 blinded reviewers. Each lesion was reconstructed with surface rendering with 3 different threshold values, and with volumetric rendering with the transition zone (mucosa) reconstructed separately. In this study, volumetric rendering with the transition zone reconstructed as a separate structure overwhelmingly was rated as best (p equals 0.0001). The added mucosal detail with volumetric rendering using this approach is significant. This technique should be considered for all virtual reality applications.

We present a simple and robust approach that utilizes planar images at different angular rotations combined with unfiltered back-projection to locate the central axes of the pulmonary arterial tree. Three-dimensional points are selected interactively by the user. The computer calculates a sub- volume unfiltered back-projection orthogonal to the vector connecting the two points and centered on the first point. Because more x-rays are absorbed at the thickest portion of the vessel, in the unfiltered back-projection, the darkest pixel is assumed to be the center of the vessel. The computer replaces this point with the newly computer-calculated point. A second back-projection is calculated around the original point orthogonal to a vector connecting the newly-calculated first point and user-determined second point. The darkest pixel within the reconstruction is determined. The computer then replaces the second point with the XYZ coordinates of the darkest pixel within this second reconstruction. Following a vector based on a moving average of previously determined 3- dimensional points along the vessel's axis, the computer continues this skeletonization process until stopped by the user. The computer estimates the vessel diameter along the set of previously determined points using a method similar to the full width-half max algorithm. On all subsequent vessels, the process works the same way except that at each point, distances between the current point and all previously determined points along different vessels are determined. If the difference is less than the previously estimated diameter, the vessels are assumed to branch. This user/computer interaction continues until the vascular tree has been skeletonized.