Pathologic tissue damage frequently is associated with partial or complete loss of structural elements and is also reflected by alterations of the diffusivity of water molecules. Since single-shot echo-planar imaging (EPI) suffers from image degradation we seek to assess multi-shot EPI-based diffusion tensor imaging (DTI) for its potential to contribute to the diagnostic work-up of patients with neurologic disorders. Volunteer studies in subjects without any neurologic disorders served to document the reliability of this technique. Initial clinical applications in patients suffering from clinically definite relapsing-remitting MS were performed. Our measurements revealed high quality diffusion images of the entire brain with spatial resolution superior to that of conventional EPI scans. Critical regions, such as the brain stem or the cerebellum, were clearly visualized. The diffusion patterns observed were quite variable. It was found that certain diffusion characteristics are indicators of structural changes, and the underlying pathological phenomena may be related to pathological activity that has not been detected by other MRI techniques.

By using multiple receiver coils in conjunction with parallel MR signal reception, the speed of image acquisition can be dramatically increased. In this work the feasibility of such parallel imaging (PI) methods for fast brain imaging was studied along with their potential application for diffusion- weighted imaging (DWI). All measurements were performed by using a four-element prototype-surface coil. Parallel image reconstruction in the image domain was performed off-line on a dedicated MR image processing workstation. For appropriate image quality, coil sensitivity maps must be of sufficient quality or must be properly filtered. Thus, a novel filtering method was employed. By means of diffusion-weighted single- shot PI, the advantage of motion insensitivity of single shot echo-planar imaging (EPI) and the increased k-space velocity of multi-shot EPI can be combined; hence, they need not be phase navigated. Therefore, the images are free from ghostings, and artifacts arising from resonance offsets (e.g., B₀, susceptibility artifacts, and chemical shift) are less prominent than in conventional single-shot EPI. Furthermore, image blurring was markedly reduced. Preliminary results in neuroimaging promise that PI can become a helpful tool for rapid imaging in the CNS, although further improvement of coil sensitivity is required for sufficient SNR in parallel DWI.

Diffusion- and perfusion-weighted magnetic resonance imaging (DWI, PWI) allows the diagnosis of ischemic brain injury at a time when ischemic lesions may not yet be detectable in computer tomography or T2-weighted (T2w) MRI. However, regions with pathologic apparent diffusion coefficients (ADC) do not necessarily match with regions of prolonged mean transit times (MTT) or pathologic relative cerebral blood flow (rCBF). Mismatching parts are thought to correlate with tissues that can be saved by appropriate treatment. Ten patients with cerebral ischemia underwent standard T1w and T2w imaging as well as single-shot echo planar imaging (EPI) DWI, and PWI. Multidimensional histograms were constructed from T2w images, DWI, ADC, rCBF, and MTT maps. After segmenting different tissues, signal changes of ischemic tissues relative to unaffected parenchyma were calculated. Combining different information allowed the segmentation of lesions and unaffected tissues. Acute infarcts exhibited decreased ADC values as well as hypo- and hyperperfused areas. Correlating ADC, T2w, and rCBF with clinical symptoms allowed the estimation of age and perfusion state of the lesions. Combining DWI, PWI, and standard imaging overcomes strongly fluctuating parameters such as ADC values. A multidimensional parameter-set characterizes unaffected and pathologic tissues which may help in the evaluation of new therapeutic strategies.

A histogram-based segmentation technique was extended to exploit information acquired by manifold MRI techniques. An automated method was used to combine T₂-weighted imaging, diffusion-weighted imaging (DWI), and derived maps of the quantitative apparent diffusion coefficient (ADC). DWI allows the early detection of cerebral ischemia, and the calculated ADC value may provide information on pathophysiologic changes. Different optionally shaped clusters were characterized as separate local density maxima in the resultant 3D histogram. Cluster borders were determined by detecting density minima. Distinct but related clusters could be merged in the histogram using the Euclidian distance and a score describing the spatial neighborhood of pixels in the image. In healthy volunteers, gray matter, white matter, muscle, skin, adipose tissue, and cerebrospinal fluid were clearly identified by the automated analysis. In stroke patients, ischemic regions were reliably segmented irrespective of shape, size, and location. The time course of relative ADC changes in ischemic lesions was determined. Results were confirmed by a radiologist. The proposed automatic segmentation algorithm can be used without restrictions for the fast analysis of any multidimensional dataset. The method has proved to be reliable for determining quantities containing information on the physiologic state of tissue, such as the ADC.

3D reconstruction of airway trees can provide intuitive and useful information for understanding human lung structure and function as well as assessing the presence of disease and its response to therapy. However, due to the partial volume effects, small airways may be undetectable on some slices of lung CT images. As a result, airway tree reconstructed by the existing methods based only on segmentation results will have broken airway branches and exhibit discontinuities. This paper proposes a complete airway tree reconstruction scheme to resolve this problem, which consists of two major steps. First, the broken segments are detected by incorporating the airway tree topology with the a priori knowledge of anatomy. Topology analysis is employed to provide information related to the a priori anatomical knowledge, so that it can be applied to lung images as guidance. Then, morphology-based and linear interpolation are applied so that unconnected airway segments will gradually merge together so that a complete airway tree can be reconstructed. The process stops when all the detected discontinuities have been processed. 3D visualization is also provided to demonstrate the effectiveness of the proposed scheme.

A new technique for airway measurement that simultaneously estimates airway tilt angle, caliber, and wall thickness is presented. An idealized (circular cross-section) airway model is parameterized by airway caliber, wall thickness, and tilt angle. Using a 2D CT slice and the full-width-half-max principle we form an estimate of the inner and outer airway wall locations. We then fit ellipses to the inner and outer airway walls using a direct least squares fit and use the major and minor axes of the ellipses to estimate the tilt angle. The airway model is then initialized with tilt angle and size estimates and convolved with the 3D scanner PSF to generate a predicted image. We compare predicted versus actual images to determine the goodness-of-fit. The initial airway parameters are refined using a multi-dimensional, unconstrained, non-linear minimization routine (Nelder-Mead). When optimization converges, airway model parameters estimate the airway inner and outer radii, and tilt angle. Results using a plexiglass phantom show that tilt angle could be estimated to within plus or minus 5 degrees, the inner radius to within about one half of a pixel, and the outer radius to within one pixel.

A 3D anatomic atlas can be used to analyze pulmonary structures in CT images. To use an atlas to guide segmentation processing, the image being analyzed must be aligned and registered with the atlas. We have developed a 3D surface- based registration technique to register pulmonary CT volumes. To demonstrate the method, we have constructed an atlas from a CT image volume of a normal human male. The atlas is registered to new images in two steps: (1) a global transformation, and (2) a local elastic transformation. In the local transformation, the image and atlas volumes are divided into small subimages called cubes. The similarity between cubes in the image and atlas is measured to find the best match displacement vectors. These displacement vectors are processed using Burr's dynamic model to give a smoothed deformation vector for each voxel in the image. This method has been tested by three intra-subject registrations and three inter-subject registrations from four different normal human subjects. The results show that lung surface-based registration can register the internal lobar fissures from the atlas to the image within about 2.73 +/- 2.05 mm for intra- subject registration, and 5.96 +/- 4.99 mm for inter-subject registration. This registration can be used as an initialization for additional segmentation processing.

The purpose of this research was to use contrast enhanced spiral CT and image processing techniques to provide physiological information about the blood supply of lung lesions. This is useful for assessing solitary pulmonary nodules as well as evaluating treatment efficacy for lung cancer patients using tumor angiogenesis inhibition factors. A dynamic, contrast-enhanced spiral CT protocol was used to acquire a series of scans through the lesion both prior to contrast injection as well as at t equals 45, 90, 180 and 300 s post-injection. Lesion boundaries were then segmented for a single representative image at each time. After registration, the enhancement for each voxel was determined by subtracting the attenuation values of the pre-contrast reference voxels from the post-contrast values. Enhancement maps were then generated using these difference values. The mean and standard deviation enhancement values for each lesion were also measured. From preliminary analysis, the mean enhancement value was 100 plus or minus 24 HU for malignant lesions and 55 plus or minus 33 HU for benign lesions. Enhancement maps were generated and color-coded to demonstrate the variation of enhancement within lesions. This work presents preliminary results using contrast enhanced spiral CT imaging to detect differences between enhancing and non-enhancing lesions.

This paper presents a computerized approach to characterize pulmonary nodules as benign or malignant based on contrast enhancement patterns extracted from serial three-dimensional (3-D) thoracic CT images. In this approach the registration procedure of sequential 3-D pulmonary images consisted of the rigid transformation between two sequential region-of-interest (ROI) images including the pulmonary nodule. The normalized mutual information was used as a voxel-based similarity measure in the registration. After the motion correction between successive ROI images, the enhancement rate within a core of the segmented 3-D nodule image was estimated from the difference between the pre- and post-contrast images. We analyzed a data set of twelve 3-D thoracic CT images with pulmonary nodules in this study. Based on the Wilcoxon rank sum test, the median enhancement of the malignant lesions was significantly higher than that of the benign lesions (p less than 0.01). The preliminary results of the approach are very promising in characterizing pulmonary nodules based on quantitative measures of the contrast enhancement.

To diagnosis the lung cancer as to determine if it has malignant or benign nature, it is important to understand the spatial relationship among the abnormal nodule and other pulmonary organs. But the lung field has very complicated structure, so it is difficult to understand the connectivity of the pulmonary organs using Thin-section CT images. This method consists of two parts. The first is the classification of the pulmonary structure based on the anatomical information. The second is the quantitative analysis that is then applicable to differential diagnosis, such as differentiation of malignant or benign abnormal tissue.

To improve the care of lung-cancer patients, we are devising a diagnostic paradigm that ties together three-dimensional (3D) high-resolution computed-tomographic (CT) imaging and bronchoscopy. The system expands upon the new concept of virtual endoscopy that has seen recent application to the chest, colon, and other anatomical regions. Our approach applies computer-graphics and image-processing tools to the analysis of 3D CT chest images and complementary bronchoscopic video. It assumes a two-stage assessment of a lung-cancer patient. During Stage 1 (CT assessment), the physician interacts with a number of visual and quantitative tools to evaluate the patient's 'virtual anatomy' (3D CT scan). Automatic analysis gives navigation paths through major airways and to pre-selected suspect sites. These paths provide useful guidance during Stage-1 CT assessment. While interacting with these paths and other software tools, the user builds a multimedia Case Study, capturing telling snapshot views, movies, and quantitative data. The Case Study contains a report on the CT scan and also provides planning information for subsequent bronchoscopic evaluation. During Stage 2 (bronchoscopy), the physician uses (1) the original CT data, (2) software graphical tools, (3) the Case Study, and (4) a standard bronchoscopy suite to have an augmented vision for bronchoscopic assessment and treatment. To use the two data sources (CT and bronchoscopic video) simultaneously, they must be registered. We perform this registration using both manual interaction and an automated matching approach based on mutual information. We demonstrate our overall progress to date using human CT cases and CT-video from a bronchoscopy- training device.

Virtual bronchoscopy reconstructions of the airway noninvasively provide useful morphologic information of structural abnormalities such as stenoses and masses. To date, virtual bronchoscopy has been mainly applied to the central airways. In this paper, we show how virtual bronchoscopy can be applied to more peripheral airways by making use of the capabilities of a multi-slice helical CT scanner.

This paper proposes a method for tracking the camera motion of the real endoscope by using the virtual endoscopy system. One of the most important advantages of the virtual endoscopy is that the virtual endoscopy can visualize information of other organs that are existing under the wall of the target organ. If it is possible to track the viewpoint and the view direction of real endoscopy (fiberscope) in the examination of the patient and to display various information obtained by the virtual endoscopy onto the real endoscopic image, we construct a very useful system for assisting examination. When a sequence of real endoscopic images is inputted, tracking is performed by searching a sequence of viewpoints and view directions of virtual endoscope that correspond to camera motions of the real endoscope. First we roughly specify initial viewpoints and view directions that correspond to the first frame of the real endoscopic image. The method searches the best viewpoint and view direction by calculating matching ratio between a generated virtual endoscopic image and a real endoscopic image within the defined search area. Camera motion is also estimated by analyzing video images directly. We have applied the proposed method to video images of real bronchoscopy and X-ray CT images. The result showed that the method could track the camera motion of real endoscope.

This paper proposes a method to specify interested regions including points, lines, surfaces and mass regions through a volume rendered image,directly and interactively, and its application to virtual endoscopy system. Measurement function is one of the most important functions in the virtual endoscopy. We should specify a target region on the virtual endoscopic image for measurements. It is hard to specify target regions on the organ wall in a volume rendered image, since the organ is not explicitly segmented from an input image when observed using the volume rendering. The proposed method enables the user to specify interested regions directly by analyzing change of accumulated opacity along a casting ray. When the user specify a point on a volume rendered image, we cast a ray that passes through a specified point of an image plane from a viewpoint. We considered the position that has the highest accumulated opacity as the three-dimensional position of the specified point. Line and surface regions are obtained by iterating the point specification method. A mass region is obtained by finding the interval on the ray where the opacity is greater than zero. We have implemented those specifying methods to our virtual endoscopy system. The result showed that we could specify points, lines, surfaces and mass regions on volume rendered images.

Computed tomography (CT) based virtual cystoscopy (VC) has been studied as a potential tool for screening bladder cancer. It is accurate in localizing tumor of size larger than 1 cm and less expensive, as compared to fiberoptic cystoscopy. However, it is invasive and difficult to perform due to using Foley catheter for bladder insufflating with air. In a previous work, we investigated a magnetic resonance imaging (MRI) based VC scheme with urine as a natural contrast solution, in which a MRI acquisition protocol and an adaptive segmentation method were utilized. Both bladder lumen and wall were successfully delineated. To suppress motion artifact and insight pathological change on the bladder wall images, a multi-scan MRI scheme was presented in this study. One transverse and another coronal acquisitions of T1-weighted that cover the whole bladder were obtained twice, at one time the bladder is full of urine and at another time it is near the empty. Four bladder volumes extracted from those 4 datasets were registered first using a flexible three- dimensional (3D) registration algorithm. Then, associated 4 lumen surfaces were viewed simultaneously with the help of an interactive 3D visualization system. This MRI-based VC was tested on volunteers and demonstrated the feasibility to mass screening for bladder cancer.

In this paper, we propose a new 3D visualization technique for virtual colonoscopy. Such visualization methods could have a major impact since they have the potential for non-invasively determining the presence of polyps and other pathologies. We moreover demonstrate a method which presents a surface scan of the entire colon as a cine, and affords the viewer the opportunity to examine each point on the surface without distortion. We use the theory of conformal mappings from differential geometry in order to derive an explicit method for flattening surfaces obtained from 3D colon computerized tomography (CT) imagery. Indeed, we describe a general finite element method based on a discretization of the Laplace- Beltrami operator for flattening a surface onto the plane in an angle preserving manner. We also provide simple formulas which may be used in a real time cine to correct for distortion. We apply our method to 3D colon CT data provided to us by the Surgical Planning Laboratory of Brigham and Women's Hospital. We show how the conformal nature of the flattening function provides a flattened representation of the colon which is similar in appearance to the original. Finally, we indicate a few frames of a distortion correcting cine which can be used to examine the entire colon surface.

In our previous work, we developed a virtual colonoscopy system on a high-end 16-processor SGI Challenge with an expensive hardware graphics accelerator. The goal of this work is to port the system to a low cost PC in order to increase its availability for mass screening. Recently, Mitsubishi Electric has developed a volume-rendering PC board, called VolumePro, which includes 128 MB of RAM and vg500 rendering chip. The vg500 chip, based on Cube-4 technology, can render a 256³ volume at 30 frames per second. High image quality of volume rendering inside the colon is guaranteed by the full lighting model and 3D interpolation supported by the vg500 chip. However, the VolumePro board is lacking some features required by our interactive colon navigation. First, VolumePro currently does not support perspective projection which is paramount for interior colon navigation. Second, the patient colon data is usually much larger than 256³ and cannot be rendered in real-time. In this paper, we present our solutions to these problems, including simulated perspective projection and axis aligned boxing techniques, and demonstrate the high performance of our virtual colonoscopy system on low cost PCs.

Virtual Colonoscopy is a minimally invasive procedure to detect polyps in the colon using three dimensional tomographic imaging. An important step in analyzing the image data is accurate localization of the lumen surface. Polyps protruding into the colon lumen can then be identified by analyzing the surface either manually or using computer assisted techniques. We have developed a method for lumen segmentation based on a 3D geometric deformable model (GDM) to provide an improved representation of the lumen surface. The results show the GDM can remove the surface holes or tunnels that can be created with thresholding techniques and demonstrates the ability to accurately locate the lumen surface in high contrast air- tissue boundaries while preserving relatively lower contrast boundaries of the houstra and polyps.

Our institution has been using an internally developed system for the analysis of Computed Tomographic Colonography (CTC) since 1994. This system has gone through several major revisions during that period. Careful application of 'total quality management' (TQM) principles have been utilized to enhance such aspects of performance as patient comfort, latency between examination and reporting of results, capability and reliability of 'picture archive system' (PACS) and network components, as well as reliability of results. CTC is now being practiced at our institution for clinical screening and research applications. To date, 1500 patients have been scanned. On an average day, six patients are scanned for research and/or clinical purposes. Research patient data remain on the CTC workstation for future analysis by the Radiologists while clinical patient data are analyzed as soon as the data have been received at the CTC workstation. An enlarged dynamic axial stack augmented by multiple interactive, off axis reformatted and perspective volume rendered endoluminal views have proven to be the most effective reading mode. Well over 1000 patient scans have been analyzed utilizing this specific protocol. When compared to corresponding patient BE and/or Colonoscopy procedures, CTC findings of potential cancers and polyps have compared very favorably.

Since the introduction of Computed Tomographic Colonography (CTC) in 1995, many advances in computer equipment and software have become available. Despite these advances, the promise of colon cancer prevention has not been realized. A colorectal screening tool that performs at a high level, is acceptable to patients, and can be performed safely and at low cost holds promise of saving lives in the future. Our institution has performed over two hundred seventy five clinical CTC examinations. These scans, which each entail a supine and a prone acquisition, only differ from our research protocol in the necessity of an expeditious interpretation. Patients arrive for their CTC examination early in the morning following a period of fasting and bowel preparation. If a CTC examination has a positive finding, the patient is scheduled for colonoscopic polypectomy that same morning. To facilitate this, the patients are required to continue fasting until the CTC examination has been interpreted. It is therefore necessary to process the CTC examination very quickly to minimize patient discomfort. A positive CTC result occurred in fifteen percent of examinations. Among these positive results, the specificity has been in excess of ninety five percent. Additionally, life threatening extra-colonic lesions were discovered in two percent of the screened population.

This paper presents a new method for measuring longitudinal strain of the heart using harmonic phase magnetic resonance imaging (HARP-MRI). The heart is tagged using 1-1 SPAMM at end-diastole with tagging surfaces parallel to the imaging plane. Two image sequences are acquired for a short-axis slice with two different encodings in the direction orthogonal to the imaging plane. A method to compute a sequence of longitudinal strain estimates from this data is described.

In this study the effects of different pacing protocols on left ventricular (LV) torsion was evaluated over the full cardiac cycle. A systolic and diastolic series of Magnetic Resonance Imaging scans were combined and used to calculate the torsion of the LV. The asynchronous contraction resulting from ventricular pacing interferes with the temporal evolution of LV torsion. From these experiments we have shown that measuring torsion is an extremely sensitive indicator of the existence of ectopic excitation. The torsion of the left ventricle was investigated under three different protocols: (1) Right atrial pacing, (2) Right ventricular pacing and (3) Simultaneous pacing from the right ventricular apex and left ventricular base. The temporal evolution of torsion was determined from tagged magnetic resonance images and was evaluated over the full cardiac cycle. The peak twist T_max for the RA paced heart was 11.09 (+/- 3.54) degrees compared to 6.06 (+/- 1.65) degrees and 6.09 (+/- 0.68) degrees for the RV and Bi-V paced hearts respectively. While biventricular pacing has been shown to increase the synchrony of contraction, it does not preserve the normal physiological twist patterns of the heart.

Although several methods exist for the analysis of tagged MRI images of the left ventricle (LV), analysis of the right ventricle (RV) remains challenging due to its complex anatomy and significant through plane motion. We present here preliminary results of our new motion analysis method, both for RV and LV, in healthy human volunteers. In this method, following standard myocardial and tag segmentation of cardiac gated cine tagged MR images; a 4D B-spline based parametric motion field was computed for a volume of interest encompassing both ventricles. Using this motion field, 3D displacements and strains were calculated on the RV and LV. We observed that for both chambers the circumferential strain (Ecc) decreased with a constant rate throughout systole. The systolic strain rate displayed spatial similarity not only for the LV but also for the RV. For RV free wall, mean systolic Ecc was -0.19 +/- 0.05 with an average coefficient of variability of 20%. The 4D B-spline based motion analysis technique for tagged MRI yields compatible results for the LV and gives consistent circumferential strain measures for the RV free wall. Tagged MRI based RV mechanical analysis can be used along with LV results for a more complete cardiac evaluation.

This paper presents a near automatic process for separating vessels from background and other clutter as well as for separating arteries and veins in contrast-enhanced MR angiographic (CE-MRA) images, and an optimal method for the 3D visualization of vascular structures. The anatomic separation process utilizes fuzzy connected object delineation principles and algorithms. Its first step is the segmentation of the entire vessel structure from the background and other clutter via absolute fuzzy connectedness. Its second step is to separate artery from vein within this entire vessel structure via relative fuzzy connectedness. After 'seed' points are specified inside artery and vein in the vessel-only image, the regions of the bigger aspects of artery and vein are separated in the initial iteration. Further regions are added with subsequent iterations so that the detailed aspects of artery and vein are included in later iterations. This approach has been applied to EPIX Medical Inc's CE-MRA data. 3D images/movies of vessels, arteries, and veins have been created. Shell renditions are colored differently for arteries and veins in a composite display. This approach can produce separated artery and vein images with high quality, reliability, and minimal user interaction, in a clinical setting. All case studies were performed on a Gateway Pentium PC under Linux, and the whole procedure per study was completed in a few minutes.

Measurement of stenosis due to atherosclerosis is essential for interventional planning. Currently, measurement of stenosis from magnetic resonance angiography (MRA) is made based on 2D maximum intensity projection (MIP) images. This methodology, however, is subjective and does not take full advantage of the 3D nature of MRA. To address these limitations we present a deformable model for reconstructing the vessel surface with particular application to the carotid artery. The deformable model is based on a cylindrical coordinate system of a curvilinear axes. In this coordinate system, the location of each point on the surface of the deformable model is described by its axial, circumferential and radial position. The points on the surface deform in the radial direction so as to minimize discontinuity in radial position between adjacent points while maximizing the proximity of the surface to local edges in the image. The algorithm has no bias towards either narrower or wider cross- sectional shapes and is thus appropriate for the measurement of stenosis. Axes of the vessels are indicated manually or determined by axes detection methods. Once completed, the surface reconstruction lends itself directly to 3D methods for measuring cross-sectional diameter and area.

Functional magnet resonance imaging (fMRI) has become a standard non invasive brain imaging technique delivering high spatial resolution. Brain activation is determined by magnetic susceptibility of the blood oxygen level (BOLD effect) during an activation task, e.g. motor, auditory and visual tasks. Usually box-car paradigms have 2 - 4 rest/activation epochs with at least an overall of 50 volumes per scan in the time domain. Statistical test based analysis methods need a large amount of repetitively acquired brain volumes to gain statistical power, like Student's t-test. The introduced technique based on a self-organizing neural network (SOM) makes use of the intrinsic features of the condition change between rest and activation epoch and demonstrated to differentiate between the conditions with less time points having only one rest and one activation epoch. The method reduces scan and analysis time and the probability of possible motion artifacts from the relaxation of the patients head. Functional magnet resonance imaging (fMRI) of patients for pre-surgical evaluation and volunteers were acquired with motor (hand clenching and finger tapping), sensory (ice application), auditory (phonological and semantic word recognition task) and visual paradigms (mental rotation). For imaging we used different BOLD contrast sensitive Gradient Echo Planar Imaging (GE-EPI) single-shot pulse sequences (TR 2000 and 4000, 64 X 64 and 128 X 128, 15 - 40 slices) on a Philips Gyroscan NT 1.5 Tesla MR imager. All paradigms were RARARA (R equals rest, A equals activation) with an epoch width of 11 time points each. We used the self-organizing neural network implementation described by T. Kohonen with a 4 X 2 2D neuron map. The presented time course vectors were clustered by similar features in the 2D neuron map. Three neural networks were trained and used for labeling with the time course vectors of one, two and all three on/off epochs. The results were also compared by using a Kolmogorov-Smirnov statistical test of all 66 time points. To remove non- periodical time courses from training an auto-correlation function and bandwidth limiting Fourier filtering in combination with Gauss temporal smoothing was used. None of the trained maps, with one, two and three epochs, were significantly different which indicates that the feature space of only one on/off epoch is sufficient to differentiate between the rest and task condition. We found, that without pre-processing of the data no meaningful results can be achieved because of the huge amount of the non-activated and background voxels represents the majority of the features and is therefore learned by the SOM. Thus it is crucial to remove unnecessary capacity load of the neural network by selection of the training input, using auto-correlation function and/or Fourier spectrum analysis. However by reducing the time points to one rest and one activation epoch either strong auto- correlation or a precise periodical frequency is vanishing. Self-organizing maps can be used to separate rest and activation epochs of with only a 1/3 of the usually acquired time points. Because of the nature of the SOM technique, the pattern or feature separation, only the presence of a state change between the conditions is necessary for differentiation. Also the variance of the individual hemodynamic response function (HRF) and the variance of the spatial different regional cerebral blood flow (rCBF) is learned from the subject and not compared with a fixed model done by statistical evaluation. We found that reducing the information to only a few time points around the BOLD effect was not successful due to delays of rCBF and the insufficient extension of the BOLD feature in the time space. Especially for patient routine observation and pre-surgical planing a reduced scan time is of interest.

A linear time invariant model is applied to functional fMRI blood flow data. This model assumes that the fMRI stochastic output sequence can be determined by a constant plus a linear filter (hemodynamic response function) of several fixed deterministic inputs and an error term, assumed stationary with zero mean and error spectrum. An on-off finger tapping experiment was performed where the subject repetitively tapped their fingers for 30 seconds and remained still for 30 seconds. Thirty three disjoint frequency bands, 3 wave numbers wide were chosen to analyze the data. At each band an F- statistical image was constructed to test ((alpha) equals .05/33) whether power from the input signal induced a response in the output signal. Activation was seen at frequency .0154 Hz close to the frequency for maximum power of the input signal, .0167 Hz in the contralateral motor strip and motor cortex. In conclusion, (1) No assumptions are made about the filter. (2) Several different deterministic inputs may be applied. (3) Problems with temporal correlation are avoided by performing the statistics in the Fourier domain. (4) Testing can be performed for differences in the hemodynamic transfer function at different spatial locations under different experimental conditions.

The aim of this work was to use dynamic contrast enhanced MR image (DEMRI) data to generate 'parameter images' which provide functional information about contrast agent access, in bone sarcoma. A simulated annealing based technique was applied to optimize the parameters of a pharmacokinetic model used to describe the kinetics of the tissue response during and after intravenous infusion of a paramagnetic contrast medium, Gd-DTPA. Optimization was performed on a pixel by pixel basis so as to minimize the sum of square deviations of the calculated values from the values obtained experimentally during dynamic contrast enhanced MR imaging. A cost function based on a priori information was introduced during the annealing procedure to ensure that the values obtained were within the expected ranges. The optimized parameters were used in the model to generate parameter images, which reveal functional information that is normally not visible in conventional Gd-DTPA enhanced MR images. This functional information, during and upon completion of pre-operative chemotherapy, is useful in predicting the probability of disease free survival.

Functional MRI (fMRI) is a non-invasive technique widely used to map brain-functions. Nevertheless, its hemodynamic basis and spatial precision with which fMRI reflects sites of neuronal activity are not completely understood. We therefore combined fMRI, based on the blood oxygenation level dependent (BOLD) effect, with optical recording of intrinsic signals (ORIS), a technique, which has a better spatial and temporal resolution. Furthermore, ORIS can distinguish between localized changes in deoxyhemoglobin, and more widespread changes in cerebral blood volume/flow. In gerbils hemodynamic responses over the contralateral barrel cortex were studied with both methods, using identical stimulation of a single vibrissae and identical integration and correlation analysis strategies. Analysis of integration maps and of the spatial distribution and temporal correlation with the block-design of vibrissal stimulation revealed that the BOLD signal, at the site of neuronal activation, does not reflect a depletion of deoxyhemoglobin, as generally assumed. Instead, its positive polarity is likely due to an increase in cerebral blood volume (CBV) whose highly dynamic effect on the BOLD signal exceeds that of the increase in deoxyhemoglobin remaining elevated during prolonged stimulation. This is so, because we show, that blood flow does wash out deoxyhemoglobin but at a rate which is to decrease the deoxyhemoglobin concentration in the voxel below resting level. The wash out causes an accumulation of deoxyhemoglobin in the draining venous side, but at a time window which can be clearly distinguished from the specific activity by applying an analysis strategy based on correlation functions. Therefore, draining veins do not appear as confounding problem. This knowledge could be useful to model the BOLD effect more accurately and improve the spatial resolution of fMRI.

fMRI has provided a new option to study cognitive phenomena. Recent developments in medical image processing and analysis allow researchers to study more elaborate cognitive tasks from a wide perspective. These techniques include Statistical Parametric Mapping, Subspace Modeling and Maximum Likelihood Estimation, and Spatio-temporal Analysis using Random Fields. Their common weakness is the assumption of the statistical independence among the image pixels. We have developed a multivariate segmentation method to functional MRI analysis for human brain function study based on the second-order statistics of images. It consists of four steps: (1) detecting the number of the distinctive image regions, (2) generating the scores and determining their rank, (3) forming score plots and clustering in the feature space, (4) projecting clusters from the feature space to the image space to generate object images. We have validated this method on the simulated and fMRI images. The theoretical and experimental results obtained by using this method were in good agreement. The relations between this method and other multivariate image analysis methods are discussed.

The composition of atherosclerotic lesions in the carotid arteries is believed to be an important predictor of stroke risk. Several MR contrasts may be necessary to discriminate between different plaque components, and multispectral analysis can used to integrate the information obtained from these multiple contrasts. This study presents the use of registered MR and histological images of carotid endarterectomy specimens as a tool for the quantitative assessment of maximum likelihood classification and other segmentation algorithms. Carotid endarterectomy specimens were imaged in a 1.5T GE Signa scanner. PD, T1, T2, diffusion spin echo weightings were obtained. MR images were registered with digitized images of the corresponding histology. A pathologist identified regions of collagen, calcification, cholesterol, hemorrhage on the histological images. Training and ground truth regions were selected. The accuracy of the maximum likelihood classification was assessed on a pixel by pixel basis using truth regions identified on histological images. The accuracy of multispectral analysis was calcification (73%), fibrin (68%), cholesterol (62%), fibrous plaque (53%). This technique was limited by registration inaccuracies caused by partial volume effects and histological artifacts. Despite these limitations, accuracy results were reasonable. This technique, with continued improvements, provides a framework for evaluating the accuracy of different segmentation algorithms.

Average myocardial perfusion is remarkably consistent throughout the heart wall under resting conditions and the velocity of blood flow is fairly reproducible from artery to artery. Based on these observations, and the fact that flow through an artery is the product of arterial cross-sectional area and blood flow velocity, we would expect the volume of myocardium perfused to be proportional to the cross-sectional area of the coronary artery perfusing that volume of myocardium. This relationship has been confirmed by others in pigs, dogs and humans. To test the body size-dependence of this relationship we used the hearts from rats, 3 through 25 weeks of age. The coronary arteries were infused with radiopaque microfil polymer and the hearts scanned in a micro- CT scanner. Using these 3D images we measured the volume of myocardium and the arterial cross-sectional area of the artery that perfused that volume of myocardium. The average constant of proportionality was found to be 0.15 +/- 0.08 cm³/mm². Our data showed no statistically different estimates of the constant of proportionality in the rat hearts of different ages nor between the left and right coronary arteries. This constant is smaller than that observed in large animals and humans, but this difference is consistent with the body mass-dependence on metabolic rate.

We developed methods to quantify arterial structural and mechanical properties in excised rat lungs and applied them to investigate the distensibility decrease accompanying chronic hypoxia-induced pulmonary hypertension. Lungs of control and hypertensive (three weeks 11% O₂) animals were excised and a contrast agent introduced before micro-CT imaging with a special purpose scanner. For each lung, four 3D image data sets were obtained, each at a different intra-arterial contrast agent pressure. Vessel segment diameters and lengths were measured at all levels in the arterial tree hierarchy, and these data used to generate features sensitive to distensibility changes. Results indicate that measurements obtained from 3D micro-CT images can be used to quantify vessel biomechanical properties in this rat model of pulmonary hypertension and that distensibility is reduced by exposure to chronic hypoxia. Mechanical properties can be assessed in a localized fashion and quantified in a spatially-resolved way or as a single parameter describing the tree as a whole. Micro-CT is a nondestructive way to rapidly assess structural and mechanical properties of arteries in small animal organs maintained in a physiological state. Quantitative features measured by this method may provide valuable insights into the mechanisms causing the elevated pressures in pulmonary hypertension of differing etiologies and should become increasingly valuable tools in the study of complex phenotypes in small-animal models of important diseases such as hypertension.

In victims of electrical trauma, electroporation of cell membrane, in which lipid bilayer is permeabilized by thermal and electrical forces, is thought to be a substantial cause of tissue damage. It has been suggested that certain mild surfactant in low concentration could induce sealing of permeabilized lipid bilayers, thus repairing cell membranes that had not been extensively damaged. With an animal model of electrically injured hind limb of rats, we have demonstrated and validated the use of radiotracer imaging technique to assess the physiology of the damaged tissues after electrical shock and of their repairs after applying surfactant as a therapeutic strategy. For example, using Tc-99m labeled pyrophosphate (PYP), which follows calcium in cellular function and is known to accumulate in damaged tissues, we have established a physiological imaging approach for assessment of the extent of tissue injury for diagnosis and surgical planning, as well as for evaluation of responses to therapy. With the use of a small, hand-held, miniature gamma camera, this physiological imaging method can be employed at patient's bedside and even in the field, for example, at accident site or during transfer for emergency care, rapid diagnosis, and prompt treatment in order to maximize the chance for tissue survival.

Contrast enhanced MRI (CE-MRI) is increasingly being used for tumor characterization as well as therapy response monitoring. We are investigating the diagnostic value of fast, multi-slice [1.5X1.5X5(mm)] CE-MRI with a time interval of 2(s) to enables estimating perfusion parameters. A clinical visualization tool is being developed that estimates curve parameters on a per voxel basis to be able to analyze the heterogeneous tumor physiology. MRI image noise strongly affects the (perfusion) related start-of-enhancement and time- to-peak estimates especially in low contrast curves. We therefore developed a robust estimation method. It was tested on realistic simulations and in vivo CE-MRI data and compared to existing methods. The new method is shown to be more accurate, precise and robust than common least-squares optimization based methods.

Factor analysis of medical image sequences (FAMIS), in which one concerns the problem of simultaneous identification of homogeneous regions (factor images) and the characteristic temporal variations (factors) inside these regions from a temporal sequence of images by statistical analysis, is one of the major challenges in medical imaging. In this research, we contribute to this important area of research by proposing a two-step approach. First, we study the use of the noise- adjusted principal component (NAPC) analysis developed by Lee et. al. for identifying the characteristic temporal variations in dynamic scans acquired by PET and MRI. NAPC allows us to effectively reject data noise and substantially reduce data dimension based on signal-to-noise ratio consideration. Subsequently, a simple spatial analysis based on the criteria of minimum spatial overlapping and non-negativity of the factor images is applied for extraction of the factors and factor images. In our simulation study, our preliminary results indicate that the proposed approach can accurately identify the factor images. However, the factors are not completely separated.

Since the positron emission tomographic (PET) image shows the abnormal activity of brain by the different parts of the brain respond to different stimuli and the patient's response to noise, illumination, change in mental concentration, and other activity, the functional PET image data is much helpful for clinical diagnosis. However, the PET image is also a high noise image that the quality of the PET image and the diagnosis accuracy is affected by the noise. To improve the quality problem of PET image, a novel subband denoising technique is provided in this paper. The method is based on the subband transformation and the statistical features in each subbands of the PET image.

Representative results from simulated, laboratory and physiological studies are presented, demonstrating the ability to extract important features of dynamic behavior from dense scattering media. These results were obtained by analyzing a time series of image data. Investigations on the human forearm clearly reveal the ability to identify and correctly locate principal features of the vasculature. Characterization of these features using linear and nonlinear time-series analysis methods can produce a wealth of information regarding the spatio-temporal features of the dynamics of vascular reactivity.

Intrinsic and exogenous fluorescent molecules may be used as specific markers of disease processes, or metabolic status. A variety of fluorescent markers have been successfully used for transparent tissue, in-vitro studies, and in cases where the markers are located close to the tissue surface. For example, given fluorescence lifetime measurements of a fluorophore such as bis(carboxylic acid) dye, the known relationship of pH on its lifetime may be used to determine the pH of tissue at the fluorophore's location. For fluorophore depths greater than approximately one millimeter in normal tissue, such as might be encountered in in vivo studies, multiple scattering makes it impossible to make direct measurements of characteristics such as fluorophore lifetime. In a multiple scattering environment, the collected intensity depends heavily on the scattering and absorption coefficients of the tissue at both the excitation and emission frequencies. Thus, to obtain values for specific fluorophore characteristics such as the lifetime, a theoretical description of the complex photon paths is required. We have applied Random-walk theory to successfully model photon migration in turbid medias such as tissue. We show how time-resolve intensity measurements may be used to determine fluorophore location and lifetime even when the fluorophore site is located many mean photon scattering lengths from the emitter and detector.

The purpose was to evaluate differences in dynamic changes of the lung aeration (air-tissue ratio) between augmented modes of ventilation (AMV) and controlled mechanical ventilation (CMV) in normal subjects. 4 volunteers, ventilated with the different respirator protocols via face mask, were scanned using the EBCT in the 50 ms mode. A software analyzed the respirator's digitized pressure and volume signals of two subsequent ventilation phases. Using these values it was possible to calculate the onset of inspiration or expiration of the next respiratory phase. The calculated starting point was then used to trigger the EBCT. The dynamic changes of air- tissue ratios were evaluated in three separate regions: a ventral, an intermediate and a dorsal area. AMV results in increase of air-tissue ratio in the dorsal lung area due to the active contraction of the diaphragm, whereas CMV results in a more pronounced increase in air-tissue ratio of the ventral lung area. This study gives further insight into the dynamic changes of the lung's biomechanics by comparing augmented ventilation and controlled mechanical ventilation in the healthy proband.

The measurement of blood flow is important to understanding the physiological effects of heating on tissue. The primary effect of thermal therapies on solid tumors is the collapse of the blood vessels supplying the tumor, resulting in a thermal lesion due to the cessation of blood flow. We investigated the effect of heating on VX2 tumors implanted in the rabbit thigh, over the course of a one-hour treatment. A method of measuring the blood flow over an entire tissue slice using dynamic contrast-enhanced CT was developed. The distribution of the blood flow values was displayed as a single image in which a spectrum of pseudo-colors was used to encode blood flow values. This blood flow map provides both a visual and quantitative means of assessing blood flow changes and hence the tissue damage over time. From the blood flow maps we defined thermal lesions as tissue regions which had blood flow in the range of 0 - 2 ml/min/100g. Using this definition, the ratio of the thermal lesion area from pre-treatment to 60 minutes post-treatment, increased by a factor of 6, whereas the same ratios for the normal and viable tumor tissue remained essentially constant.

Our aim is to derive quantitative measurements from Myocardium Perfusion Scintigraphic (MPS) exams for drug trials on Myocardium re-perfusion. We are considering 19 patients imaged 6 times to measure the effects of various conditions of SPECT acquisition (Sestamibi stress and Thallium rest). We are also measuring the stability of various indices of perfusion evolution. Our method is based on intra-patients image matching techniques for follow-up and on inter-patients matching with a reference model based on 100 normal subjects to define perfusion abnormalities. We are measuring intensity differences between normalized images of 10% for Sestamibi and 14% for Thallium. Correlation between image acquisitions is 95% for Sestamibi and 88% for Thallium. Our most stable index is the deficit load, being the integral over stress defects of perfusion deficit. For our 19 cases, deficit load average is 8% of global normal perfusion (GNP), standard deviation between 2 acquisitions is about 0.5% GNP with a -0.4% GNP systematic bias, and correlation between 2 acquisitions is 99.8%. The stability of the index is demonstrated and we expect that a deficit load variation of more than 2% GNP is significant of an evolution, which has to be confirmed by ongoing retrospective drug trials.

We developed a novel clinical tool (PERFIT) for automated 3-D voxel-based quantification of myocardial perfusion, validated it with a wide spectrum of angiographically correlated cases, compared it to previous approaches, and tested its agreement with visual expert reading. A multistage, 3-D iterative inter- subject registration of patient images to normal stress and rest cardiac templates was applied, including automated masking of external activity before final fit. The reference templates were adjusted to the individual left ventricles by template erosion, for further shape correction. 125 angiographically correlated cases including multi-vessel disease, infarction, and dilated ventricles were tested. In addition, standard polar maps were generated automatically from the registered data. Results of consensus visual reading (V) and PERFIT (P) were compared. The iterative fitting was successful in 245/250 (99%) stress and rest images. PERFIT found defects on stress in 2/29 normal patients and 95/96 abnormal patients. Overall correlation between V and P findings was r equals 0.864. In all abnormal groups (n equals 96), PERFIT average defect sizes expressed as the percentage the myocardial volume were 9.6% for rest and 22.3% for stress, versus 11.4% (rest) and 23% (stress) for visual reading. Automatic quantification by PERFIT is consistent with visual analysis; it can be applied to the analysis whole spectrum of clinical images, and can aid physicians in interpretation of myocardial perfusion.

Our purpose is to assess the pulse flow propagation from series of digital coronary angiograms. The local dilation- contraction pattern along the vessel is a measure for the elasticity and endothelian function. A small distensibility could be an indication for the presence of atherosclerosis also in cases where the angiogram is not abnormal. We have developed an analytical model of the pulse flow propagation in coronary arteries. In the model the artery is a straight elastic tube and is not moving due to the motion of the heart. The pulsatile flow of contrast agent is modeled for clinically relevant parameters and predictions of coronary angiograms are obtained for various characteristics of elasticity such as modulus, compliance and ratio. In the clinical angiograms, we compute the local vessel diameter from frame to frame. Because of the heart motion it is not so easy to track the vessel diameter on the same spot. Motion estimation and compensation are required. Algorithms for these processing steps are implemented. We have obtained satisfactory model simulations and predictions of angiograms. The simulated-dilation- contraction patterns help to understand the more complicated clinical angiograms. We have obtained various pulse flow patterns from coronary angiograms of a small patient population.

Synchronization of contrast administration and CT imaging can maximize the signal differences between arteries and background in first pass studies. In this paper, a bolus propagation model is developed for tomographic angiography, especially CT angiography that relies on bolus peak prediction, real-time CT observation and adaptive table transport. The bolus propagation model is configured as a network of vessels and organs, each of which is described based on the vascular transport operator theory. The traditional vascular operator is augmented by introducing a longitudinal length parameter to depict the bolus propagation along a vessel. A parameter adjustment algorithm is also designed for the extended vascular operator, assuming real- time measurements of concentration-time curves are available at multiple locations along the vessel. As a result, the bolus concentration can be computed with respective to both the time since bolus injection and the location along the vessel. Numerical simulation and patient studies with CT and MRI are performed to evaluate the feasibility and utility of the bolus propagation model. Individualization of the bolus propagation model are tested. Theoretical and practical values are in excellent agreement. This bolus propagation model has a significant potential for optimization of tomographic angiography. In particular, this model can be applied to CT angiography so that the intravenous bolus peak and the X-ray imaging aperture are matched with an adaptive table transport.

In cardiac surgeries it is frequently necessary to carry out interventions in internal heart structures, and where the blood circulation and oxygenation are made by artificial ways, out of the patient's body, in a procedure known as extracorporeal circulation (EC). During this procedure, one of the most important parameters, and that demands constant monitoring, is the blood flow. In this work, an ultrasonic pulsed Doppler blood flowmeter, to be used in an extracorporeal circulation system, was developed. It was used a 2 MHz ultrasonic transducer, measuring flows from 0 to 5 liters/min, coupled externally to the EC arterial line destined to adults perfusion (diameter of 9.53 mm). The experimental results using the developed flowmeter indicated a maximum deviation of 3.5% of full scale, while the blood flow estimator based in the rotation speed of the peristaltic pump presented deviations greater than 20% of full scale. This ultrasonic flowmeter supplies the results in a continuous and trustworthy way, and it does not present the limitations found in those flowmeters based in other transduction methods. Moreover, due to the fact of not being in contact with the blood, it is not disposable and it does not need sterilization, reducing operational costs and facilitating its use.

Typically, indicator dilution theory is applied to time- density curves acquired from dynamic contrast images, including x-ray planar and fast CT, for determination of blood circulation parameters. The original theory developed by Zierler and now applied to image curves assumes that the time- density curves are flow-sampled, i.e. each particle of indicator is counted for a time proportional to its velocity so that particles traveling within streamlines with higher or lower flow contribute equally to the measured concentration. However, curves obtained from images are instead cross- sectionally sampled, i.e, each indicator particle is counted for a time proportional to its residence time within the ROI, (inversely proportional to its velocity), so that particles traveling within low flow streamlines contribute relatively more to the measured concentration than do particles within higher flow streamlines. We illustrate some of the potential pitfalls encountered when applying the conventional theory directly to image curves and propose a strategy for accounting for the cross-sectional error in the transit time estimate. The correction method was applied to a known simulated network system for verification and illustration of its usefulness. Finally, to illustrate its applicability, the method was implemented on experimental image curves obtained from a microfocal x-ray imaging system.

In the rat, almost 20% of the total body heat-loss occurs by sympathetically mediated increases in blood flow through a system of arteriovenous anastomoses (AVAs) in the skin of the tail which are absent at the base and abundant at the tip. To study the mechanisms of thermoregulation in the rat tail we monitored online the blood vessel temperature and the arterial and venous vessel size and their mutual vascular volume interactions using in vivo MRA. During a gradual rise in rectal temperature from 36 degrees Celsius to 40 degrees Celsius, tail surface temperatures were measured at ventral (T_a) and lateral (T_v) sits overlying the respective vascular bundles. At the base, middle and tip, diameter of the ventral artery and the lateral veins of the heat-loaded animal increased clearly upon rising body temperature. Calculation of (T_a - T_v) in function of the rectal temperature during heating showed that at the tail base (T_a - T_v) was maximum at rectal temperature of 38 degrees Celsius and minimum at 39 degrees Celsius. At the middle and the tip of the tail, a steady rise of (T_a - T_v) was observed. If we assume that vasodilatation is a synchronical process along the length of the tail, then the difference in (T_a - T_v) is due to the presence of AVAs.

Stroke models, if used in drug evaluation studies, should have a predictable and reproducible course and outcome. While most drug trials focus on the lesion outcome, our study shows the importance of studying lesion growth instead of lesion outcome. In the study reported here, the time course of a photochemically induced neocortical infarct is studied in rats, using diffusion-weighted magnetic resonance imaging, while the rats were submitted to a rigorous control of physiological parameters, ensuring constant body temperature, blood gases (pO₂ and pCO₂), arterial pressure, heart rate and plasma glucose levels. Under such a stable physiological condition, rats were imaged as soon as possible after lesion up to 6 hours, which is the most important period to determine the slope of further lesion growth and final outcome. The data show that the initial size of the lesion is important for the further outcome of the stroke, both in lesion size and severity of the ischemic damage, as reflected by changes in the Apparent Diffusion Coefficient.

Simultaneously acquired EEG and BOLD (Blood Oxygenation level dependent contrast) MRI allowed to study on line the neurophysiological changes in rat brain during epileptic seizures. MRI and EEG data were acquired with a specially designed high quality MR RF-antenna with incorporated non- invasive carbon EEG electrodes. The problem of severe pollution of the EEG data due to MR gradient switching during simultaneous EEG/MRI acquisitions was solved by a specially designed automated effective filtering algorithm. We measured continuously EEG data, and T₂*-weighted coronal MRI sections of rat brain before and after the injection of pentetrazol (43 mg/(kg body weight) PTZ; convulsive dose 97%), an epilepsy inductor. In this way, we could correlate the abnormalities in the EEG traces, with changes in the MRI BOLD signal intensities. Immediately after PTZ induction and before epileptic discharges were observed on the EEG traces, the cortex displayed an increase in BOLD signal intensity (increase in blood flow). Much later and correlated with epileptic discharges on the EEG traces, the ventromedial hypothalamic nuclei showed an increased BOLD signal while the BOLD signal intensity dropped in the entire brain, except for the hypothalamus. The decreased BOLD signal reflected general hypoxia and subsequent ischemia as a consequence of the sustained depolarization of neurons during the seizure.

Heart valve insufficiencies can optimally be assessed using transesophageal, triggered, three-dimensional ultrasound imaging. The dynamic ultrasound data contain morphological as well as functional components which are recorded and displayed simultaneously. It allows the visualization of intracardiac motion which is an important parameter to detect abnormal flow caused by defect valves. A realtime reconstruction is desired to get a spatial impression on the one hand and to interactively clip parts of the volume on the other hand. OpenGL Volumizer is used for visualization. Scalability of the visualization was tested with respect to different workstations and graphics resources using a Multipipe Utility library (MPU). The combination of both APIs enables a visualization of volumetric and functional data with frame rates up to 10 frames per second. By using the proposed method, it is possible to visualize the jet in the original color-coding which is employed during a conventional two- dimensional examination for displaying the velocity values. A good scalability from low cost up to high end graphic workstations is given by the use of the MPU. The quality of the resulting 3D images allows exact differentiation of heart valve insufficiencies to support the diagnostic procedure.

Cardiac imaging with conventional computed tomography (CT) has gained significant attention in recent years. New hardware development enables a CT scanner to rotate at a faster speed so that less cardiac motion is present in acquired projection data. Many new tomographic reconstruction techniques have also been developed to reduce the artifacts induced by the cardiac motion. Most of the algorithms make use of the projection data collected over several cardiac cycles to formulate a single projection data set. Because the data set is formed with samples collected roughly in the same phase of a cardiac cycle, the temporal resolution of the newly formed data set is significantly improved compared with projections collected continuously. In this paper, we present an adaptive phase- coded reconstruction scheme (APR) for cardiac CT. Unlike the previously proposed schemes where the projection sector size is identical, APR determines each sector size based on the tomographic reconstruction algorithm. The newly proposed scheme ensures that the temporal resolution of each sector is substantially equal. In addition, the scan speed is selected based on the measured EKG signal of the patient.

99m-Tc-pertechnetate SPECT studies of thyroid phantoms, with volumes in 13-198 cc range, and of a number of patients were performed. A modified reconstruction and analytic methods were applied to the data analysis. It involved: (1) transformation of the initial ECT data by a suitable mathematical function to compress the dynamical range of the data; (2) tomographic reconstruction resulting in transaxial (TV) images; (3) decompression of TV images via a reverse function; (4) Gaussian smoothing of the decompressed TV (dTV) images; (5) subtraction of a Compton dTV images from photopeak dTV images, (6) parallel operation on dTV images: Sobel edge detection and impulse filtering; (7) combining the filtered images via the AND operator; (8) edge tracing in the combined image by a human operator; (9) volume estimation by a computer. Based on the phantom studies it was established that the proposed technique yielded volumes with the relative error not exceeding 10%. For patient studies, the obtained volumes were also compared with palpation estimation. It has been found that inter-observer variability of the thyroid volume estimate is a function of physician's experience in utilization of manual palpation and it may result in the relative errors exceeding 50%.

Past measurements of arterial branching geometry have indicated that the branching geometry is somewhat consistent with an optimal trade-off between the work needed to build and maintain the arterial tree and the work needed to operate the tree as a transport system. The branching geometry is also consistent with the mechanism that acutely adjusts the lumen diameter by way of maintaining a constant shear stress by dilating (or constricting) the arteries via the nitric oxide mechanism. However, those observations also indicate that there is considerable variation about the predicted optimization, both within any one individual and between individuals. Possible causes for this variation include: (1) measurement noise -- both due to the imprecision of the method but also the preparation of the specimen for applying the measurement technique, (2) the fact that the measurement task presents a major logistic problem, which increases as the vessel size decreases (but the number of branches correspondingly doubles at each branching) and results in progressive under-sampling as the vessel size decreases, (3) because of the logistic task involved the number of arterial trees analyzed is also greatly limited, and (4) there may indeed be actual heterogeneity in the geometry which is due to slight variation in implementation of the 'rules' used to construct a vascular tree. Indeed, it is this latter possibility that is of considerable physiological interest as it could result in the observed heterogeneity of organ perfusion and also provide some insight into the relative importance of 'initial ' conditions (i.e., how the vascular tree initially develops during embryogenesis) and the adaptive mechanisms operative in the maturing individual. The use of micro-CT imaging to provide 3D images of the intact vascular tree within the intact organ overcomes or minimizes the logistic problems listed above. It is the purpose of this study to examine whether variability in the branching geometry is constant over the length of an artery or whether this progressively amplifies along the length of the artery.

In this paper, we propose a new method for left ventricle (LV) motion estimation. The method is based on a point-constrained optical flow algorithm for heart motion estimation. The constrained points are the corresponding point pairs in subsequent images. The motion estimation of these characteristic points can be obtained using techniques such as shape-based tracking of perceptually salient points on the LV boundary. Knowing the motion of characteristic points the proposed algorithm is used to estimate motion at other image points. Using this approach the result of the optical flow algorithm is constrained by displacement of characteristic LV boundary points. The constraints are implemented by setting the values of the optical flow of characteristic points to certain fixed values. The spatial and temporal continuity functionals are specified so that the fixed values influence their neighborhoods and the optical flow field is interpolated between the constrained points. The experiments have been performed using ECG-gated MR images of the beating heart. The method has been applied to the image sequence and estimation of the motion field has been computed.

The dynamic image recognition is applied to the robot engineering sciences. This technology (the optical flow method) was used the computer aided diagnosis of the esophageal fluoroscopy. The dynamic images were obtained by means of using digital video recorder. The motion vectors are calculated using the computer vision algorithm (the gradient based method) from these dynamic images in each pixels. The motion vectors were visualized in color images and allow images. The results showed that the motion of normal esophageal walls were manly in the horizontal direction, whereas a large oblique direction component was observed in that of abnormal esophageal walls. The dynamic image recognition was useful in clinical study for the analysis of esophageal wall motion.