An important application of functional imaging is the estimation of regional blood flow and volume using residue detection of vascular indicators. An indicator-dilution model applicable to tissue regions distal from the inlet site was developed. Theoretical methods for determining regional blood flow, volume, and mean transit time parameters from time-absorbance curves arise from this model. The robustness of the parameter estimation methods was evaluated using a computer-simulated vessel network model. Flow through arterioles, networks of capillaries, and venules was simulated. Parameter identification and practical implementation issues were addressed. The shape of the inlet concentration curve and moderate amounts of random noise did not effect the ability of the method to recover accurate parameter estimates. The parameter estimates degraded in the presence of significant dispersion of the measured inlet concentration curve as it traveled through arteries upstream from the microvascular region. The methods were applied to image data obtained using microfocal x-ray angiography to study the pulmonary microcirculation. Time- absorbance curves were acquired from a small feeding artery, the surrounding microvasculature and a draining vein of an isolated dog lung as contrast material passed through the field-of-view. Changes in regional microvascular volume were determined from these curves.

Electron beam computed tomography (EBCT) is a potentially useful modality to quantitate regional pulmonary flow (RPF) with minimal invasiveness, in part because it has good spatial and temporal resolution. The present studies used a single compartment model of indicator transport and EBCT to measure regional tissue flow in the lungs of human subjects. The model postulates that flow is proportional to maximal enhancement and assumes complete tissue accumulation of indicator before significant indicator washout (WO). EBCT flow studies were retrospectively analyzed with respect to RPF in 10 adult patients who had undergone clinically indicated or research cardiovascular studies. Time density curves from the left atrial (LA) cavity and one-third segments of left (LL) and right (RL) lungs (A: anterior, M: middle, and P: posterior segments) were used to calculate RPF. Washout was determined as the percent of the LA curve at the time of peak parenchymal opacification using gamma curve fits to both tissue data and the LA curve data. Mean +/- standard deviation RPF in ml/min/ml was 0.8 +/- 0.4, 1.1 +/- 0.4, and 1.3 +/- 0.4 for A, M, and P respectively for one-third regions in the left lung. Similar results were found in the right lung. No difference in RPF was found when images were measured either by including the largest of visible parenchymal vessels or when such vessels were excluded. Flow in A of LL and RL was less than that in M or P. Average WO was about 10%, with a range of 0-41% of the LA curve area. There was no significant difference between one-third segment WO using pairwise comparison on the left and right sides when tested separately. RPF values were greater in the posterior vs anterior regions of these supine patients. In conclusion, EBCT can detect gravity related flow differences in the human lung. EBCT has potential for clinical assessment of absolute regional pulmonary flow determination in animals and man.

We present here our initial findings regarding the utility of functional x-ray CT imaging in determining the presence of pulmonary emboli. Recently, x-ray CT has been reported to be a promising technique in detecting pulmonary emboli through direct visualization of the clot as a filling defect of the reconstructed vascular lumen with CT scanning occurring during i.v. contrast drip. To determine whether functional imaging via the dynamic mode of electron beam CT might add to the sensitivity and specificity of pulmonary emboli diagnosis through the visualization of pulmonary parenchymal blood flow and its associated temporal parameters, we scanned 32 pigs and report here our findings on 17 pigs evaluated to date. Findings to date show that the evaluation of flow deficits detected via electron beam CT with a small 2-3 sec. bolus contrast injection has the potential to provide improvement in embolus detection over visual inspection of this section CT/continuous infusion contrast where the viewer is looking for unenhanced regions in the pulmonary arteries.

This report outlines a method of automatically segmenting pulmonary perfusion regions from ultrafast CT data. Segmenting was performed by a k-means clustering of unenhanced, maximal enhancement and both parameters. Time density curves from distinct regions were generated and perfusion determined. In two example subjects with pulmonary embolism hypoperfused regions were identified. In a normal subject slight gravity related gradient hypoperfusion was also detected. These preliminary results suggest that automated segmentation of UFCT pulmonary perfusion scans is possible and could greatly improve the clinical utility of the scanning technology.

Various imaging modalities permit direct observation of the pulmonary arterial tree within the intact lung. We have been concerned with finding a means for efficient organization of the data such that they can reveal certain aspects of the hemodynamic function of the tree. Commonly, pulmonary arterial morphometric data have been summarized by grouping the individual vessel segments according to generation or order and then averaging the dimensions within each generation or order. The most effective criteria for grouping has been a question, and some criteria are not applicable to imaging methods having limited resolution. We have considered an alternative approach in which we begin with the concept that the bifurcating, volume filling characteristics of the tree put constraints of the structure such that the assignment of orders or generations may be superfluous. The scale independent, or fractal, appearance of the tree suggests that one might consider the three vessel segments joined at a bifurcation to be the fundamental repeating morphometric unit descriptive of the tree. The analysis is based on the information in the diameters of the three vessels at each bifurcation. These diameters, D₁ the parent vessel diameter, and D₂ and D₃, the two daughter vessel diameters are used to calculate (beta) ₁ which is the harmonic mean of (beta) ₁ equals log2/[log2D₁ - log(D₁ + D₂], where (beta) ₁ is the quantitative descriptor of each bifurcation of the tree. Within the range of resolution of the imaging modality, a statistical sample of the values of (beta) ₁ can provide an estimate of (beta) ₁. To put the utility of (beta) ₁ in perspective, we introduce the concept of cumulative vascular volume, which is the arterial volume upstream from all of the locations within the arterial tree that have the same intravascular pressure. The distribution of intravascular pressure from arterial inlet to capillary inlet as a function of cumulative vascular volume can be expressed in terms of (beta) ₁. Thus, a sample containing a sufficiently large number of bifurcations can be used to relate the structural image data to pulmonary arterial tree hemodynamic function. Microfocal pulmonary angiographic data provide examples of the application of this concept.

Phase contrast magnetic resonance (PC MR) imaging provides an accurate, non- invasive method for blood flow quantification. Unlike conventional MR images that are derived from the amplitude of the proton signal, velocity maps are reconstructed from the phase information of the MR signal. When a magnetic field gradient is applied along the axis of a vessel, intravascular magnetic spins accumulate a phase- shift that is proportional to flow velocity. The phase-shifts are mapped into a 2D array composed of flow velocities. The velocities over an entire cross-sectional area of a blood vessel can then be summed to quantitate actual blood flow. We have used PC MR imaging to quantitate flow in a flow phantom and human subjects. In flow phantom studies, a significant correlation was found between PC MR flow measurements made proximal and distal to a bifurcation (r² equals 0.999, N equals 5). In 6 human subjects, we found right pulmonary artery (PA) blood flow comprised 53% +/- 1% (mean +/- SEM) of total PA blood flow with the remaining 47% +/- 1% provided by the left PA (difference not statistically significant). Blood flow in the descending aorta, distal to the takeoffs of arteries to the head and upper extremeties, equaled 75% +/- 5% of the blood flow in the ascending aorta. PC MR imaging promises to be a useful tool in the evaluation of blood flow. Advantages of this method include the ability to profile flow velocities over the entire cross-sectional area of the vessel and non-invasive analysis of structures not accessible by other imaging modalities.

A fast, volume scanning, CT method is used to explore the feasibility of quantitating functional aspects of the in situ splanchnic microcirculation. Anesthetized pigs were scanned during and following the injection of contrast agent into the aorta. The indicator dilution curves generated by the passage of contrast medium through an imaged region of interest in the gut wall or through the liver parenchyma, were used to compute regional tissue perfusion and intravascular blood content of the tissue. Splanchnic perfusion was modulated by intra-arterial injection of Bradykinin and by the intragastric infusion of alcohol or hydrochloric acid. The results are consistent with values obtained with more invasive traditional methods for estimating these parameters under similar experimental conditions. We conclude that the resolution of the CT imaging method permits quantitative evaluation of changes in those splanchnic microcirculation following physiologic stimuli. The importance of bowel motion is apparent in these analyses. Indeed, the poorly periodic motion of the gut, even though it is slower than that of the heart wall, presents a greater problem than does the rapid motion of the heart wall, which is gateable because of its cycle-to-cycle reproducibility.

A project is underway to develop automated methods of fusing cerebral magnetic resonance angiography (MRA) and x-ray angiography (XRA) for creating accurate visualizations used in planning treatment of vascular disease. We have developed a vascular phantom suitable for testing segmentation and fusion algorithms with either derived images (psuedo-MRA/psuedo-XRA) or actual MRA or XRA image sequences. The initial unilateral arterial phantom design, based on normal human anatomy, contains 48 tapering vascular segments with lumen diameters from 2.5 millimeter to 0.25 millimeter. The initial phantom used rapid prototyping technology (stereolithography) with a 0.9 millimeter vessel wall fabricated in an ultraviolet-cured plastic. The model fabrication resulted in a hollow vessel model comprising the internal carotid artery, the ophthalmic artery, and the proximal segments of the anterior, middle, and posterior cerebral arteries. The complete model was fabricated but the model's lumen could not be cleared for vessels with less than 1 millimeter diameter. Measurements of selected vascular outer diameters as judged against the CAD specification showed an accuracy of 0.14 mm and precision (standard deviation) of 0.15 mm. The plastic vascular model produced provides a fixed geometric framework for the evaluation of imaging protocols and the development of algorithms for both segmentation and fusion.

The objective of this study was to establish the accuracy of an holographic display approach for detection of stenoses in coronary arteries. The rationale for using an holographic display approach is that multiple angles of view of the coronary arteriogram are provided by a single 'x-ray'-like film, backlit by a special light box. This should be more convenient in that the viewer does not have to page back and forth through a cine angiogram to obtain the multiple angles of view. The method used to test this technique involved viewing 100 3D coronary angiograms. These images were generated from the 3D angiographic images of nine normal coronary arterial trees generated with the Dynamic Spatial Reconstructor (DSR) fast CT scanner. Using our image processing programs, the image of the coronary artery lumen was locally 'narrowed' by an amount and length and at a location determined by a random look-up table. Two independent, blinded, experienced angiographers viewed the holographic displays of these angiograms and recorded their confidence about the presence, location, and severity of the stenoses. This procedure evaluates the sensitivity and specificity of the detection of coronary artery stenoses as a function of the severity, size, and location along the arteries.

The assessment of blood flow in the gastrointestinal mucosa might be an important factor for the diagnosis and treatment of several diseases such as ulcers, gastritis, colitis, or early cancer. The quantity of blood flow is roughly estimated by computing the spatial hemoglobin distribution in the mucosa. The presented method enables a practical realization by calculating approximately the hemoglobin concentration based on a spectrophotometric analysis of endoscopic true-color images, which are recorded during routine examinations. A system model based on the reflectance spectroscopic law of Kubelka-Munk is derived which enables an estimation of the hemoglobin concentration by means of the color values of the images. Additionally, a transformation of the color values is developed in order to improve the luminance independence. Applying this transformation and estimating the hemoglobin concentration for each pixel of interest, the hemoglobin distribution can be computed. The obtained results are mostly independent of luminance. An initial validation of the presented method is performed by a quantitative estimation of the reproducibility.

A significant increase in diagnostic incidence of prostate cancer underscores the need to accurately stratify and quantify the cancers to facilitate appropriate therapy. Currently, there is no reliable method to preoperatively predict pathological stage and thus malignant potential of prostate cancer. Tumor volume and microvessel density have been shown postoperatively to be accurate predictors of cancer metastatic potential. 3D visualization and analysis of image volumes produced from series of immunocytochemically stained pathological sections may improve our understanding of the relationships of the tumor to angiogenesis, i.e., to the microvessel density of the tumor. Sequential thinly sliced (approximately 7 microns) pathological sections of the prostate will be differentially stained with fluorescent antibodies to clotting factor VIII- related antigen, which labels the endothelial cells of the vessels, facilitating automated color separation for visualization of the microvessels. Digitized images of the sections can be synthesized into 3D volumes and measured to quantify vessel quantity and density. Using 3D colorwash and transparency display techniques, anatomic and densitrometric relationships between the tumor and microvessels can be visualized. The microvessel density can be measured using image processing algorithms and compared to measurements made by pathologists. Advanced approaches to imaging the prostate in vivo include dynamic MRI techniques using contrast agents to accurately detect and quantify the region of prostate cancer. The cancerous region can be correlated with histologic specimens using the same methods described for measurement of microvessel density. This detailed information could lead to improved methods to properly stratify patients with diagnosed prostate cancer.

An airway cast was made and imbedded in a solid polyurethane block of a contrasting color. The block was sequentially milled and photographed. The sequential photographs were scanned to create an image database which was analyzed on VIDA; a multidimensional image analysis software package. The technique shows promise as a semi-automated process for generating a high resolution morphometric database from airway casts.

We present an algorithm for the construction of a 3D bronchial tree representation based on an asymmetric dichotomy model proposed by Weibel. The model has a total of ten branch generations and three independent parameters for each branch. The bronchial tree model is constructed by dividing each branch of a generation into two subbranches, except for the first few generations where some branches divided into three subbranches. The bronchial tree model is used as a testground to simulate different CT acquisition/reconstruction schemes and examine the role of CT parameters, such as slice thickness and overlap/spacing, on the measurement of fractal properties of the lung. A digital version of the model with high resolution (0.2 mm/pix) is created first. Acquisition schemes are then simulated by dividing this high resolution version of the model into cross-sections and summing one or more cross- sections to form a series of 2D slices. Three acquisition schemes are simulated by choosing different ways to generate these 2D slices. To reconstruct the 3D bronchial tree structure from a series of 2D slices, interpolation is applied between the 2D slices to recover the original volume. Two types of interpolations are used here: (1) slice repetition (2) linear interpolation. Fractal measures of the 3D bronchial tree structure are estimated from the reconstructed images and compared with the same measures from the high resolution model.

Accurate physiological measurements of the parameters like branching angles, branch lengths, and diameters of bronchial tree structures help in addressing the mechanistic and diagnostic questions related to obstructive lung disease. In order to facilitate these measurements, bronchial trees are reduced to a central axis tree. The approach we take employs first setting up a theoretical computerized tree structure, and then applying a 3D analysis to obtain the required anatomical data. A stick model was set up in 3D, with segment endpoints and diameters as input parameters to the model generator. By fixing the direction in which the slices are taken, a stack of 2D images of the generated 3D tree model is obtained, thereby simulating bronchial data sets. We design a two pass algorithm to compute the central axis tree and apply it on our models. In the first pass, the topological tree T is obtained by implementing a top-down seeded region growing algorithm of the 3D tree model. In the second pass, T is used to region growth along the axes of the branches. As the 3D tree model is traversed bottom-up, the centroid values of the cross sections of the branches are stored in the corresponding branch of T. At each bifurcation, the branch point and the three direction vectors along the branches are computed, by formulating it as a nonlinear optimization problem that minimizes the sum of least squares error of the centroid points of the corresponding branches. By connecting the branch points with straight lines, we obtain a reconstructed central axis tree which closely corresponds to the input stick model. We also studied the effect of adding external noise to out tree models and evaluating the physiological parameters. We conclude with the results of our algorithm on real airway trees.

Fractal geometry is increasingly being used to model complex naturally occurring phenomena. There are two types of fractals in nature-geometric fractals and stochastic fractals. The pulmonary branching structure is a geometric fractal and the intensity of its grey scale image is a stochastic fractal. In this paper, we attempt to quantify the texture of CT lung images using properties of both types of fractals. A simple algorithm for detection of abnormality in human lungs, based on 2D and 3D fractal dimensions, is presented. This method involves calculating the local fractal dimensions, based on intensities, in the 2D slice to aid enhancement. Following this, grey level thresholding is performed and a global fractal dimension, based on structure, for the entire data is estimated in 2D and 3D. High resolution CT images of normal and abnormal lungs were analyzed. Preliminary results showed that classification of normal and abnormal images could be obtained based on the differences between their global fractal dimensions.

The nuclear magnetic resonance (NMR) properties of lung are markedly affected by the alveolar air-tissue interface, which produces internal magnetic field inhomogeneity because of the different magnetic susceptibilities of air and water. This internal magnetic field inhomogeneity results in a marked shortening of the free induction decay (FID) (in the time domain) and in inhomogeneous NMR line broadening (in the frequency domain). The signal loss due to internal magnetic field inhomogeneity can be measured as the difference Δ between the spin-echo signals obtained using temporally symmetric and asymmetric spin-echo sequences; the degree of asymmetry of the asymmetric sequence is characterized by the asymmetry time τ_a. In accordance with predictions based on the analysis of theoretical models, experiments in excised rat lungs (studied at various inflation levels) have shown that Δ depends on τ_a and is very low in degassed lungs. When measured at τ_a equals 6 ms, the difference signal (Δ_6ms) increases markedly with alveolar opening but does not vary significantly during the rest of the inflation-deflation cycle. In edematous (oleic acid-injured) lungs, the values of Δ_6ms measured at low inflation levels are significantly below those observed in normal lungs. These results suggest that Δ_6ms is very sensitive to alveolar recruitment and relatively insensitive to alveolar distension. Therefore, measurements of Δ_6ms may provide a means of assessing the relative contributions of these two factors to the pressure-volume behavior of lung. Such measurements may contribute to the characterization of pulmonary edema (for example, by detecting the loss of alveolar air-tissue interface due to alveolar flooding, by differentiating interstitial from alveolar pulmonary edema, and by assessing the effects of positive airway pressures). NMR lineshape measurements can also provide valuable information regarding lung geometry and the characterization of pulmonary edema.

Volumetric data sets in the lung are formed by stacking sequential computed tomography (CT) slices. The resulting volumetric image file, however, is nonuniformly resolved; the in-plane resolution is greater than that in the z-axis (slice thickness) dimension. The purpose of this study was to determine the effect of branch orientation within the image volume on the measurement and resolvability of airway branches. An isolated canine lobe was sequentially scanned at three orientations with a Siemens Somotom Plus S CT Scanner using a 124.4 mm field of view, 137 kVp, 220 mAs, a 2 mm slice thickness, and a 1 mm table feed. A grid size of 256², resulted in an in- plane pixel dimension of 0.49 mm. The lobe was inflated to an airway pressure of 20 cm H₂O with the main bronchus of the lobe aligned approximately perpendicular to the scan plane. The entirety of the lobe was scanned at this orientation. The inflated lobe was then reoriented 90 degree(s) clockwise (as if the lobe was sitting on a clock face) and again the entirety of the lobe was scanned. The lobe was rotated a third time 90 degree(s) in plane from the first lobe alignment and again rescanned. Airway trees were segmented for each orientation and there were significant differences in the number of resolved branches among the three segmented trees. Measurement of airways over four millimeters in diameter was not affected by orientation. Airways smaller than two millimeters in diameter showed surprising similarity in measured diameter, but all airways were not resolved at all orientations. The angle of orientation of the individual airways with respect to the scan axis was therefore calculated to determine the angle at which a branch of given diameter was no longer resolved. There was surprising similarity of measured diameter in these smaller airways with orientation, if the branch was resolved. The orientation at which branches were resolved was quite variable, suggesting a high degree of randomness in the segmented branches at this level of branch resolution.

Accurate assessment of airway physiology, evaluated in terms of geometric changes, is critically dependent upon the accurate imaging and image segmentation of the 3D airway tree structure. We have previously reported a knowledge-based method for 3D airway tree segmentation from high resolution CT (HRCT) images. Here we report a substantially improved version of our method. In the current implementation, the method consists of several stages. First, the lung borders are automatically determined in the 3D set of HRCT data. The primary airway tree is semi-automatically identified. In the next stage, potential airways are determined in individual CT slices using a rule- based system that uses contextual information and a priori knowledge about pulmonary anatomy. Using 3D connectivity properties of the pulmonary airway tree, the 3D tree is constructed from the set of adjacent slices. The method's performance and accuracy were assessed in five 3D HRCT canine images. Computer-identified airways matched 226/258 observer-defined airways (87.6%); the computer method failed to detect the airways in the remaining 32 locations. By visual assessment of rendered airway trees, the experienced observers judged the computer-detected airway trees as highly realistic.

A general methodology has been developed for computer interpretation of medical images, based on an explicit anatomical model. A test system for analyzing posterior- anterior (PA) chest x-rays has been implemented. The inferencing and control system identifies the major lung structures in the image, and then flags any suspected abnormalities. Image and model data are transformed into a feature space where they are represented in terms of edge descriptions. The inference engine compares the image and model in feature space to label the edges anatomically, and check for normality. The control system schedules events within the inference engine and coordinates interaction with the model and image processing routines. The control architecture is blackboard-based, with a separate data frame for each structure to be identified. The anatomical model uses fuzzy sets to provide ranges of feature values which are considered normal or indicative of a particular abnormality. This allows the inference engine to give a confidence score and linguistic description to each decision. Mediastinum, cardiac border, domes of the diaphragm, ribs and lung outline have been modeled. Their automatic identification allows diagnostic checks such as the cardiothoracic ratio, comparison of right and left lungs to identify lobular collapse and inspection of interfaces in terms of shape and clarity. The inference engine provides simple comments on its findings, making it suitable for pre- and double-checking of images.

There have been considerable research efforts in the area of vocal tract modeling, but there is still a very small body of information regarding direct 3D measurements of the vocal tract shape. The purpose of this study was to acquire, using MRI, an inventory of 3D vocal tract airway shapes that correspond to a particular set of vowels and consonants. The 3D shapes were analyzed to find the cross-sectional areas along the centerline extending from the glottis to the mouth to produce an 'area function'. These area functions were then used as input to a computer model of 1D acoustic wave propagation in the vocal tract. The final result was a simulation of the speech waveform radiated from the mouth and closely similar to the speech waveform recorded from the same subject.

Issues about the accuracy of ROI definition methods for FDOPA PET studies were investigated. An MRI-based ROI method and manually defined ROI method were compared using a computer simulated brain phantom and four real PET studies. The results indicate the error or discrepancy between MRI-based ROI and manually defined ROI is small (< equals 5%) at different head orientations and different noise levels. The VOI is not sensitive to the orientation, but the mid-plane ROI is also fairly reliable.

We have previously developed a knowledge-based method of factor analysis to analyze dynamic nuclear medicine image sequences. In this paper, we analyze dynamic PET cerebral glucose metabolism and neuroreceptor binding studies. These methods have shown the ability to reduce the dimensionality of the data, enhance the image quality of the sequence, and generate meaningful functional images and their corresponding physiological time functions. The new information produced by the factor analysis has now been used to improve the estimation of various physiological parameters. A principal component analysis (PCA) is first performed to identify statistically significant temporal variations and remove the uncorrelated variations (noise) due to Poisson counting statistics. The statistically significant principal components are then used to reconstruct a noise-reduced image sequence as well as provide an initial solution for the factor analysis. Prior knowledge such as the compartmental models or the requirement of positivity and simple structure can be used to constrain the analysis. These constraints are used to rotate the factors to the most physically and physiologically realistic solution. The final result is a small number of time functions (factors) representing the underlying physiologic processes and their associated weighting images representing the spatial localization of these functions. Estimation of physiological parameters can then be performed using the noise-reduced image sequence generated from the statistically significant PCs and/or the final factor images and time functions. These results are compared to the parameter estimation using standard methods and the original raw image sequences. Graphical analysis was performed at the pixel level to generate comparable parametric images of the slope and intercept (influx constant and distribution volume).

In the physiologic modeling of dynamic positron emission tomography (PET) data, one is typically interested in the average reconstructed activity for voxels within the boundaries of some volume of interest (VOI), the uncertainty of this value, and possibly correlations with other VOIs. These calculations have been partially carried out in the past by drawing appropriate 2D regions on a number of reconstructed images in a PET volume, and then summing the voxel values within these regions. Summing voxel values in this fashion provides a value for activity within a volume, but does not allow calculation of the statistical uncertainty. To perform the latter task, calculations must be performed on the raw tomographic data, and thus the 2D regions must be specified on the originally acquired tomographic slices. We have developed software enabling clinicians to specify regions along any preferred viewing axis, and yet perform calculations on these regions to obtain the complete VOI covariance matrix. In this process, 2D regions are first drawn on slices of arbitrary orientation. Stacks of these regions are then tiled together to form a closed 3D surface model for each VOI. Cross sections of these VOIs in the originally acquired orientation are obtained by intersecting the 3D surface models with a series of appropriately transformed slicing planes. The resliced regions are then projected into tomographic sinogram space and the activity and uncertainty is calculated for each region. Knowledge of the complete covariance matrix allows combination of these 2D region activity values into 3D volume activity values and uncertainties.

The purpose of this research was to compare a physiological model of ⁸²Rb in the myocardium with two reduced order models with regard to their ability to assess physiological parameters of diagnostic significance. A three compartment physiological model of ⁸²Rb uptake in the myocardium was used to simulate kinetic region of interest data from a positron emission tomograph (PET). Simulations were generated for eight different blood flow rates reflecting the physiological range of interest. Two reduced order models which are commonly used with myocardial PET studies were fit to the simulated data and the parameters of the reduced order models were compared with the physiological parameters. Then all three models were fit to the simulated data with noise added. Monte carlo simulations were used to evaluate and compare the diagnostic utility of the reduced order models.

Technological advances in brain imaging have revolutionized diagnosis in neurology and neurological surgery. Major imaging techniques include magnetic resonance imaging (MRI) to visualize structural anatomy, positron emission tomography (PET) to image metabolic function and cerebral blood flow, magnetoencephalography (MEG) to visualize the location of physiologic current sources, and magnetic resonance spectroscopy (MRS) to measure specific biochemicals. Each of these techniques studies different biomedical aspects of the brain, but there lacks an effective means to quantify and correlate the disparate imaging datasets in order to improve clinical decision making processes. This paper describes several techniques developed in a UNIX-based neurodiagnostic workstation to aid the noninvasive presurgical evaluation of epilepsy patients. These techniques include online access to the picture archiving and communication systems (PACS) multimedia archive, coregistration of multimodality image datasets, and correlation and quantitation of structural and functional information contained in the registered images. For illustration, we describe the use of these techniques in a patient case of nonlesional neocortical epilepsy. We also present out future work based on preliminary studies.

An automated method based on maximizing a 2D correlation coefficient between images is proposed to realign consecutive images obtained in functional MR image sequences. Dynamic Gadolinium-enhanced osteosarcoma studies are analyzed to study the influence of interframe motion and assess the registration method. The alignment procedure is evaluated by subtraction of the pre-injection image and by factor analysis of medical image sequences. The effect of motion correction is demonstrated by both techniques and the correlation method is compared to a procedure based on external markers.

A feasibility study was conducted to segment 1.5T fMRIs into gray matter and large veins using individual pixel intensity and temporal phase delay as two correlated parameters in gradient echo images. The time-course of each pixel in gradient echo images acquired during visual stimulation with a checkerboard flashing at 8Hz was correlated to the stimulation 'on'-'off' sequence to identify activated pixels. The temporal delay of each activated pixels was estimated by fitting its time-course to a reference sinusoidal function. The mean signal intensity difference of the activated pixels was computed by subtracting the average of the 'on' images from the average of the 'off' images. After replacing each activated pixel with 2D features (i.e., intensity and time-delay), a clustering method based on a K-means algorithm was employed to classify vein and tissue pixels. Good demarcation between large veins and activated gray matter was achieved with this method.

We study changes of EEG coherence induced by cognitive activation of the brain in healthy subjects. Coherence is considered a correlation coefficient between two electrical signals and may reflect the functional relationships between brain areas. Local coherence is calculated on paired adjacent scalp EEG electrodes whereas inter- regional coherence also includes nonadjacent paired electrodes. We developed a color computer animation display method allowing us to visualize the intensity and regional spatial distribution of coherence changes in time during a cognitive activation process.

NIFTY (NeuroImaging Functional Toolkit) is a tool designed to perform quantitative analysis and visualization of neurofunctional magnetic resonance image (fMRI) data sets. NIFTY is an OSF/Motif application which utilizes the AVW (a visualization workshop) imaging library developed in our laboratory and includes algorithms for robust image registration, statistical analysis, and mapping of neurofunctional data sets. Anisotropic diffusion routines can be used to enhance the signal-to-noise- ratio of these images. Tools capable of histogram equalization, thresholding, volume rendering, atlas matching, and a large number of other functions can then be used to visualize the data. NIFTY's development will offer a robust and flexible system of essential functions integrated into an interactive, graphically-oriented program, allowing neuroscientists the means by which to process, visualize, and interpret their data.

This conference on physiology and function covers a wide range of subjects, including the vasculature and blood flow, the flow of gas, water, and blood in the lung, the neurological structure and function, the modeling, and the motion and mechanics of organs. Many technologies are discussed. I believe that the list would include a robotic photographer, to hold the optical equipment in a precisely controlled way to obtain the images for the user. Why are 3D images needed? They are to achieve certain objectives through measurements of some objects. For example, in order to improve performance in sports or beauty of a person, we measure the form, dimensions, appearance, and movements.

We have developed a method for measuring the detailed in vivo three dimensional geometry of the left and right ventricles using cine-magnetic resonance imaging. From data in the form of digitized short axis outlines, the normal vectors, principal curvatures and directions, and wall thickness were computed. The method was evaluated on simulated ellipsoids and on human MRI data. Measurements of normal vectors and of wall thickness were very accurate in simulated data and appeared appropriate in patient data. On simulated data, measurements of the principal curvature k₁ (corresponding approximately to the short axis direction of the left ventricle) and of principal directions were quite accurate, but measurements of the other principal curvature (k₂) were less accurate. The reasons behind this are considered. We expect improvements in the accuracy with thinner slices and improved representation of the surface data. Gradient echo images were acquired from 8 dogs with a 1.5T system (Philips Gyroscan) at baseline and four months after closed chest experimentally produced mitral regurgitation (MR). The product (k₁ + k₂) X wall thickness averaged over all slices at end-diastole was significantly lower after surgery (n equals 8, p < 0.005). These geometry changes were consistent with the expected increase in wall stress after MR.

There are several theoretic advantages to synchronizing positive pressure breaths with the cardiac cycle, including the potential for improving distribution of pulmonary and myocardial blood flow and enhancing cardiac output. We evaluated the effects of synchronizing respiration to the cardiac cycle using a programmable ventilator and electron beam CT (EBCT) scanning. The hearts of anesthetized dogs were imaged during cardiac gated respiration with a 50msec scan aperture. Multislice, short axis, dynamic image data sets spanning the apex to base of the left ventricle were evaluated to determine the volume of the left ventricular chamber at end-diastole and end-systole during apnea, systolic and diastolic cardiac gating. We observed an increase in cardiac output of up to 30% with inspiration gated to the systolic phase of the cardiac cycle in a nonfailing model of the heart.

Magnetic resonance imaging (MRI) is an extremely versatile technique for noninvasive imaging of the anatomy, structure, and physiological function of the heart and other soft tissues and organs. Although mathematical models have often been used to enhance the information content of medical images, these models are most often based on the physics of the imaging system rather that the properties of the target organ or tissue. We use finite element (FE) models of regional mechanical and electrical function in the intact heart to compute 3D distributions of important physiological field variables, such as myocardial stress, that cannot be imaged directly. A parametric model of the heart based on the physical properties of the organ as a material continuum provides a general and convenient way to synthesize clinical data, such as multidimensional images, with experimental tests, such as biomechanical and histological studies.

A 3D, nonlinear finite element (FE) model of the diastolic canine heart was constructed from multislice magnetic resonance images (MRI). The model was solved using the p-version of the FE method to predict stress and deformation in diastole. Finite element models were employed in an 'inverse' problem to estimate the nonlinear material properties of intact diastolic myocardium. Additionally, FE models were employed to examine the effects of RV pressure overload on LV pressure-volume (P- V) relationships in the pathologic heart. A 3D, nonlinear FE model had 1,704 degrees of freedom and 8 elements, it had a maximum principal stress value of 429,127 dynes/cm² and a minimum principal stress of -344,599 dynes/cm² at p equals 6. Average nonlinear material parameters estimated for 6 dogs were E equals 28,722 +/- 15,984 dynes/cm² and c equals 0.00182 +/- 0.00232 cm²/dyne. Examination of the effects of RV pressure increase on LV P-V relationships indicated substantially different effects of RV pressure overload on the different pathologic conditions (p < 0.005 by ANOVA) with increasing RV pressure having a more pronounced effect on the dilated heart than the hypertrophied heart. When the mechanical effects of the pericardium were included in the model, at higher RV pressures, all of the pressure-volume (P-V) curves became similar indicating that at higher RV pressures, the P-V curves were independent of ventricular shape and material properties and depended only on the RV pressure. In conclusion, FE models of the heart were constructed from MRI images of the heart and were employed to study diastolic ventricular function in the normal and pathologic heart.

The syndrome of constrictive pericarditis (CP) presents a diagnostic challenge to the clinician. This study was undertaken to determine whether cine computed tomography (CT), a cardiac imaging technique with excellent temporal and spatial resolution, can reliably demonstrate the unique abnormalities of pericardial anatomy and ventricular physiology present in patients with this condition. A second goal of this study was to determine whether the presence of diseased thickened pericardium, by itself, imparts cardiac impairment due to abnormalities of ventricular diastolic function. Methods: Twelve patients with CP suspected clinically, in whom invasive hemodynamic study was consistent with the diagnosis of CP, underwent cine CT. They were subdivided into Group 1 (CP, N equals 5) and Group 2 (No CP, N equals 7) based on histopathologic evaluation of tissue obtained at the time of surgery or autopsy. A third group consisted of asymptomatic patients with incidentally discovered thickened pericardium at the time of cine CT scanning: Group 3 (ThP, N equals 7). Group 4 (Nl, N equals 7) consisted of healthy volunteer subjects. Results: Pericardial thickness measurements with cine CT clearly distinguished Group 1 (mean equals 10 +/- 2 mm) from Group 2 (mean equals 2 +/- 1 mm), with diagnostic accuracy of 100% compared to histopathological findings. In addition, patients in Group 1 had significantly more brisk early diastolic filling of both left and right ventricles than those in Group 2, which clearly distinguished all patients with, from all patients without CP. Patients in Group 3 had pericardial thicknesses similar to those in Group 1 (mean equals 9 +/- 1 mm, p equals NS), but had patterns of diastolic ventricular filling that were nearly identical to Group 4 (Nl). Conclusions: The abnormalities of anatomy and ventricular function present in the syndrome of constrictive pericarditis are clearly and decisively identified by cine CT. This allows a reliable distinction between patients with constrictive pericarditis and those with cardiomyopathy. The presence of diseased thickened pericardium does not by itself impart impairment of ventricular diastolic function. Thus, definitive diagnosis of constrictive pericarditis requires demonstration of both abnormal anatomy and physiology.

Following myocardial infarction, the size of the infarcted region and the systolic functioning of the noninfarcted region are commonly assessed by various cross- sectional imaging techniques. A series of images representing successive phases of the cardiac cycle can be acquired by several imaging modalities including electron beam computed tomography, magnetic resonance imaging, and echocardiography. For the assessment of patterns of ventricular contraction, images are commonly acquired of ventricular cross-sections normal to the 'long' axis of the heart and parallel to the mitral valve plane. The endocardial and epicardial surfaces of the myocardium are identified. Then the ventricle is divided into sectors and the volumes of blood and myocardium within each sector at multiple phases of the cardiac cycle are measured. Regional function parameters are derived from these measurements. This generally mandates the use of a polar or cylindrical coordinate system. Various algorithms have been used to select the origin of this coordinate system. These include the centroid of the endocardial surface, the epicardial surface, or of a polygon whose vertices lie midway between the epicardial and endocardial surfaces of the myocardium (centerline method). Another algorithm has been developed in our laboratory. This uses the centroid (or center of mass) of the myocardium exclusive of the ventricular cavity. Each of these choices for origin of coordinate system can be derived from the end- diastolic image or from the end-systolic image. Alternately, new coordinate systems can be selected for each phase of the cardiac cycle. These are referred to as 'floating' coordinate systems. A series of computer models have been developed in our laboratory to study the effects of each of these choices on the regional function parameters of normal ventricles and how these choices effect the quantification of regional abnormalities after myocardial infarction. The most sophisticated of these is an interactive program with a graphical user interface which facilitates the simulation of a wide variety of dynamic ventricular cross sections. Analysis of these simulations has led to a better understanding of how polar coordinate system placement influences the results of quantitative regional ventricular function assessment. It has also created new insight into how the appropriateness of the placement of such a polar coordinate systems can be objectively assessed. The validity of the conclusions drawn from the analysis of simulated ventricular shapes was validated through the analysis of outlines extracted from cine electron beam computed tomographic images. This was done using another interactive software tool developed specifically for this purpose. With this tool, the effects on regional function parameters of various choices for origin placement can be directly observed. This has proven to reinforce the conclusions drawn from the simulations and has led to the modification of the procedures used in our laboratory. Conclusions: The so-called floating coordinate systems are superior to fixed ones for quantification of regional left ventricular contraction in almost every respect. The use of regional ejection fractions with a coordinate system origin located at the centroid of the endocardial surface can lead to 180 degree errors in identifying the location of a myocardial infarction. This problem is less pronounced with midline and epicardium- based centroids and does not occur when the centroid of the myocardium is used. The quantified migration of myocardial mass across sector boundaries is a useful indicator of an inappropriate choice of coordinate system origin. When the centroid of the myocardium falls well within the ventricular cavity, as it usually does, it is a better location for the origin for regional analysis than any of the other centroids analyzed.

Study of the motility of the gut and flow in the gut lumen is complex. We have used simple models of the colon to suggest that focal, segmental narrowing of the lumen of a large-bore tube produces an increase in distance traveled by solids and liquids compared to a tube of the same volume which has smooth cylindrical walls. This suggests that the colonic septae may promote efficient transit of solids and liquids even in an organ with large diameter.

The goal of this study was to demonstrate quantitatively, using ultrasound (US) recording techniques, the extent of motion of the sacroiliac joint achieved using manual medicine techniques. Initial judgements of perceived (i.e., felt) SI mobility during manual examination were made on 22 subjects. Baseline no movement ultrasound images (static) were obtained of the left and right SI joints at two levels-- posterior-superior-iliac-spine and inferior (PSIS, INF)--and two projections (AP and LAT). Manual medicine spring testing of the SI joint was then performed while ultrasound recordings (on video) were made. The differences between baseline separation of the SI joint and displacement distance during spring testing were measured by six radiologists who typically read US images. Significant movement of at least one SI joint was demonstrated in 91% of the subjects using ultrasound recordings. The extent of movement appeared to corroborate the experience of manual medicine practitioners.

3D visualization of CT sectional images in a video workstation with a medical imaging analysis system is very helpful to the surgeon in the selection of the optimal donor site for autogenous grafts. The sites of interest were represented on the monitor as free, interactively movable objects which could be observed three-dimensionally from all perspectives. By means of superimposition, turning, and penetration of these objects, the ideal donor site for the graft, in our examples parts from the left and right iliac crest, could be determined. An additional method for this determination is computer- assisted generation of a graft pattern from the CT data set for cases where no graftable object in the volume of interest can be found. In a special procedure a graft from biocompatible material can then be duplicated from this pattern. A reconstructive operation with 3D planning was performed on 12 patients with osseous defects in the area of the jaws and facial cranium. In the search for appropriate grafts from the patient's own body the iliac crest, with its specific volume, was selected for all patients.

Ultrasound imaging of the temporomandibular joint has been problematic due to the lower frequency of the transducers used up to the present time. Imaging of temporomandibular joint structures being utilizable for diagnosis and therapy was only possible through time-consuming and expensive radiological image yielding procedures (computertomography, magnetic resonance imaging). 84 temporomandibular joints in 42 patients were examined clinically, radiologically, by axiographic tracing, magnetic resonance imaging and ultrasound imaging. An ultrasound unit was used with a high- frequency 13MHz transducer. The temporomandibular joint was examined preauricularily; by this the lateral section of the joint could be represented. The image sequences in functional condylus movements were taped via a video output into a film recorder. Selected ultrasound images from the beginning to the end of the movement could then be digitalized and read into a personal computer to be evaluated. The computer then calculated a line of movement and the angle of the joint's course. By ultrasound imaging the joint space could be represented and measured clearly. Compared with the space measured in the magnetic resonance image the value determined by ultrasonography was a tenth power more exact. The computer-supported image analysis of the condylus movements led to an exact presentation of the condylus course. The sonographically determined condylar guidance corresponded to the value traced by axiography with high significance within a range of 3 degrees. The temporomandibular joint's disc could be localized just as exactly as with the magnetic resonance imaging. The use of a 13MHz transducer offers a new low-cost method of noninvasive dynamic imaging of important temporomandibular joint structures. The possibility of video and computer support enables movement analysis and opens new possibilities in the morphological and functional evaluation of the temporomandibular joint.

Electron beam computed tomography is unparalleled in its ability to consistently produce high quality dynamic images of the human heart. Its use in quantification of left ventricular dynamics is well established in both clinical and research applications. However, the image analysis tools supplied with the scanners offer a limited number of analysis options. They are based on embedded computer systems which have not been significantly upgraded since the scanner was introduced over a decade ago in spite of the explosive improvements in available computer power which have occured during this period. To address these shortcomings, a workstation-based ventricular analysis system has been developed at our institution. This system, which has been in use for over five years, is based on current workstation technology and therefore has benefited from the periodic upgrades in processor performance available to these systems. The dynamic image segmentation component of this system is an interactively supervised, semi-automatic surface identification and tracking system. It characterizes the endocardial and epicardial surfaces of the left ventricle as two concentric 4D hyper-space polyhedrons. Each of these polyhedrons have nearly ten thousand vertices which are deposited into a relational database. The right ventricle is also processed in a similar manner. This database is queried by other custom components which extract ventricular function parameters such as regional ejection fraction and wall stress. The interactive tool which supervises dynamic image segmentation has been enhanced with a temporal domain display. The operator interactively chooses the spatial location of the endpoints of a line segment while the corresponding space/time image is displayed. These images, with content resembling M-Mode echocardiography, benefit form electron beam computed tomography's high spatial and contrast resolution. The segmented surfaces are displayed along with the imagery. These displays give the operator valuable feedback pertaining to the contiguity of the extracted surfaces. As with M-Mode echocardiography, the velocity of moving structures can be easily visualized and measured. However, many views inaccessible to standard transthoracic echocardiography are easily generated. These features have augmented the interpretability of cine electron beam computed tomography and have prompted the recent cloning of this system into an 'omni-directional M-Mode display' system for use in digital post-processing of echocardiographic parasternal short axis tomograms. This enhances the functional assessment in orthogonal views of the left ventricle, accounting for shape changes particularly in the asymmetric post-infarction ventricle. Conclusions: A new tool has been developed for analysis and visualization of cine electron beam computed tomography. It has been found to be very useful in verifying the consistency of myocardial surface definition with a semi-automated segmentation tool. By drawing on M-Mode echocardiography experience, electron beam tomography's interpretability has been enhanced. Use of this feature, in conjunction with the existing image processing tools, will enhance the presentations of data on regional systolic and diastolic functions to clinicians in a format that is familiar to most cardiologists. Additionally, this tool reinforces the advantages of electron beam tomography as a single imaging modality for the assessment of left and right ventricular size, shape, and regional functions.

This paper presents a neural network-based method for intrathoracic airway detection and segmentation from 3D HRCT images. Two feed-forward neural networks are independently trained to identify various airway appearances in 3D CT images. While the first network identifies potential airways located adjacent to vessels, the second network identifies potential airways by assessing the existence of walls surrounding airways, The two networks are combined to construct a dual-network classifier taking its inputs from a 21 X 21 moving subimage window: (1) raw gray-level subimage and (2) 4 directional profiles. By design, each network provides a superset of airways that are present in the CT images and only the airways identified by both networks are considered reliable. After the networks are trained by the generalized delta rule with momentum using limited number of airway/nonairway samples apart from the validation data sets, the generalization performance of the networks is assessed with two independent standards consisting of 282 and 167 observer-traced airways. The performance of the current method is compared with that of the conventional seeded region growing method. Our validation results indicate that the presented method indeed provide enhanced detection of peripheral airways compared to the conventional region growing method.

We characterize the trabecular structure with the aid of fractal dimension. We use alternating sequential filters (ASF) to generate a nonlinear pyramid for fractal dimension computations. We do not make any assumptions of the statistical distributions of the underlying fractal bone structure. The only assumption of our scheme is the rudimentary definition of self-similarity. This allows us the freedom of not being constrained by statistical estimation schemes. With mathematical simulations, we have shown that the ASF methods outperform other existing methods for fractal dimension estimation. We have shown that the fractal dimension remains the same when computed with both the x-ray images and the MRI images of the patella. We have shown that the fractal dimension of osteoporotic subjects is lower than that of the normal subjects. In animal models, we have shown that the fractal dimension of osteoporotic rats was lower than that of the normal rats. In a 17 week bedrest study, we have shown that the subject's prebedrest fractal dimension is higher than that of the postbedrest fractal dimension.

Image segmentation is an important aspect of biomedical image analysis. Here we discuss the implementation of a system for segmentation of confocal electron microscope images of DNA within from rat liver cell nuclei. Segmentation is achieved using controlled continuity splines or 'snakes' proposed by Kass, Witkin, and Teraopoulos. A snake is an energy minimizing spline guided by external constraint forces and influenced by image forces that pull it toward features such as lines and edges. They are active contour models. They lock onto nearby edges, localizing them accurately.

Correlation analysis of the far-zone speckle intensity has been examined in application to biotissue structure investigation. Asymptotic value of the exponential factor of intensity structure function has been proposed as the informative parameter. Generation of the 2D images of this factor distribution is discussed for different illumination conditions. Mapping of the light-scattering biotissue structure images has been carried out for the samples of normal and psoriatic human epidermal strippings.

Adhesion molecules present in the cellular membrane of the endothelium provide sites of leukocyte adherence as a first step in the process of leukocyte migration into the interstitium. New evidence suggests the same adhesion proteins may be responsible for the spread of metastatic tumors by providing a location for tumor cell attachment. A method was sought to quantitate the degree of adhesion molecule expression in the pulmonary capillary endothelium using a recently developed animal model which allows for viewing the lung surface in vivo. Videoimages of the pulmonary vascular system were gathered using this new lung chamber technique. A fully automated digital image processing and analysis (DIPA) system was also developed to estimate the level of intercellular adhesion molecule-1 (ICAM-1) expression on the capillary endothelial cells in these videoimages. Fluorescent microspheres were immunologically bound to the ICAM-1 molecules present on the endothelial cell surface. The DIPA system then located and quantified the fluorescent spots present in the videoimages. The ability of this system to locate and measure the fluorescence was compared with human measurements of the same images.

Imaging methods that have relatively high scan repetition rates can be used to generate indicator dilution curves in spatially defined regions of interest. This image-based analysis of indicator dilution processes has several potential advantages over catheter sampled indicator curve analysis. These include the fact that sampling can be performed in many locations within the vascular system where sampling catheters cannot be placed. Moreover, the multitudes of simultaneous sampling sites that can be accessed with CT would be prohibitively invasive if performed by multiple sampling catheters. Image—based indicator dilution curve analysis does have several limitations and/or special considerations that must be acknowledged and dealt with. It is the purpose of this workshop to highlight some of these special issues and to explore solutions to them. These issues indicate that use of these techniques requires consideration of the application as well as the imaging method.