The lung area has very complicated structure which consists of the bronchus, the pulmonary artery, and the pulmonary vein. So it is difficult for even medical doctors to understand the spatial relationships among the tumor, the bronchus, and the blood vessels. Here we present a 3D image analysis method of lung areas using thin slice CT images, and we apply this system to the differential diagnosis such as malignant or benign decision of the abnormal tissue. This system consists of two steps. The first step is the analysis of the structure of the lung area, and the second step is the visualization of classified pulmonary organs for quantitative analysis.

Time sequenced regional lung parenchymal analysis is dependent upon the successful registration of lung data sets that corrects for local distortions. An elastic model based image registration scheme is proposed and developed in 2D. It models the registration as a deformation process of an elastic material. An example of real lung image registration is given to show the process. The performance of the registration method is assessed by generating a pair of synthetic images with known matching points and comparing the actual matching points with the ones from our method. Three types of geometric transformation are applied to simulate different effects of distortions and the results show good promise to register images with both global and local distortions.

The quantification of the visco-elastic properties (resistance (R), inertia (L) and compliance (C)) of the different chest wall compartments (pulmonary rib cage,diaphragmatic rib cage and abdomen) is important to study the status of the passive components of the respiratory system, particularly in selected pathologies. Applying the viscoelastic-electrical analogy to the chest wall, we used an identification method in order to estimate the R, L and C parameters of the different parts of the chest, basing on different models; the input and output measured data were constituted by the volume variations of the different chest wall compartments and by the nasal pressure during controlled intermittent positive pressure ventilation by nasal mask, while the parameters of the system (R, L and C of the different compartments) were to be estimated. Volumes were measured with a new method, recently validated, based on an opto-electronic motion analyzer, able to compute with high accuracy and null invasivity the absolute values and the time variations of the volumes of each of the three compartments. The estimation of the R, L and C parameters has been based on a least-squared criterion, and the minimization has been based on a robustified iterative Gauss-Newton algorithm. The validation of the estimation procedure (fitting) has ben performed computing the percentage root mean square value of the error between the output real data and the output estimated data. The method has been applied to 2 healthy subjects. Also preliminary results have been obtained from 20 subjects affected by neuromuscular diseases (Duchenne Muscular Dystrophy (DMD) and Spinal Muscle Atrophy (SMA)). The results show that: (a) the best-fitting electrical models of the respiratory system are made up by one or three parallel RLC branches supplied by a voltage generator (so considering inertial properties, particularly in the abdominal compartment, and not considering patient/machine connection); (b) there is a significant difference between DMD and SMA groups (the value of resistance and rigidity of the thorax is much higher in SMA patients); (c) the inclusion of the connection patient-ventilator make the models ill-conditioned. We conclude that this method allows a quantitative evaluation of rib cage and abdominal passive characteristics with a good accuracy and through a dynamic measurement and that it could give significant data in physiology and clinics.

The purpose is to integrate time-attenuation curves from Electron-Beam CT with flow-time curves from spirometry in the analysis of airflow obstruction. A pressure-sensitive switch was connected between a spirometer mouthpiece and a modified EBCT scanner keyboard. The onset of expiratory flow causes pressure changes which simultaneously trigger EBCT and spirometric acquisitions. Subjects performed a forced expiratory maneuver, during which EBCT images of the lung were obtained every 500 ms using 130 kVp, 630 mA, 100 ms scan time and 3 mm collimation. From EBCT images, time-attenuation curves were generated for each of three zones (non-dependent, middle and dependent lung) using small ROIs (12 mm²) placed over approximately the same anatomic regions of lung. The resulting time- attenuation curves and flow-time curves were then superimposed. Two normal subjects, two subjects with emphysema and three lung transplant subjects have been studied to date. In normal subjects, lung attenuation increases steadily during the first 4 - 6 seconds of expiration, whereas in patients with emphysema, lung attenuation was relatively constant over the course of expiration. Lung transplant subjects show both of these characteristics--normal characteristics for the transplant lung and emphysematous characteristics for the native lung. Lung transplant subjects may also demonstrate some dynamics between transplant and diseased lung. Respiratory-triggered EBCT can be used to simultaneously acquire time-attenuation and flow-time data. This has been used to characterize dynamic airflow patterns in patients with respiratory disease.

We are using the unique features of electron beam CT (EBCT) in conjunction with respiratory and cardiac gating to explore the use of non-radioactive xenon gas as a pulmonary ventilation contrast agent. The goal is to construct accurate and quantitative high-resolution maps of local pulmonary ventilation in humans. We are evaluating xenon-enhanced computed tomography in the pig model with dynamic tracer washout/dilution and single breath inhalation imaging protocols. Scanning is done via an EBCT scanner which offers 50 msec scan aperture speeds. CT attenuation coefficients (image gray scale value) show a linear increase with xenon concentration (r equals 0.99). We measure a 1.55 Hounsfield Unit (HU) enhancement (kV equals 130, mA equals 623) per percentage increase in xenon gas concentration giving an approximately 155 HU enhancement with 100% xenon gas concentration as measured in a plexiglass super-syringe. Early results indicate that a single breath (from functional residual capacity to total lung capacity) of 100% xenon gas provides an average 32 +/- 1.85 (SE) HU enhancement in the lung parenchyma (maximum 50 HU) and should not encounter unwanted xenon side effects. However, changes in lung density occurring during even short breath holds (as short as 10 seconds) may limit using a single breath technique to synchronous volumetric scanning, currently possible only with EBCT. Preliminary results indicate close agreement between measured regional xenon concentration-time curves and theoretical predictions for the same sample. More than 10 breaths with inspirations to as high as 25 cmH₂O airway pressure were needed to clear tracer from all lung regions and some regions had nearly linear rather than mono-exponential clearance curves. When regional parenchymal xenon concentration-time curves were analyzed, vertical gradients in ventilation and redistribution of ventilation at higher inspiratory flow rates were consistent with known pulmonary physiology. We present here a works in progress, showing results from two pigs illustrating the high resolution and detailed regional information obtainable with careful attention to cardiac and respiratory gating during a multi-breath washout period.

It was the goal of this study to see if relatively noninvasive CT studies could provide a quantitative index of acute lung rejection in single lung transplantation. Using volume scanning fast CT, the change in cross-sectional area of the major pulmonary arteries from systole to diastole, regional lung perfusion and ventilation was measured in 12 dogs with left lung allotransplantation before and during rejection and four dogs with left lung autotransplantation. All dogs were anesthetized and scanned in a fast computed tomography scanner (dynamic spatial reconstructor--DSR) during several ventilatory cycles and again during injection of contrast medium into the right atrium. There was significant reduction of regional air content, ventilation, perfusion and pulmonary artery compliance during rejection of the transplanted lung. The severity of these changes related linearly with the histological indices of rejection. It is concluded that minimally invasive dynamic CT imaging of transplanted lung can be used to detect acute rejection and its severity.

A robust, automated technique has been developed for estimating total lung volumes from chest computed tomography (CT) images. The technique includes a method for segmenting major chest anatomy. A knowledge-based approach automates the calculation of separate volumes of the whole thorax, lungs, and central tracheo-bronchial tree from volumetric CT data sets. A simple, explicit 3D model describes properties such as shape, topology and X-ray attenuation, of the relevant anatomy, which constrain the segmentation of these anatomic structures. Total lung volume is estimated as the sum of the right and left lungs and excludes the central airways. The method requires no operator intervention. In preliminary testing, the system was applied to image data from two healthy subjects and four patients with emphysema who underwent both helical CT and pulmonary function tests. To obtain single breath-hold scans, the healthy subjects were scanned with a collimation of 5 mm and a pitch of 1.5, while the emphysema patients were scanned with collimation of 10 mm at a pitch of 2.0. CT data were reconstructed as contiguous image sets. Automatically calculated volumes were consistent with body plethysmography results (< 10% difference).

We are developing new image processing methods to combine a sequence of x-ray angiographic images into a single `stacked' output image with optimal visualization of fully opacified arteries. In this preliminary report, we use simulation to optimize acquisitions and image stacking methods for efficient use of contrast agent and x-ray dose. Using a Monte Carlo approach, we couple a convective-dispersive model of bolus transport to a model of x- ray image acquisition. We investigate three image processing algorithms: maximum intensity projection (MIP), matched filtering (MAT), and finite impulse response (FIR) and compare them to the unstacked (UnST) result. Since MAT is not practically applied, we have developed FIR which is a practical approximation to MAT. Comparing output images of equivalent CNR to the unstacked case with 100% contrast agent volume, FIR can reduce the contrast material dose by approximately 65% with the same x-ray exposure and frame rate. FIR is the preferred image processing algorithm because of its flexibility and high CNR result.

In this paper, we describe an approach to 3D reconstruction of the coronary tree based on combined use of biplane coronary angiography and intravascular ultrasound (IVUS). Shortly before the start of a constant-speed IVUS pullback, radiopaque dye is injected into the examined coronary tree and the heart is imaged with a calibrated biplane X-ray system. The 3D centerline of the coronary tree is reconstructed from the geometrically corrected biplane angiograms using an automated segmentation method and manual matching of corresponding branching points. The borders of vessel wall and plaque are automatically detected in the acquired pullback images and the IVUS cross sections are mapped perpendicular to the previously reconstructed 3D vessel centerline. In addition, the twist of the IVUS probe due to the curvature of the coronary artery is calculated for a torsion-free catheter and the whole vessel reconstruction is rotationally adjusted using available anatomic landmarks. The accuracy of the biplane reconstruction procedure is validated by means of a left coronary tree phantom. The feasibility of the entire approach is demonstrated in a cadaveric pig heart.

Electron beam computed tomography (EBCT), also known as ultrafast-CT or cine-CT, uses a unique scanning architecture which allows for multiple high spatial resolution electrocardiographic triggered images of the beating heart. A recent study has demonstrated the feasibility of qualitative comparisons between EBCT derived 3D coronary angiograms and invasive angiography. Stenoses of the proximal portions of the left anterior descending and right coronary arteries were readily identified, but description of atherosclerotic narrowing in the left circumflex artery (and distal epicardial disease) was not possible with any degree of confidence. Although these preliminary studies support the notion that this approach has potential, the images overall were suboptimal for clinical application as an adjunct to invasive angiography. Furthermore, these studies did not examine different methods of EBCT scan acquisition, tomographic slice thicknesses, extent of scan overlap, or other segmentation, thresholding, and interpolation algorithms. Our laboratory has initiated investigation of these aspects and limitations of EBCT coronary angiography. Specific areas of research include defining effects of cardiac orientation; defining the effects of tomographic slice thickness and intensity (gradient) versus positional (shaped based) interpolation; and defining applicability of imaging each of the major epicardial coronary arteries for quantitative definition of vessel size, cross-sectional area, taper, and discrete vessel narrowing.

Typically, quantitation of 2D angiographic data has been limited to assessment of luminal edges from a single or, at most, two views. Use of densitometric image information, which can give data more closely related to the 3D geometry of the vessel, has been largely ignored because of imaging problems associated with conventional contrast agents. Recently, in phantom studies, we have demonstrated that the use of CO2 as an angiographic contrast agent may eliminate many of the obstacles previously limiting the use of densitometric approaches to quantitating 2D angiograms. This may allow us to provide an evaluation more closely related to the 3D geometry of the vasculature of interest. In this paper, we discuss our approach to CO2 angiography and provide illustrative angiographic studies demonstrating the potential breakthrough CO2 angiography may provide in extracting 3D information from 2D data sets.

The primary focus of speech production research is directed towards obtaining improved understanding and quantitative characterization of the articulatory dynamics, acoustics, and cognition of both normal and pathological human speech. Such efforts are, however, frequently challenged by the lack of appropriate physical and physiological data. A great deal of attention is, hence, given to the development of novel measurement/instrumentation techniques which are desirably non invasive, safe, and do not interfere with normal speech production. Several imaging techniques have been successfully employed for studying speech production. In the first part of this paper, an overview of the various imaging techniques used in speech research such as x-rays, ultrasound, structural and functional magnetic resonance imaging, glossometry, palatography, video fibroscopy and imaging is presented. In the second part of the paper, we describe the results of our efforts to understand and model speech production mechanisms of vowels, fricatives, and lateral and rhotic consonants based on MRI data.

As part of an analysis by synthesis approach to studying vocal intensity control in falsetto register, volumetric imaging of the vocal tract (the upper airway from the glottis to the lips) using electron beam computed tomography was performed on a classically trained singer, a countertenor, who uses a falsetto singing technique. Eight pitch and loudness conditions were imaged, a subset of which will be presented here. Each set of scans consisted of contiguous 3 mm axial `slices' encompassing the arch of the hard palate superiorly and the first tracheal ring inferiorly. Images were analyzed in three stages: image segmentation, 3D airway reconstruction and airway measurement. The vocal tract airway was segmented from surrounding tissue by assigning airway voxels a unique gray scale value. Reconstruction of the vocal tract in three dimensions was accomplished using shape based interpolation on the segmented images. Cross-sectional areas and vocal tract length were acquired from shape based interpolated data. Vocal tract area functions derived from these measurements were used to simulate the subject's phonations, which in turn allowed estimation of glottal and supraglottal contributions to vocal intensity.

Horizontal supraglottic laryngectomy is a surgical procedure to remove a mass lesion located in the region of the pharynx superior to the true vocal folds. In contrast to full or partial laryngectomy, patients who undergo horizontal supraglottic laryngectomy often present with little or nor involvement to the true vocal folds. This population provides an opportunity to examine the acoustic consequences of altering the pharynx while sparing the laryngeal sound source. Acoustic and magnetic resonance imaging (MRI) data were acquired in a group of four patients before and after supraglottic laryngectomy. Acoustic measures included the identification of vocal tract resonances and the fundamental frequency of the vocal fold vibration. 3D reconstruction of the pharyngeal portion of each subjects' vocal tract were made from MRIs taken during phonation and volume measures were obtained. These measures reveal a variable, but often dramatic difference in the surgically-altered area of the pharynx and changes in the formant frequencies of the vowel/i/post surgically. In some cases the presence of the tumor created a deviation from the expected formant values pre-operatively with post-operative values approaching normal. Patients who also underwent radiation treatment post surgically tended to have greater constriction in the pharyngeal area of the vocal tract.

Videostroboscopy is an examination method during which a video-recording of the moving vocal folds is obtained. This examination, is very important because it yields a permanent record of the moving vocal folds and it allows the diagnosis of abnormalities which contribute to voice disorders. In this paper we use image processing algorithms in order to find and quantify the motion of the vocal folds during phonation. In order to achieve this, a new approach for tracking of the vocal folds is presented. More specifically, an active contours (snakes) based procedure is developed to automatically, with minimum user intervention, delineate the contour of the vocal folds in each frame of the videostroboscopic image sequence. After this delineation, an elastic registration algorithm is used to find the motion of the vocal folds between adjacent frames of a video sequence. For this purpose, a new simulated annealing based algorithm is used to match elastically the vocal fold contours from frame-to-frame. After the motion of each point of the vocal folds has been estimated, an affine model is used to parameterize the motion from frame-to-frame. Least-squares estimation is used to fit this model. The parameters of such a model (rotation, translation, and deformation along two principle axes) will hopefully help quantify and discriminate videostroboscopic recordings based on the motion present.

Vocal tract imaging for the vowels /i/ and /a/ using both EBCT and MRI was carried out for one subject (29 yr old male, native of midwestern United States) using an Imatron C-150 electron beam CT scanner and a GE Signa 1.5 Tesla scanner, respectively. Each image set was analyzed using a general display and quantitation package called VIDA^TM (Volumetric Image Display and Analysis). The image analysis consisted of segmenting the airspace from the surrounding tissue, obtaining a 3D vocal tract shape via shape based interpolation, and finally using an iterative bisection algorithm to determine the vocal tract area function. The results show that the 3D representations of the vocal tract shapes derived from EBCT show subtle deformations of the airway by articulatory structures and teeth that are not observed in the MRI based representations. Shaded surface renderings of each vocal tract shape and for each imaging technique are shown and the apparent trade-offs between the two imaging methods are discussed.

Evaluation of most normal and patho-pulmonary physiology has relied upon indirect measures of pulmonary function which yield global estimates of underlying structural and functional deficits, which are usually very heterogeneous in nature. Early signs of disease are not recognizable by these techniques, nor are they usually recognizable by the manifestation of physical symptoms. As X-ray CT technology has improved, imaging has held a promise to provide the detailed information here-to-fore missing in standard pulmonary function evaluations. A full solution to the imaging and analysis problem requires true dynamic volumetric approaches to facilitate tracking the lung through space during respiratory maneuvers, and following the radiopacified blood and airflow tracers as they pass through the pulmonary vascular bed or wash in or out of the alveolar air spaces. However, high- resolution, high-speed, stacked single-slice approaches for lung imaging have brought the state-of-the-art to a point where quantitative airway evaluation can play an important role in the study of lung disease and normal lung physiology, if one limits the evaluation to those airway segmented sliced in true cross-sections, or to the evaluation of those airway segments for which a true cross-sectional image can be reformatted from the original stacked sections. This paper presents a software system, called ASAP (for Airway Segmentation and Analysis Program), which provides a rapid, minimally-interactive method for objective identification of airway borders and the reporting of associated geometric measures of diameters and wall thicknesses. We demonstrate that this system yields highly reproducible results both within and between observers, and quantitative measures are accurate to within the resolution of the scanner when phantoms of known geometry are evaluated. Results included here demonstrate that the well-accepted half-max criteria for border definition is a rough approximation, which when applied to structures such as intrathoracic airways yields incorrect results. Our analysis shows that the inner and outer wall detection thresholds must be customized based upon the size of the structure of interest.

Major airway obstruction (trachea, right and left main bronchi) is an important cause of morbidity and mortality. Management requires adequate assessment of the position, extent and severity of the obstructing or stenotic segment. The objective of this study was to evaluate 3D reconstruction of the major airways using volumetric image display and analysis (VIDA), in subjects with major airflow obstruction. We have evaluated five subjects with major airway obstruction using Electron Beam Computed Tomography (EBCT) with a contiguous 3 mm slice thickness at total lung capacity. The digital information was transferred to a Sun Workstation (SPARC 5) for data analysis using VIDA. From this data set, the airway dimensions were calculated using a method for airway centerline determination and slice reformatting so as to section the airway perpendicular to its local long axis. Once appropriately sectioned, a number of different methods were used in edge finding. The airways were also presented as a surface rendered 3D image in either still or movie format. Finally, all subjects underwent flexible bronchoscopy to assess the abnormalities by direct visualization, with results of the bronchoscopic assessment being compared to the VIDA measurements. In all subjects, the volumetric image display and analysis gave anatomically correct and detailed images, which could be accurately measured. This information enabled appropriate pre-planning of operative corrective procedures, that included laser therapy, stent placement and balloon bronchoplasty. We conclude that the volumetric image display and analysis provides useful and reliable information for the management of major airflow obstruction.

Standard airway evaluation by oropharyngeal visualization, neck and jaw mobility tests and radiological assessment does not identify all patients at risk for failed intubation. Our objective in this study was to determine if acoustic reflection measurement of the upper airway can identify patients who are difficult to intubate and discriminate between them and the readily intubated population. We evaluated adults having a documented history of unexpected failed (n equals 14) or successful (n equals 14) endotracheal intubation using six combinations of body (sitting, supine) and neck (flexed, neutral, extended) position. Standard airway assessment revealed no differences between the study groups. Results of acoustic reflection analysis revealed multiple position dependent differences between the study groups. Absolute criteria to discriminate between the two groups were found with the patients laying in supine position with the neck extended. All patients who had been successfully intubated had pharyngeal volumes greater than 43.4 ml (mean +/- standard deviation, 55.0 +/- 8.8 ml), whereas pharyngeal volumes were less than 37.5 ml in all patients who had unexpected failed intubation (mean +/- standard deviation, 18.9 +/- 10.6 ml). This difference was significant (p < 0.05, Two Way ANOVA with post-hoc t testing). Acoustic reflection was 100% sensitive and 100% specific at distinguishing between patients who experienced unexpected failed intubation and those who had successful intubation. This potential screening test is fast (2 - 3 minutes), simple to perform and requires minimal patient cooperation.

We investigated the use of the Generalized Linear Least Squares (GLLS) method for fast estimation of myocardial blood flow (MBF) with N-13 ammonia Positron Emission Tomography (PET). GLLS was based on a high order integral equation converted from the state variable differential equations describing the kinetics of a PET tracer. Two error sources, spillover and the measurement noise, were studied. The estimation of the spillover coefficients between plasma time activity curve (TAC) and tissue TAC was incorporated into the GLLS. The GLLS procedure was modified accordingly. It was found, in computer simulation, that spillover correction incorporated GLLS provided as reliable MBF estimation as the model fitting method accounting for spillover. Since the linear kinetic model relating the plasma TAC to the tissue TAC was equivalent to the one relating the integral of plasma TAC (accumulated counts) to the integral of the tissue TAC, the GLLS method could be directly applied to the accumulated PET counts. It was found that the direct use of the accumulated counts reduced the random fluctuation observed in TAC data from PET images; and that this noise reduction significantly improved the accuracy of estimated MBF.

Current positron emission tomography techniques for the measurement of cerebral blood flow assume that voxels represent pure material regions. In this work, a method is presented which utilizes anatomical information from a high resolution modality such as MRI in conjunction with a multicompartment extension of the Kety model to obtain intravoxel, tissue specific blood flow values. In order to evaluate the proposed method, noisy time activity curves (TACs) were simulated representing different combinations of gray matter, white matter and CSF, and ratios of gray to white matter blood flow. In all experiments it was assumed that registered MR data supplied the number of materials and the fraction of each present. For each TAC, three experiments were run. In the first it was assumed that the fraction of each material determined by MRI was correct, and, in the second two, that the value was either too high or too low. Using the tree annealing method, material flows were determined which gave the best fit of the model to the simulated TAC data. The results indicate that the accuracy of the method is approximately linearly related to the error in material fraction estimated for a voxel.

In this study, we applied factor analysis of dynamic structures (FADS) techniques in dynamic PET images to (1) extract the arterial time activity curve (TAC) from human adult or small monkey dynamic FDG PET; and (2) investigate the use of FADS generated factor images and factor curves to detect large subject movements during dynamic GA-68 EDTA scans of brain tumor patients. The results showed that a blood sample constraint improved the accuracy of FADS technique in extracting the `pure' blood pool TAC from dynamic PET studies that have large spillover problems. The studies of GA-68 EDTA brain PET studies showed that three major factors were extracted from images using FADS. For studies with little patient movement, a standard pattern of three factor curves and three factor images were obtained. However, large patient movement changed the outcomes of FADS results. We conclude that (1) FADS technique with a blood sample allows the extraction of the `pure' blood pool TAC directly from quantitative PET images without requiring multiple blood samples, region-of- interest drawing or spillover correction; and (2) FADS technique provides a sensitive way to detect large patient movements in dynamic PET studies.

To study the effects of RMP7, an bradykinin analog, on BBB permeability changes in brain tumor patients, Ga-68 EDTA PET was used for quantitative determination of tumor tissue permeability, two dynamic Ga-68 EDTA PET scans (with and without RMP7) were performed for each patient. Changes between the results of the two scans were used to assess the effect of the drug. MRI was also performed to provide the anatomical information. Patient movement during the dynamic PET studies were corrected by frame-to-frame PET image registration. Due to different orientation among PET and MRI images, all PET images were co-registered to the MRI images of the corresponding patient. All frames of a dynamic scan were summed before registering to the MRI images. The main features for registration were boundaries of skull and tumor. After registration, region of interest (ROI) was defined on MRI images and was copied to registered PET dynamic images. The whole tumor radioactivity was calculated based on each plane's tumor ROI. The transport constant from plasma to tissue (Ki) that is related to BBB permeability was estimated by a two-compartmental model. The PET-MRI registration method developed by Lin et al was found to work well for Ga-68 PET-MRI image registration. Estimated whole tumor K_i was 0.0002 +/- 0.0019 (s.d.)ml/min/g for the baseline study and was increased by 46 +/- 42 (s.d.)% with RMP7.

A tissue database was established by using multidimensional clusters of exact longitudinal (T₁) and transversal (T₂) relaxation times and spin density, allowing the automatic segmentation and characterization of healthy and pathologic tissue. All parameters were simultaneously acquired employing a modified Multi-Echo pulse sequence. Initial clinical results showed a good differentiation between normal brain tissue and pathologic tissue like edema and meningioma. Inhomogeneous tumors such as high-grade glioblastoma were difficult to characterize automatically. The implementation of a diffusion-weighted modified Tanner-Stejskal pulse sequence allows the acquisition of the Apparent Diffusion Constant (ADC), which has been incorporated for the first time into a multidimensional information set as a new tissue-characterizing parameter. This parameter is sensitive to changes in the mobility of water in and between different cell compartments resulting from metabolic cell disorders like ischemic or edematous processes. To reproduce the known results of animal experiments, where as early as 30 min after an ischemic event the measurement of the ADC led to a diagnosis, diffusion-weighted imaging had to be implemented on a standard clinical scanner. The correction of unavoidable motion artifacts, which occur when applying diffusion- weighted spin echo sequences on standard clinical scanners, require the implementation of a special sequence using the navigator echo method followed by a correction algorithm of the raw data in Fourier space. Initial results showed a significant improvement in the differentiation of healthy and pathologic tissue classes.

We have developed a software package which combines the simulation and analysis of dynamic brain images acquired by positron emission tomography. In the spatial domain, standard digitized phantoms are used to create spatial maps which represent various structures in the brain. In the temporal domain, kinetic models are used to simulate the dynamics at the pixel level for each time-activity curve. Physical characteristics such as spatial resolution and noise are then modeled on individual image frames. This software has been used in the case of [18F]fluorodeoxyglucose studies to produce realistic simulations which were further analyzed to confirm previous findings based on real clinical data.

The purpose of this presentation is to report our image registration and analysis experience with a large-array biomagnetometer for magnetoencephaography (MEG) and a high resolution positron emission tomography (PET) scanner in the presurgical localization of spike sources of intractable epilepsy patients and in studying the relationship between metabolic and physiologic abnormalities detected in regions of interest during the evaluation. Findings of 12 epilepsy patients who underwent MEG and PET studies at UCSF are discussed.

Though excellent spatial resolution (on the order of 1 mm) is obtainable in functional MRI (fMRI), its temporal resolution is limited to about 1 second by hemodynamics. On the other hand, magnetoencephalography (MEG) and electroencephalography (EEG) provide millisecond temporal resolution but a relatively crude (on the order of 1 cm) spatial resolution to localized sources. Thus, techniques that could combine the high temporal resolution of MEG or EEG with the high spatial resolution of fMRI would be of great significance in imaging the spatiotemporal distribution of neuronal activation. With the ultimate objective of combining fMRI and EEG activation studies, we have conducted experiments to determine how pixels activated in fMRI correlate with underlying EEG sources in a given subject during visual stimulation. Results of a three-subject study suggest good correlation between the center-of-gravity of activated pixels seen in fMRI and the center-of-gravity of regions localized through EEG measurements.

Various biomedical imaging sensors, including ElectroMagnetic Tomography, are being combined to study, assess, and localized neurological (dys)function. The interest for this combination stems from the broad variety and complementarity of information brought out by (functional-) Magnetic Resonance Imaging, Magnetic Resonance Spectroscopy, Computed Tomodensitometry, Single Photon Emission Tomography, Positron Emission Tomography and ElectroMagnetic Tomography. Besides allowing morphology, metabolism and function to be studied simultaneously, this complementarity is also expected to show best when studying pathologies reflected by metabolic or electromagnetic dysfunctions. An example of clinical application for epilepsy assessment and surgery planning is presented, along with suggestions for further potential developments.

In this investigation, motion analysis is carried out on an ultrasonic image sequence of atheromatous carotid plaque using the block matching algorithm. Each image is partitioned into a number of fixed size blocks. Fractal dimension is then estimated in each block of the current frame and used as the feature parameter. The position of the corresponding block in the previous frame of the image sequence is identified based on the similarity of that parameter. A novel Two-Pass searching scheme is employed to look for the best matching block. Fractal dimension has the advantage of relatively invariant to changes in scale and intensity. Also, it demands much lesser computation time than other conventional statistical parameters. The algorithm has been used to track the movement of a carotid plaque being captured in an ultrasonic image sequence. Its performance is compared with the results obtained in a previous study.

The purpose of this study was to develop a valid, reliable and accurate system of measurement of abdominal aortic aneurysms, using volumetric analysis of x-ray computed tomographic data. This study evaluates illustrative cases, and compares measurements of AAA phantoms, using standard 2D versus volumetric methods. To validate the volumetric analysis, four phantom aneurysms were constructed in a range of diameters (4.5 - 7.0 cm) which presents the greatest management challenge to the clinician. These phantoms were imaged using a Toshiba Xpress SX helical CT. Separate scans were obtained at conventional (10 mm X 10 mm) and thin slice (5 mm X 5 mm) collimations. The thin slices were reconstructed at 2 mm intervals. Data from each of the 96 scans were interpreted using a standard 2D approach, then analyzed using task-oriented volumetric software. We evaluate patient assessments, and compare greatest outer diameters of phantoms, by standard versus volumetric methods. Qualitative differences between solutions based on standard versus volumetric analysis of illustrative patient cases are substantial. Expert radiologists' standard measurements of phantom aneurysms are highly reliable (r² equals 0.901 - 0.958; p < 0.001), but biased toward significant overestimation of aneurysm diameters in the range of clinical interest. For the same phantoms, volumetric analysis was both more reliable (r² equals 0.986 - 0.996; p < 0.001), and more accurate, with no significant bias in the range of interest. Volumetric analysis promotes selection of more valid management strategies, by providing vital information not otherwise available, and allowing more reliable and accurate assessment of abdominal aortic aneurysms. It is particularly valuable in the presence of aortic tortuosity, vessel eccentricity, and uncertain involvement of critical vessels.

Variations in blood volume in the myocardium through the cardiac cycle have previously been considered constant. More recent studies have indicated a considerably variation from end diastole to end systole. These studies were nearly all performed under non-physiological conditions using muscle preparations or ex situ cardiac preparations. This study was designed to assess the dynamic changes of the intramyocardial blood volume in the intact animal under normal flow conditions using single photon emission computed tomography (SPECT). Radiolabeled, 15 micrometers diameter, microspheres were emoblized in the myocardial microcirculation of dogs with subsequent scans in a TRIAD single-photon-emission-computed- tomography scanner. Gated images were obtained at 63 msec intervals encompassing the entire heart. Transmural voxel (equals volume element) brightness was measured in all tomographic images reflecting global and regional count density in the myocardium. There was a significant decrease in the blood volume from end diastole to end systole (10.8 cc/100 mL muscle volume; p < 0.00001). The decrease from diastole (ED) to systole (ES) in image brightness at the apex, mid ventricle, and base were: -5.7% (p < 0.01, apex vs. base), -4.7% (p < 0.01, mid ventricle vs. base) and +2.2%, respectively. Conclusions: (1) respiratory and ECG gated SPECT images allow measurement of intramyocardial blood volume changes throughout the cardiac cycle in the intact animal; (2) myocardial blood content is maximum at ED; (3) these findings progressively diminished in magnitude from apex to base.

In this paper, we describe a method for the reconstruction of the surface of the left ventricle from a set of lacunary data (that is an incomplete, unevenly sampled and unstructured data set). Global models, because they compress the properties of a surface into a small set of parameters, have a strong regularizing power and are therefore very well suited to lacunary data. Globally deformable superquadrics are particularly attractive, because of their simplicity. This model can be fitted to the data using the Levenberg-Marquardt algorithm for non-linear optimization. However, the difficulties we experienced to get temporally consistent solutions as well as the intrinsic 4D character of the data led us to generalize the classical 3D superquadric model to 4D. We present results on a 4D sequence from the Dynamic Spatial Reconstructor of the Mayo Clinic, and on a 4D IRM sequence.

Medical imaging scanners now exist that can generated 4D cardiac images (time sequences of 3D volumes). Since the heart is an organ that exhibits motion, examining its image characteristics with a 4D image can give useful information about its condition. For multi- dimensional image segmentation, semi-automatic methods have many advantages over manual segmentation. This paper describes a procedure for performing semi-automatic image segmentation and analysis upon a 4D cardiac image. This procedure involves the input of user- defined information (cues) at certain time points of the sequence. These cues are then automatically interpolated or extrapolated for the remaining time points. The analysis system interprets the completed sequence of cues to generate a list of image processing functions that can subsequently segment and analyze the 4D image. This paradigm has been implemented using INTERSEG, an existing 3D cue-based analysis system. This cue-based analysis procedure permits 4D cardiac image segmentation with a small amount of user interaction. Performance of the proposed 4D image analysis system compares favorably to results generated by defining cues on each individual volume, as well as manual techniques. Further, the 4D approach requires significantly less interaction time than a 3D-only approach.

This paper presents three approaches to the problem of obtaining the left ventricular boundaries from cardiac MR data. The first presents a new model based approach for the detection of the endocardium from 4D MR cardiac images. The method proposed here links shape modeling and edge detection to provide a compact representation of the endocardium. A spatio-temporal edge detector has been designed to incorporate the temporal information available in 4D images. This edge detector has a stronger response to dynamic edges than static edges. Since the ventricle is a dynamic shape, boundaries detected using this edge detector are far better than those detected using a spatial edge detector. The output of our edge detector is iteratively corrected using a spherical harmonic model. This model based approach allows us to overcome the problems of noise and missing boundary information. Our system is fully automated and its output consists of the extracted boundary in each slice and a 3D surface model for each time instant. Quantitative evaluation is done by comparing the results of the algorithm with manually extracted ground truth for 12 data sets. The second approach uses filters applied across the detected tag lines to remove the tags from SPAMM-tagged MR data to allow existing boundary detection algorithms to function with minimal changes. The third approach uses the Fuzzy c-Spherical Shell algorithm directly on tagged (and untagged) data to determine the approximate LV center.

Clinical and experimental data suggest that delayed reperfusion of the infarct related artery may limit infarct expansion without increasing myocardial salvage. In order to assess the potential mechanisms involved, an acute closed chest canine model of myocardial infarction and delayed reperfusion was studied. Nineteen dogs underwent 3D computed tomography in the Dynamic Spatial Reconstructor (a fast, volume imaging, CT scanner) at baseline and three and four hours later to estimate left ventricular chamber volumes, global distensibility and regional myocardial stiffness. A control group was scanned without intervention. An occlusion group underwent four hours of coronary artery occlusion. A reperfusion group underwent three hours of coronary artery occlusion followed by one hour of reperfusion. Similar infarct sizes were seen in the occlusion and reperfusion groups. Globally reperfusion was associated with increased left ventricular end diastolic pressure and prolongation of global relaxation. Regionally reperfusion was associated with increased myocardial stiffness, intramyocardial blood volume and wall thickness within the infarct zone relative to the not reperfused myocardium.

Our work focuses on functional heart analysis during acute myocardial infarction based on time-sequence data derived with a high-resolution ECG technique. This data stream can be interpreted as a sequence of potential deviation images. The analysis is performed by both visualizing the potential deviation onto the thorax as well as by shape analysis of the underlying ECG signals using a topologic map. The algorithm deals with the measurement of similarity between different pathological signal types. In contrast to other techniques, the whole ECG signal, coded as a feature vector, is used as input for the self-organizing map. The results show that this approach is suitable for handling unsharp class transitions common to the medical domain.

The left ventricular (LV) wall motion is the most challenging and interesting task in cardiac evaluation. In this paper, an integrated system that measures and displays left ventricular wall motion is presented. Based on the 3D reconstruction of ventricle from nine rotational cross- sectional images acquired with multiplane transesophageal echocardiography (TEE), a quantitative and visual expression of the motion of LV is presented. Nine images were obtained with the transducer rotating around a central axis passing through LV. A sequence of image processing operations have been developed for detecting left ventricular boundaries from TEE images obtained with different angle in a whole cardiac cycle. The algorithm which integrates 2D boundary information into 3D volume representation is designed based on automata theory. The phantom study for computing the scaling factors between the image metrics and the physical metrics shows a good correlation between the computed results and the specimens in the in vitro study. Finally, the 3D shape visualization of the reconstructed moving ventricle is presented. The performance of proposed experiments shows good feasibility of the new application of TEE in cardiac evaluation.

Echocardiography is used for detection and analysis of heart dysfunction. Different techniques have been used to quantify wall motion, but so far no gold standard has emerged. The purpose of this study is to improve the echocardiographic quantitative analysis for measurement of visualization of ventricular wall motion abnormalities. The endocardial wall is detected using active contour techniques. By using distance transform algorithms, 2D velocity vectors at every point on the endocardial border are calculated. The vectors represent the radial wall motion, and they are visualized by a color overlay on the endocardial tissue image. Quantitative information about wall motion can be extracted at every contour point. Anatomical M-mode images are generated from the same contour points in order to validate the quantitative velocity information. The presented technique improves existing methods by visualizing wall motion real-time giving the user phase information. This is helpful in localizing regional wall motion dysfunction. The velocity values are calculated from the digital scanner image data, and the use of the high frame-rate capabilities of modern scanners gives the application high sensitivity.

In 2D echocardiography, wall motion analysis is important in detection of heart dysfunction. The methods used suffer from being only approximations due to the complex movement of the heart. The purpose of our study is to use 3D ultrasound in order to improve the quantitative analysis of the left ventricular wall motion. The wall motion is calculated using the 3D Euclidean distance transform, resulting in a set of vectors normal to the endocardial surface. The endocardial wall velocity values are color coded and the surface is rendered creating a cincloop of the left ventricle with color intensities indicating the wall motion. The 3D motion coding technique allows for a precise quantitative analysis of the heart function. Infarcted areas are shown to be marked out with different color intensities due to the muscle activity reduction in that region. The 3D reconstruction improves the diagnosis by visualizing and localizing the whole dysfunctional region in relation to the rest of the ventricular structure.

The purpose of this work is to develop a method of estimation of the field of vector velocity to evaluate motion of the heart in view of an assistance with the medical diagnosis. This method is based on the application of a new functional of motion estimation for density images like X- ray CT. Assuming that the images are proportional to some conserved quantity, a new penalty function is defined for motion estimation of a deforming body. We use the theory of linear elasticity to propose a new constraining term. We choose to solve the penalty function with a finite element method. Examples of experiments using simulated images of deforming body are presented. The method is able to take into account compressible or incompressible motion according to the parameter's values.

In the present work, distinct structures appearing in biomedical images are modeled as fractals. Within an image, the relevant structures are associated to a fractal dimension. Changes in the dimension values, as a function of time, reflect alterations of structural properties. Accurate and robust estimation of this dimension, leads to a precise characterization of changes undergone by the structure. The Continuous Pyramidal Alternating Sequential Filter method is proposed as a robust and accurate fractal dimension estimator. A study on bedrest data of human subjects was conducted. Bedrest is an accepted model for the study of osteoporosis. Here the spine is modeled as a fractal structure. Fractal model were also applied towards analysis of breast cancer and brain tumors. Results from these different studies confirm that fractals can suitably model a variety of biological structures. These studies also suggest that fractal models can be effectively utilized to detect temporal changes undergone by the structures.

Tissue characterization using texture analysis is gaining increasing importance in medical imaging. We present a completely automated method for discriminating between normal and emphysematous regions from CT images. This method involves extracting seventeen features which are based on statistical, hybrid and fractal texture models. The best subset of features is derived from the training set using the divergence technique. A minimum distance classifier is used to classify the samples into one of the two classes--normal and emphysema. Sensitivity and specificity and accuracy values achieved were 80% or greater in most cases proving that texture analysis holds great promise in identifying emphysema.

This paper proposes an approach to advance the utility of physical modeling techniques for medical applications by correlating finite element based models with the mechanical anatomy characteristic of a clinical patient. A methodology is presented to model the patient-specific mechanical response of brain tissue in vivo. The resultant model is parameterized in terms of clinical CT and MRI imaging sequences acquired for each patient. Applications of the proposed technique to the areas of brain tumor growth modeling and predicting tissue shifts during stereotactic neurosurgery, are described. Results are presented for an implementation of our approach to the problem of predictive brain tumor modeling.

In this study, we designed and implemented a temporal image database for outcome analysis of lung nodules based on spiral CT images. The software package is composed of three parts. They are, respectively, a database management system which stores patient image data and nodule information; a user-friendly graphical user interface which allows a user to interface with the image database; and image processing tools that are designed to segment out lung nodules in the CT image with a simple mouse click anywhere inside a nodule. The image database uses the relational Sybase database system. Patient images and nodule information are stored in separate tables. Software interface has been designed to allow a user to retrieve any patient study from the picture archiving and communication system into the image database.

The care and efficacy of treatment for chronic wounds is typically determined by observing and measuring the wound's response to a given treatment protocol. The traditional measures of wound morphology typically include photographs taken over time, alginates for determining wound volume, and rulers or concentric circles to estimate a wound's diameter. Although the traditional wound morphology measures are generally non-invasive, they are subjective and non-repeatable. Information on tissue response is generally limited to gross metabolic measurements acquired through standard diagnostic testing, bacteriological information from biopsied material and transcutaneous oximetry taken at the periphery of the wound. Information related to tissue response is generally acquired using invasive techniques. This paper describes a non-invasive method for assessing wound morphology and response being used to assess and study chronic wounds at the USAF Medical Center at Wright-Patterson AFB. This new technique exploits the properties of laser surface scanning and magnetic resonance spectroscopy to acquire its measurements. The method used employs a Cyberware^TM laser surface scanner to capture both range and color information from the patient's wound surface. The color and range data are then registered to 1 mm accuracy for visualization of the patient's surface. The Magnetic Resonance Spectroscopy (MRS) data are then captured for the same wound using a surface localization and spectra collection protocol. The MRS data includes phosphorous MRS as an indicator of cellular energy balance. Spatial registration is used to combine the Cyberware and MRS datasets. The resulting data are then presented as a 3D volume with additional parameters, such as surface area, volume, and perimeter, portrayed for the total wound and specific tissue types. Results to date for our approach include the development of an automatic feature extraction algorithm that recognizes and extracts a wound edge from the laser surface scanner data. Additional tissue type characteristics, such as granulation, epithelialization, etc., are also identified by the feature extraction algorithm. The overall goal of this research is to provide a non-invasive, reliable method for wound quantification. Numerous patients with chronic wounds (i.e., diabetic ulcers, radiation wounds, etc.) require methods such as this for determining efficacy of wound treatments. These treatments often range from growth factors therapy to hyperbaric treatment for wound care. The results of this research will provide a technique for measuring some of the underlying biochemical mechanisms of wound healing in humans.

This study aims to extract quantifiable indices characterizing ultrasound propagation and scattering in skeletal muscle, from data acquired using a real-time linear array scanner in a paediatric muscle clinic, in order to establish early diagnosis of Duchenne muscular dystrophy in young children, as well as to chart the progressive severity of the disease. Approximately 40 patients with gait disorders, aged between 1 and 11 years, were scanned with a real-time linear array ultrasound scanner, at 5 MHz. A control group consisted of approximately 50 boys, in the same age range, with no evidence or history of muscle disease. Results show that ultrasound quantitative methods can provide a tight clustering of normal data, and also provide a basis for charting the degree of change in diseased muscle. The most significant (quantitative) parameters derive from the frequency of the attenuation and the muscle echogenicity. The approach provides a discrimination method that is more sensitive than visual assessment of the corresponding image by even an experienced observer. There are also indications that the need for traumatic muscle biopsy may be obviated in some cases.

A computerized dedicated system for patient positioning and monitoring in radiotherapy is herein presented. The system is based on the motion analysis system ELITE, exploiting its real-time operation features. Core of the method is the analytical comparison between the 3D co-ordinates of a set of body landmarks with a corresponding reference configuration, previously acquired and memorized. The system provides in real-time the identification in the environment of the selected anatomical landmarks, highlighted by means of infrared light reflecting passive markers, their 3D co-ordinates calculation and the analytical comparison with the reference configuration. The detection of patient's breathing phases, considering the 3D co-ordinates of three suitable markers for the supine position, has allowed to synchronize the analytical comparison to patient's breathing FRC (Functional Residual Capacity) increasing accuracy and reliability of patient's position control. The real-time process is completed by displaying to the operator a synoptic graphic representation of the current marker displacements situation relative to the most recent patient's FRC. The experimental use of the developed system has brought to a significant reduction of the time required for the positioning procedure, reaching a suitable tradeoff between a quick irradiation set-up and a clinically correct patient irradiation.

Surface reconstruction and representation is an important aspect of Biomedical Image Analysis. With respect to the shape and organization of DNA within a cell nucleus, the development of improved technology for visualizing genome organization and function as it exists within a cell nucleus has led to increasing interest in understanding the genome and its expression in three dimensions. In this context, it is necessary to be able to visualize the cell nucleus within which the DNA replication occurs. Here we discuss the implementation of a 3D surface reconstruction system for shape analysis of nuclei in mammalian cells obtained through laser confocal microscopy. Finite elements based deformable models are used for the surface reconstruction. The deformable models developed are based on thin shell element formulation. An initial global shape is assigned, and deformation forces based on distance and elasticity constraints are used to fit the final shape. The deformations energy functionals are the spline deformation energy functionals. These models are applied to images of a mammalian cell nucleus to extract shape information. Based on the segmented shape, volume and surface area are computed for the nucleus. Relevance of this method in shape analysis of biomedical images is examined. The biological data for the experiments were obtained from Dr. Ronald Berezeney from the Biomedical Imaging Group.

This paper presents 3D tongue surfaces reconstructed from sixty cross-sectional slices of the tongue. Surfaces were reconstructed for sustained vocalizations of 18 American English sounds. Electropalatography (EPG) data also were collected for the sounds to compare tongue surface shape with tongue-palate contact patterns. The ultrasound data were grouped into four tongue shape categories. These classes were front raising, complete channel, back raising, two-point displacement. The first three categories contained both vowels and consonants, the last only consonants. The EPG data indicated three categories of tongue-palate contact: bilateral, cross-sectional, combination of the two. Vowels used only the first pattern, consonants used all three. The EPG data provided an observably distinction in contact pattern between consonants and vowels. The ultrasound tongue surface data did not. The conclusion was that the tongue actually has a limited repertoire of shapes, and positions them against the palate in different ways for consonants vs. vowels to create narrow channels, divert airflow and produce sound.