The output of most medical imaging systems is a display for interpretation by human observers. This paper provides a general summary of recent work on shape recognition by humans. Two broad modes of visual image processing executed by different cortical loci can be distinguished: a) a mode for motor interaction which is sensitive to quantitative variation in image parameters and b) a mode for basic-level object recognition which is based on a small set of qualitative contrasts in viewpoint invariant properties of images edges. Many medical image classifications pose inherently difficult problems for the recognition system in that they are based on quantitative and surface patch variations--rather than qualitative--variations. But when recognition can be achieved quickly and accurately it is possible that a small viewpoint invariant contrast has been discovered and is being exploited by the interpreter.

on the lung airway s1iov that it has fracta[ featuie. already notice(I by several authors : it is a complex structure vitli a high degree of recursive division (up to 23 levels). and its surface is very large (80 rn2 in average) compared to its volume. The aim of this paper is to use a fractal model to simulate the growth of a lung. In order to construct a meaningful fllO(Iel. we take into account some natural lnvs that rule the grovt1i : maximisatioii of the lung surface. minimisation of its volume. minimisation of the gas tov resistance forces due to the air circulation. border constraints. then make a COIflI)Uter simulation of the iiiorphogenesis to ol)tain a three dimensiounal model of the lung. This model shows good agreement both vith average rnorplonetric data and vith fractal measurement on real lungs. The main problem. however. remains to adjust the different parameters of the model to lit best the (lata. The advantages of such an approach is that it gives a futictionnal knowledge of the lung airway. a rather precise description of the high level branches. which are difficult to observe. and that it could he used to predict evolution under abnormal conditions (for instance border conditions). 10 / SPIE Vol. 1233 Medical Imaging IV: Image Processing (1990)

This paper considers the propagation of a pulse through a theoretical fractal network. The temporal and spectral features resulting from the decorrelation of a pulse as it transits the network including the effeci of perturbations are discussed. Application to naturally occurring networks such as the His-Purkinje system of the myocardium are considered. Both the fractal dimension (DF) and the power spectrum [P(w)are descriptive of the organizational features of the network. DF describes the geometric structure based on segment length generation and scaling factors. P(w) measures the heterogeneity of the pathways present in the network which in turn reflects the distribution of the segment length variations throughout the network. We show a direct relation between structural heterogeneity and the spectral features. Variation of perturbation size and location shows that changes in geometry occur across many size scales consistent with a fractal structure. Furthermore both location and size of the perturbation are important determinants of the extent of spectral changes produced. Simulation of the His-Purkinje conduction system demonstrates QRS complexes in good agreement with clinical observations. Additionally simulation of conduction defects in the network shows that network perturbations are reflected in the QRS and more importantly in P(w). Our results suggest that characterizing perturbations in terms of their functional wavelength (frequency) rather than their geometric dimension may be a useful method for assessing the extent of performance degradation in the system under study.

Consider the problem of localizing and identifying cell organelle in a transmission electron micrograph. Opera. . tions on regions constitute the key tasks in segmenting such images. The construction of meaningful entities from an initial fine partitioning of such an image poses problems that are generally linked to the type of objects to be identified. Restraining region manipulation algorithms to a particular class of images may simplify the process. However the loss of retargetability for the segmentation process is a serious handicap of such a solution. Segmentation must be based on a formal mechanism for reasoning about scenes (cells) their images (trans. . mission electron micrographs) objects in the scene (organelle) and their representation (image regions) if the system is to be suitable for a wide variety of domains. Mathematical morphology offers such a mechanism but the drawbacks of point set topology limit its success to low-level vision tasks. A modification based on expected morphological properties of point sets and not their fixed structure is useful for intermediate vision. This modification makes it possible to develop a theory for region manipulation. The main goal of growing and shrinking is to obtain regions with a given expected structure. The operations involve a search in the region space for candidates satisfying specified conditions and capable of producing final regions that fit the desired goal subject to given constraints. In this paper we describe

A neural network approach was applied for the differential diagnosis of interstitial lung diseases. The neural network was designed for distinguishing between 9 types of interstitial lung diseases based on 20 items of clinical and radiographic information. A database for training and testing the neural network was created with 10 hypothetical cases for each of the 9 diseases. The performance of the neural network was evaluated by ROC analysis. The optimal parameters for the current neural network were determined by selecting those yielding the highest ROC curves. In this case the neural network consisted of one hidden layer including 6 units and was trained with 200 learning iterations. When the decision performances of the neural network chest radiologists and senior radiology residents were compared the neural network indicated high performance comparable to that of chest radiologists and superior to that of senior radiology residents. Our preliminary results suggested strongly that the neural network approach had potential utility in the computer-aided differential diagnosis of interstitial lung diseases. 1_

Currently, more sophisticated image processing operations are not used on clinical workstations or CR (computed radiography) systems. One reason for this is the excessive time required to perform the image processing. One parallelism approach for improving image processing speed is to utilize a single instruction stream - multiple data stream (SIMD) computer. This paper presents the preliminary results for two classes of image processing algorithms which are useful in radiological applications: segmentation and registration.

This paper describes a 3D-image processing approach to generate a combination display of intracranial vessels and adjacent brain tissue surfaces in magnetic resonance imaging (MRI). The algorithm is based on the ray-tracing principle and may be regarded as a union of the techniques of surface integration and maximum intensity projection (MIP). Measurement methods and preprocessing steps fo acquisition of a flow-compensated vessel dataset and a T1-weighted tissue volume with isolated brain with equal partitioning are described. The method is intended as a tool for the optimization of neurosurgical planning. 1.

An image editing system has been developed that incorporates some intelligent and semi-automated techniques to improve the efficiency of the editing process. Some of the functions provided in the editor operate on the entire image sets instead of individual images. Such functions are capable of accomplishing thresholding operations or operations that remove simple objects from the data set. In order to automate the joint disarticulation process, a special joint trace function is provided. This function automatically traces the joint space in the image from given constraints. Manual editing functions are also provided to enable general editing operations. These techniques, coupled with an efficient user interface, contribute to the improvement of speed of the editing process when applied to real cases.

Three-dimensional reconstruction of coronary arteries from two views often encounters problems such as vessel overlapping and foreshortening (high inclination) in images and points mismatching between two views. These problems restrict the reconstruction accuracy and often cause erroneous results. Based on a two-view algorithm we have developed techniques to incorporate more than two views into the reconstruction. We demonstrate that inclusion of the additional information helps to uniquely match points among views and increases chances that a point can be clearly seen and accurately measured in at least two views. The reconstruction accuracy is consequently improved.

The measurement of the local curvature of arbitrary discrete 3-D medical images is complicated by the difficulties of defining a local neighborhood and then mapping the surface of the neighborhood onto the unit square as a way to unambiguously define a parameterization. Five practical methods are presented for deriving one or more measures of curvature about a point on an arbitrary discretized surface. The first 3 methods approximate the surface patch using continuous biquadratics while the next 2 methods obtain the curvature directly from the discrete data points on the surface which define the neighborhood. The 5 methods are compared in computational complexity accuracy and robustness in the presence of a noisy surface. 1.

Simulated annealing a Monte Carlolike algorithm is applied to 3D blood vessel reconstruction based on digital subtraction angiography(DSA). Since the imaging system is highly illposed because of the insufficient number of projections good reconstruction can not be achieved with only the observed data. This paper shows that even in such cases objects can be well reconstructed by simulated annealing making use of constraints based on a priori knowledge. The performance of this method for photon noise and the feasibility of 3D reconstruction are discussed through some computer simulations. 1.

This paper proposes a method to describe and identify 3-D curved objects using a complete set of 3-D range data. Curved 3-D objects in general are difficult to represent and identify because they lack distinct local properties such as edges planes cylindrical surfaces etc. which are the basic building blocks used to represent objects when using intensity or 2-D range images. In this paper we propose to use principal axes to establish a reference for describing a 3-D object. A method of obtaining an inertia matrix from the complete 3-D range data is developed. Using this method an unique set of principal axes of an object with an arbitrary 3-D position and orientation is first obtained. On the principal axis coordinate the object is described by the normalized features describing the shape of the object such as a spine compactness section size section contraction and section orientation. As experiments several 3-D objects are described using the proposed method by features two scalars and three vectors. For the purpose of identification a similarity measure between the two objects are defined based on the descriptions of objects. As the proposed method is based on the global features of objects using sufficient information from 3-D range data it provides an unique description which is suggestive of the shape of an object as well as an accurate noise-insensitive identifying method. 1.

Magnetic resonance imaging (MRI) being one of the most promising neuro-imaging modalities in medical diagnosis suffers from spatial inaccuracy when it comes to high precision processes such as neuro-stereotactic surgery. Spatial distortions of several millimeters are possible. The distortion originates from the fact that highly controllable magnetic fields are used to encode the positions of the spins. When the fields are perturbed by unknown interferences such as the para- and/or dia-magnetism of the patient or inhomogeneity in the main field the spatial encoding becomes inaccurate. However the distortions are not without helpful characteristics. For instance using an SE pulse sequence one may observe that 1) the distortion vectors are collinear 2) they are along an axis which is determined by the gradients and 3) the integral of image intensity between homologous coaxial points is unaffected by the inhomogeneity. With such an observation we propose a method that uses a pair of images from the same object that are acquired using a set of gradients having a prescribed relationship. While these two images are both distorted a set of independent first order ordinary differential equations have been derived to relate the two images. Based upon this relation we show that a distortion-compensated image can be derived. We also show some preliminary phantom study results as well as computer simulation results. The objective of this study is to arrive at spatially accurate MRI

The paper describes applications problems and approaches for image segmentation in magnetic resonance imaging. The methods which are proposed work on 3D-datasets with the goal of isolating tissue volumes. Unlike 2D-techniques which operate on multispectral image data in 3D-image segmentation only one image for one anatomical slice is available lacking essential information for tissue discrimination. Three different approaches for the task of volume segmentation are presented. The first is based on the detection of edge structures dividing the original images into anatomically relevant object regions. A 3D-region merging algorithm is applied to extract those regions which belong to the object to be segmented from the region dataset. The second method consists of a polynomial classification of imagepixels into several user-defined tissue classes. Local texture properties are used as discrimination features. The third approach region classification may be regarded as a combination between edge detection and pixel classification. On the basis of a presegmentation of the dataset into object regions a classification process tries to group the regions into different object classes making use of various region features. The latter strategy has yielded the best results and highest reliability for 3D-image segmentation. Further improvements towards minimization of user-interaction are proposed. 1.

Neural networks are well adapted to the task of grouping input patterns into subsets which share some similarity. Moreover once trained they can generalize their classification rules to classify new data sets. Sets of pixel intensities from magnetic resonance (MR) images provide a natural input to a neural network by varying imaging parameters MR images can reflect various independent physical parameters of tissues in their pixel intensities. A neural net can then be trained to classify physically similar tissue types based on sets of pixel intensities resulting from different imaging studies on the same subject. A neural network classifier for image segmentation was implemented on a Sun 4/60 and was tested on the task of classifying tissues of canine head MR images. Four images of a transaxial slice with different imaging sequences were taken as input to the network (three spin-echo images and an inversion recovery image). The training set consisted of 691 representative samples of gray matter white matter cerebrospinal fluid bone and muscle preclassified by a neuroscientist. The network was trained using a fast backpropagation algorithm to derive the decision criteria to classify any location in the image by its pixel intensities and the image was subsequently segmented by the classifier. The classifier''s performance was evaluated as a function of network size number of network layers and length of training. A single layer neural network performed quite well at

Multi-spectral image data fusion techniques for tissue classification of magnetic resonance (MR) images are presented. Using MR it is possible to obtain imagesof proton density the spin-lattice relaxation time constant ( T1) and the spin-spin relaxation time constant (T2) of the same anatomical section of the human body. In this paper we adopt a sensor fusion approach to tissue classification and segmentation in which each of the three images are treated as the output of different sensors. Regions of the images are modeled as noncausal Gaussian Markov random fields (GMRFs) and the underlying tissue label image is also assumed to follow a Gibbs distribution. Two different multi-spectral tissue labeling algorithms maximum a posteriori (MAP) estimation and the Dempster-Shafer evidential reasoning technique are presented. In the Bayesian MAP approach we use an independent opinion pool for data fusion and a deterministic relaxation to obtain the MAP solution. In practice the Bayesian approach may be too restrictive and a likelihood represented by a point probability value is usually an overstatement of what is actually known. In the Dempster-Shafer approach we adopt Dempster''s rule of combination for data fusion using belief intervals and ignorance to represent our confidence in a particular labeling and we present a new deterministic relaxation scheme that updates the belief intervals. Results obtained from real MR images are presented. 1.

To provide a means of rendering complex computer architectures languages and input/output modalities transparent to experienced and inexperienced users research is being conducted to develop a voice driven/voice response computer graphics imaging system. The system will be used for reconstructing and displaying computed tomography and magnetic resonance imaging scan data. In conjunction with this study an artificial intelligence (Al) control strategy was developed to interface the voice components and support software to the computer graphics functions implemented on the Sun Microsystems 4/280 color graphics workstation. Based on generated text and converted renditions of verbal utterances by the user the Al control strategy determines the user''s intent and develops and validates a plan. The program type and parameters within the plan are used as input to the graphics system for reconstructing and displaying medical image data corresponding to that perceived intent. If the plan is not valid the control strategy queries the user for additional information. The control strategy operates in a conversation mode and vocally provides system status reports. A detailed examination of the various AT techniques is presented with major emphasis being placed on their specific roles within the total control strategy structure. 1.

Tracking of magnetic resonance (MR) tags in myocardial tissue promises to be an effective tool in the assessment of myocardial motion. The amount of data acquired is very large and the measurements are numerous and must be precise requiring automated tracking methods. We describe a hierarchy of image processing steps that estimate both the endocardial and epicardial boundaries of the left ventricle and also estimate the spines of radial tags that emanate outward from the left ventricular cavity. The first stage determines the position of the myocardial boundaries for each of 128 rays emanating from the origin. To counter the deleterious effects of noise and the presence of the tags when determining the boundary positions we use nonlinear filtering concepts from mathematical morphology together with a prion knowledge related to boundary smoothness to improve the estimates. The second stage estimates the tag spines by matching a template in a direction orthogonal to the expected tag direction. We show results on tagged images and discuss further research directions. 1.

At present mammography is the most effective method for the early detection of breast cancer1 . Detection and classification of masses in mammograms are among the most important and difficult tasks performed by radiologists. Various studies have indicated that regular mammographic screening can reduce the mortality from breast cancer in women2. Thus mammography may become one of the largest volume x-ray procedures routinely interpreted by radiologists. The miss rate for the radiographic detection of malignant masses ranges from 12 to 30 percent. In addition although general rules exist for the visual differentiation of benign and malignant masses error does occur in the classification of masses with the current methods of radiologic characterization. Thus it is apparent that the efficiency and effectiveness of screening procedures could be increased by use of a computer system that successfully aids the radiologist in detecting and characterizing mammographic masses. We are developing computerized schemes for the automated detection and classification of masses in digital mammograms. The detection scheme utilizes the architectural symmeiry of the left and right breasts and digital bilateralsubtraction techniques in order to increase the conspicuity of the mammographic mass prior to the application of featureextraction techniques. The classification scheme involves the extraction of border information from the mammographic mass in order to quantify the degree of spiculation which is related to the likelihood of malignancy. METHODS Clinical screen/film mammograms were used in the

An algorithm for the autoitatic detection of clusters of calcifications in digital marrtinograrrts has been developed using image analysis techniques. The results for 50 complete clinical mammograms show that the algorithm achieves 100 true positive cluster detection rate with a false detection rate of 5 clus ters. 1.

The contribution of the vertebral cortex to the strength of the vertebrae is unknown. Several researchers have attempted to estimate this contribution by mechanically testing excised vertebral specimens. We have addressed the problem by creating finite element models (FEM) of the lumbar spine from contiguous scan CT studies. Seventeen women with no evidence of spinal fracture were scanned on a commercial CT scanner along with a calibration phantom to allow accurate determination of spinal bone mineral density. Using interactive software techniques two vertebral FEM are created from each scan set one with the cortex intact and the second with the cortical shell removed. Each model was analyzed under compression using finite element software and the results were compared for the intact and no cortex cases. Removal of the cortical shell was found to significantly decrease vertebral strength an average of 15. 6 3. 0 (mean standard error of the mean) in our study (p 0. 001). The decrease was variable between patients and did not appear related to trabecular mineral density. In osteoporotic patients the cortical contribution to vertebral strength may be much greater than that reported for the normal women due to the accelerated loss of trabecular bone. 1.

This paper consists of two parts. The first part considers the limitations imposed by statistical properties of ultrasound images. Through this analysis the minimum detectable tumor size from an ultrasound Bscan using the current state of the art is determined the second part describes an improvement to a successful tissue-characterization algorithm that adds several image processing steps to compute the tissuecharacterization features. The inclusion of such steps will enable the tissue-characterization algorithm to take advantage of visual cues similar to those that a clinician would use to differentiate various organs and segments of the image. This in turn expands the applicability of the present tissue-characterization algorithm from multivariate to multiorgan and multidisease cases.

Gestational age estimation from the measurements of the fetal head from the ultrasound images can be of great importance in management decisions during pregnancy. Biparietal diameter (BPD), occipital frontal diameter (OFD), and the head circumference (HC) are the most common parameters used for estimating the gestational age. Since the fetal head has an elliptical shape in the two dimensional image plane, extracting the ellipse and measuring the major axis (OFC), minor axis (BPD), and circumference (HC) are the task of image processing which will be discussed in this paper.

Based on the assertion that entropy and leakage are related this paper describes the software implementation of a texture analysis technique which is based on entropy for analyzing macromolecular leakage from microvessels. Images of vessel leakage were compared to a pre-leakage image by computing their percent changes in entropy to give a relative measure of leakage. Entropy calculations were tested on different region sizes of the images to determine the regional sources as well as topographical spread of the leakage. Since entropy can be based on the statistics of both gray level components and frequency components the FWT (Fast-Walsh Transform) FF1'' (Fast-Fourier Transform) DCT (Discrete-Cosine Transform) and histogram routines were implemented in C to investigate the effects of transform type on the entropy measure. The percent changes in entropy from the frequency analyses were found to be more significant ''than changes in entropy from the histogram approach. Moreover the FWT was found to be comparable to the FFT and DCT with regard to the entropy measure and was thus chosen as the better transformation because it decreased computation time and memory requirements. This software package successfully produced a texture analysis technique based on entropy. However the exact quantitative relationship between vessel leakage and entropy measures has not fully been established. . 1.

Accurate measurement of blood flow rates can provide a useful means for evaluation of functional significance of stenoses coronary flow reserve and changes in flow rates resulting from interventional procedures. However for pulsatile flow conditions blood flow rates determined by analysis of conventional time-density curves are unreliable. Therefore we are developing a new method based on an analysis of the distribution of contrast material along the length of the vessel i. e. the " distance-density" curve in digital subtraction angiographic (DSA) images. The distance that the contrast material travels during the time between two image acquisitions is determined by comparison of the two distance-density curves. The flow rate between the image acquisitions is calculated by multiplying this distance by the frame rate and the vessel crosssectional area which is estimated from the vessel size. Thus for high frame-rate acquisitions " instantaneous" blood floW rates can be determined. From angiograms obtained at 15 frames/see the instantaneous flow rates measured with our technique agreed with those measured with an electromagnetic method to within an average of 2. 3 cc/sec for pulsatilc llovv conditions with peak flow rates of up to 30 cc/see and average flow rates agreed to within an average of 11. 1.

X-ray angiography images of the legs are enhanced using linear and morphological image processing tech- I1iques. Salient features of the leg images include high contrast bones relatively smooth soft tissue and arteries. Filter parameters are optimized to enhance the arteries while deemphasizing the bones and other structured noise. Morphological processing has several advantages compared to linear processing including improved retention of both arterial dimensions and x-ray density. 1

In this paper we describe work-in-progress regarding a fully automatic reporting system for coronary artery stenotic lesions. In a first step all blood vessel segments are assigned their anatomical label according to a coronary anatomy model. Segment labeling is done using a constraint satisfaction technique because most anatomical coronary artery knowledge can be formulated as umary constraints only depending on local segment attributes and binary relational constraints such as thicker than left ofand above. In a second step we perform an automatic quantification of all artery trajectories. Therefore we calculate a stenosis severity score for each segment which is not only based on local properties like per cent diameter or per cent area stenosis but also takes into account the anatomical significance of the vessel. For example a stenotic lesion proximal on the Left Anterior Descending (LAD) branch is much more significant than one on its distal side branches. Results are presented on clinical coronary angiograms. 1.

A combined signal processing and knowledge-based approach was developed to graphically reconstruct the three-dimensional vascular structures from dual biplane angiograms. It was shown that the correct correspondence of segments between two orthogonal views of a simple tree model can be determined by use of a knowledge base containing a directed-graph representation of the tree structures a procedure of recognizing crossover points a requirement of consistency in connectivity and a constraint of coordinate invariance along the axis of view rotation. However the current system failed to handle errors in the segmentation caused by either the complexity of the vascular structures or the presence of noise in the images. The result suggested that the key to resolving ambiguities in the two-view reconstruction problem be the provision for extensive interactions among segmentation correspondence and knowledge base. An systematic analysis and a preliminary design of such a knowledge-based system were presented. 1.

Cardiac angiograms are preferably imaged jn orthogonal so-called biplane views. In such image pairs the vessels do not resemble each other to such a degree that correspondences between them can be derived directly. Direct computer matches on the basis of similarity have been used for closer angles like 15 or 20 degrees [9 In this recent study model knowledge for each of the standardized angio-projections is used to recognize the vessels in each view. After these two matches for identification the third dimension can be calculated using the fixed and known relation of the orthogonal vascular models. 1.

Contrast Limited Adaptive Histogram Equalization (CLAHE) which was developed to enhance regional contrast of an image was improved by using fuzzy sets. In the proposed method gray level transformation of each pixel is performed using each cumulative distribution function calculated from a histogram of pixel values in a contextual region of the target pixel. We optimized two parameters for making the cumulative distribution function that is size of contextual regions and a clipping level of the histogram for each pixel by using fuzzy logic. The results of simulation showed that the method not only yielded optimum contrast at any regions of an image but also reduced boundary artifacts appeared on images processed by a conventional CLAHE algorithm. 1.

In this paper the autonomous behaviour of neural networks detecting straight lines is investigated. At this purpose the role of the synaptic junctions in the recognition process of elementary visual stimuli is emphasized taking into account an isotropic architecture of neural network. Finally it is shown that so organized models of linear neurons have a " natural" capability to exhibit coordinate oscillations with constant norm in the state space. 1.

If photons are of constant energy and if they are emitted with constant intensity in the body say I then the intensity I of the photons reaching a film is established as I 10 e IS I''(XrY) dx. dy (1) where IA(x is the Xray attenuation coefficient of the medium (tissue or organ) at point (x from the source. An X-''ray transform is defined for a function f(x as i: f(( 1 J'' f( e . f(: ds . dt (2) It is shown that after sampling the discrete form of this transform is N N . ) ''i. nk i i 1 e '' (3) In this paper the computational aspects of this transform are studied and different algorithmic procedures are outlined. Furthermore its application in convolution and filtering is evaluated in medical image processing. Emphasis is placed on the use of residue arithmetic where this transform is used as numbertheoretic. It is compared with other potentially effective transforms as specially applied to medical image processing. 392 / SPIE Vol. 1233 Medical Imaging/V Image Processing (1990)

Although chest radiography is the most commonly performed radiographic examination and one of the most valuable and cost-effective studies in medicine it suffers from relatively high error rates in both missing pathology and false positive interpretations. Detectability of lung nodules and other structures in underpenetrated regions of the chest film can be improved by both exposure and optical compensation but current compensation systems require major capital cost or a significant change in normal clinical practice. A new optical compensation system called the " Intelligent X-Ray Illuminator" (IXI) automatically and virtually instantaneously generates a patient-specific optical unsharp mask that is projected directly on a radiograph. When a radiograph is placed on the IXI which looks much like a conventional viewbox it acquires a low-resolution electronic image of this film from which the film transmission is derived. The transmission information is inverted and blurred in an image processor to form an unsharp mask which is fed into a spatial light modulator (SLM) placed between a light source and the radiograph. The SLM tailors the viewbox luminance by decreasing illumination to underexposed (i. e. transmissive) areas of the radiograph presenting the observer with an optically unsharp-masked image. The IXI uses the original radiograph and will allow it to be viewed on demand with conventional (uniform illumination. Potentially the IXI could introduce the known beneficial aspects of optical unsharp masking into radiology at low capital

ICOS is a modular software system for the 3-dimensional reconstruction of icosohedral particles from 2-dimensional projections. Noise in the resulting reconstructions is due to many sources including ringing artifact due to finite sampling of Fourier transforms and is comparable with many of the desired components representing particulate structure. This paper describes a method for reducing ringing artifact without significant loss of resolution. Application of a Blackman window to the original Fourier transforms significantly improves the signal to noise ratio by suppressing ringing artifact but reduces resolution in the fmal reconstruction. Taking a voxel by voxel minimum of the filtered and unfiltered 3-dimensional reconstruction restores the original resolution while retaining the enhanced SNR. 2.

A method which improves the signal-to-noise ratio in derived images is discussed. It is shown that if the derived images are filtered appropriately it may be possible to improve the signal-to-noise ratio considerably. The advantage of the current method over low-pass filtering is that one may be able to retain some of the high spatial frequency detail in the processed images. The method is applied to the mapping of molecular diffusion constant by NMIR imaging. 1.

The Integrated Radiological Information System (IRIS) supports the capture and distribution of digitized x-ray images and voice reports in the form of " electronic" patient folders which can be accessed at physician workstations throughout the hospital. Each workstation has an image screen to display documents and x-ray images a control screen to access patient folders and a hands-free telephone to dictate and play back reports and enable consultation between radiologist and clinician workstations. A seven week clinical trial of IRIS was conducted at the Ottawa Civic Hospital during April and May 1989. The system operated to process cases from the Department of Emergency Medicine weekday afternoons. Observers recorded for each case how radiologists used the system. After the trial radiologists participated in an extensive debriefing interview during which they were asked to complete a number of rating scales addressing the following issues: 1) willingness to diagnose by tissue type and by type of pathology 2) seriousness of problems due to system limitations 3) the perceived usefulness of enhancement capabilities and measurement tools. Overall the system was found to be acceptable by the radiologists. There was some concern about diagnosis in soft tissue regions. Most of the system features were regarded as acceptable but there were areas which needed improvement. The suggested improvements are described where applicable. The enhancement facilities and the means of using the facilities were acceptable overall. 426 /

The accurate quantitative assessment of variations in coronary arterial dimensions and explicit visualization of the flowing blood in vivo is fundamental to early diagnosis of occlusive vascular diseases. In Magnetic Resonance Velocity Imaging the velocity of flow is usually represented by intensity maps of the three spatial components of velocity which are difficult to use for diagnostic purposes. A technique for analysing the in vivo flow in arteries from MR imaging data is presented. This method uses a specific flow related enhancement process to extract and localize the flow field within the image which is followed by global flow analysis that reveals the underlying oriented structures of the flow field. By animation our method can provide various ways of visualizing the in vivo flow and its variation towards different vessel structures and lesions. Both lateral and cross-sectional views are animated and video tapes of the operation of the on-line animation package demonstrate the effectiveness of this diagnostic tool. 1.

Bayesian methods that utilize Gibbs priors to incorporate a priori information in the statistical models used for deriving algorithms for image processing and image reconstruction have been developed. The Gibbs prior describes the local continuity of neighboring pixels and takes into account the effect of limited spatial resolution. These new approaches are capable of providing improved image quality. 1.

The three-dimensional (3-D) object data obtained from a scanner usually have unequal sampling frequencies in the x y and z directions. Generally the 3-D data is first interpolated between slices to obtain isotropic resolution reconstructed then operated on by object extraction and display algorithms. Several interpolation methods are available. Here three interpolation algorithms: hnear polynomial and shape methods are used for comparison. The performances of these three algorithms are compared based on the image quality and accuracy in volume measurement of the 3-D object. 1.

The purpose of this paper is to present a new cone beam short scan reconstruction technique for circular and non-circular detector orbits. The reconstruction is performed by a filtered backprojection method. The short scan reconstruction technique is first investigated for fan beam circular orbits and is then extended to fan beam noncircular orbits. Finally a straightforward approach is applied to cone beam geometries and a modified Feldkamp algorithm is used. Since the projection data are incomplete cone beam reconstructions are only approximations. I.

Several styles of stents are being tested in human vessels. The smallest stents are made of fine stainless steel wire which is difficult to see under fluoroscopy. Visibility can be improved by either changing designs or using image processing. Some investigators use realtime processors to enhance edges and so increase stent visibility. They could obtain better results with a two-dimensional video filter which selectively enhances stents. Such a filter should not require adjustment for orientation or expansion and must operate on video with no delay. Computer studies indicate a practical algorithm to improve stent visibility is possible. While simulations indicate that with this filter normal fluoroscopy and normal subjects stainless steel coronary stents will be visible clinical studies with a working system are necessary to evaluate its usefulness. 1.

Hand radiographs provide a valuable index of disease in arthritis and other generalized diseases such as secondary hyperparathyroidism and osteoporosis. Measures such as cortical volume intercortical width average and periarticular demineralization provide diagnostic indicators for these diseases. However visual analysis of hand radiographs is not quantitative and is compromised by both interobserver and intraobserver variation. Automation of these measures would provide repeatable comparable quantities to assist in diagnosis and disease and therapy monitoring. The computer calculations to perform these measures are straightforward. The key problem is automatic segmentation of the hand anatomy that is recognizing the pixels that correspond to specific imaged bones and joints. Our approach incorporates computer-represented hand models in addition to more traditional image processing algorithms. We describe our techniques for using a combination of predictive models and image processing evidence to automatically fmd bone and tissue boundaries and identify specific bone and joints. 2. COMPUTING ARTHRITIS MEASURES Digital scanners and radiograph digitizers make the radiograph available as a data source for computer algorithms that analyze medical imagery. This is significant because radiographs comprise more than 80 of all medical imagery at this time and they are considerably quicker and less costly than other digital modalities such as CT and Mill. Quantitative measures from digital radiographs can aid physicians in diagnosis tracking disease progress and in therapy planning and evaluation. We have begun studying diagnostic measures in arthritis

Two types of workstation are being developed at UCLA for neuroradiology a display workstation and a therapeutic workstation. The display station is used for review of images derived from neuroradiology examinations whereas the therapeutic station is for image analysis. The therapeutic workstation will be used to compare angiograms with CT and MR images and to calculate quantitative information from image sets. Parameters obtained by image analysis will be useful for planning and evaluation of interventional procedures. Current emphasis is on development of analysis tools for digital subtraction angiography. Digital densitometry and parametric imaging routines are being developed for analysis of DSA images of blood flow (with contrast injection) taken with a GE Digital Flouricon 5000 system. These routines include determination of vessel geometry regional blood volume flow rate and velocity. The starting point for software development is the CALIPSO (CALifornia Image Processing SOftware) package developed at UCLA for the Macintosh II. A Stellar GS2000 graphics mini-supercomputer will ultimately be used to allow rapid manipulation of images. The workstation will be connected to various imaging modalities through an Ethernet network. 1.

A detection-estintation method is introduced for use in digital subtraction angiography (DSA) . The method allows three dimensional (3-D) reconstruction of the vasculature by an image intensifier-based volume irnager using few subtracted projections. In the first step several 2-D projections are processed by vascular segmentation algorithms to define binary vasculature envelopes. In the second step a " logical" backprojection algorithm defines the 3-D binary envelope of the vasculature. The gray levels of the voxels within the 3-D envelope are then estimated by a constrained algebraic reconstruction technique (ART) to refine the vessel boundaries. Results of the reconstruction of an aluminum structure specially designed to test vascular reconstruction algorithms are presented. 1.

A prototype voluiie CT irnager has been tested using a vascular phantom. This systen consists of a fixed x-ray tube a conventional image intensifier coupled to a chargecoupled device (CCD) camera a computer-controlled turntable on which phantons are placed and a digital computer with a graphics station. To explore the imaging performance of the system for reconstructing a three-dimensional (3D) vascular structure a set of subtraction projections of a vascular phantom acquired over 25 projection angles with two different sizes of image intensifiers were digitized. Then these data were reconstructed using two iterative algorithms specially designed for 3D vascular structures. The quality of the reconstructed vascular images indicates that the system can offer adequate signal-to-noise ratio (SNR) for direct 3D vascular reconstruction when only a few subtraction projections are used assuming intraarterial injection of contrast. The results also suggest that the iterative vascular reconstruction algorithms work well for direct 3D vascular reconstruction. 1.

We have developed a microcomputer based system with an application specific software package which permits the direct digitization and analysis of transrectal ultrasound (TRUS) images. The system is highly flexible and enables access to a wide range of image analysis tools through relatively simple software modifications which cannot. be implemented using a standard ultrasound instrument. We have demonstrated the capability of the system by an analysis of a number of morphometric parameters and by a correlation of these measurements with the presence of prostatic cancer. We found that the measurement of the ratio of the anterior-posterior axis to transverse axis (ATR) and the presumed circular area ratio (PCAR) were significant predictors of prostatic cancer. The sensitivity of the PCAR measurement was 93 the specificity was 50 the positive predictive value was 57 and the negative predictive value was 91. This system provides researchers with an efficient economical and flexible method to aid in the analysis of ThUS images in a quantitative manner. 1.

The increased application of digital imaging techniques to diagnostic cardiology requires the resolution of several remaining problems involving the transmis sion display and storage of clinical image data. A Digital Imaging Laboratory is being assembled for the purpose of addressing these problems on a routine clinical basis in a high volume environment. The laboratory consists of independent workstations connected over a local area network which can be accessed by different users for display analysis and storage of diagnostic image data. In addition the laboratory is connected to a hospital-wide network providing communication with other clinical laboratories and other digital imaging modalities.

In the visual world moving edges in the periphery represent vital pieces of information that directs the human foveation mechanism to selectively gather information around these specific locations. This computationally efficient approach of allocating resources at key locations has inspired computer visionists to develop new target detection and hacking algorithms based on motion detection in image sequences. In this study we implemented a recursive algorithm for estimating motion vector fields for each pixel in a sequence of Digital Subtraction Angiography (DSA) images. Velocity information is used to segment the image and perform linear quadratic and acceleration-based frame interpolation to produce an apparent frame rate increase. Our results demonstrate the feasibility of low-rate digital fluoroscopy hence less exposure risks while preserving image quality. Furthermore the technique can be useful in the medical Picture Archival and Communication Systems (PACS) where image data can be compressed by storing and transmiting only the motion fields associated with the moving pixels. 1.

Abstract not available.

Some images present a quite strong brightness gradient which may hide significant details of the represented scene. A statistical method for enhancement by elimination of this gradient is proposed. In the first section we present the statistical background which is used. The second section is devoted to the description of our model. The third section consists in the description of the steps involved in the process of enhancement. In the fourth section we present results obtained on a test image. The last section gives a brief discussion on advantages and perspectives of this approach. 1.

The purpose of this project is to develop a low cost workstation for quantitative analysis of multimodality images using a Macintosh II personal computer. In the current configuration the Macintosh operates as a stand alone workstation where images are imported either from a central PACS server through a standard Ethernet network or recorded through video digitizer board. The CALIPSO software developed contains a large variety ofbasic image display and manipulation tools. We focused our effort however on the design and implementation ofquantitative analysis methods that can be applied to images from different imaging modalities. Analysis modules currently implemented include geometric and densitometric volumes and ejection fraction calculation from radionuclide and cine-angiograms Fourier analysis ofcardiac wall motion vascular stenosis measurement color coded parametric display of regional flow distribution from dynamic coronary angiograms automatic analysis ofmyocardial distribution ofradiolabelled tracers from tomoscintigraphic images. Several of these analysis tools were selected because they use similar color coded andparametric display methods to communicate quantitative data extracted from the images. 1. Rationale and objectives of the project Developments of Picture Archiving and Communication Systems (PACS) in clinical environment allow physicians and radiologists to assess radiographic images directly through imaging workstations (''). This convenient access to the images is often limited by the number of workstations available due in part to their high cost. There is also an increasing need for quantitative analysis ofthe images. During thepast decade

This paper presents some preliminary results on a class of image compression techniques. The techniques presented here are reversible, i.e. the compression-decompression sequence yields the original image. They are also very easy to implement.

Widely differing methods for data compression described in the ACR-NEMA draft are used in medical imaging. In our contribution we will review various methods briefly and discuss the relevant advantages and disadvantages. In detail we evaluate 1st order DPCM pyramid transformation and S transformation. We compare as coding algorithms both fixed and adaptive Huffman coding and Lempel-Ziv coding. Our comparison is performed on typical medical images from CT MR DSA and DLR (Digital Luminescence Radiography). Apart from the achieved compression factors we take into account CPU time required and main memory requirement both for compression and for decompression. For a realistic comparison we have implemented the mentioned algorithms in the C program language on a MicroVAX II and a SPARC station 1. 2.

Magnetic resonance imaging (MRI) provides excellent soft tissue contrast enabling the non-invasive visualization of soft lissue diseases. The quantification of tissues distinguishable in MR images significantly increases the diagnostic information available to physicians. New 3-D display workstations are available that can also make use of the tissue characteristics to generate clinically useful views of a patient. While simple tissue selection methods work with computed tomography (CT) images these same methods usually do not work with MR images. Several feasibility studies of tissue classification methods have been performed on MR images but few comparative studies of these methods have been published and little work is available on the best statistical model of tissues in MIRI. We have developed a novel method for the identification and quantification of soft tissues from MRI atherosclerosis in particular. This project is part of our work on the development of tissue characterization and identification tools to facilitate soft tissue disease diagnosis and evaluation utilizing MR imagery. Several supervised pattern recognition methods were investigated for tissue identification in MR images such as a Fisher linear discriminant and a minimum distance to the means classifier. For tissue in vivo adequate histology can be difficult to collect. We used cluster analysis methods to generate the necessary training information. ISODATA was modified to use hierarchical stopping rules to determine the true number of tissues in the images. This new method was