The effects of thermal radiation and quantum detector noise on biomagnetic imaging are discussed. The thermal noise due to the reception of blackbody radiation by the magnetometer sensing coil is described. Quantum limits on detector performance are calculated by first deriving the noise in a SQUID magnetometer based on a small signal circuit model. The quantum limits of SQUID noise are then applied to the noise model. The results of the calculations show that thermal radiation noise is negligible compared to quantum noise which, in turn, is about a factor of 50 less than the minimum biomagnetic fields now measurable. The models also show that the noise depends on the effective volume of the pickup coil and the relationship of imaging resolution to pickup coil geometry is described.

The paramagnetic chelate complex, gadolinium-diethylene-triamine-pentaacetic acid, Gd-DTPA, and superparamagnetic particles, such as those composed of dextran coated magnetite, function as magnetic resonance contrast agents by changing the relaxation rates, 1/T1 and 1/T2. The effects that these agents have upon MR signal intensity are determined by: the inherent biophysical properties of the tissue being imaged, the concentration of the contrast agent and the data acquisition scheme (pulse sequence parameters) employed. Following the time course of MR signal change in the first minutes after the injection of contrast agent(s) allows a dynamic assessment of organ functions in a manner analogous to certain nuclear medicine studies. In order to study renal function, sequential MR fast scan images, gradient echo (TR=35/TE=7 msec, flip angle=25 degrees), were acquired, one every 12 seconds, after intravenous injection of Gd-DTPA and/or dextran-magnetite. Gd-DTPA, which is freely filtered at the glomerulus and is neither secreted nor reabsorbed, provides information concerning renal perfusion, glomerular filtration and tubular concentrating ability. Dextran-magnetite (200 A diameter), which is primarily contained within the intravascular space shortly after injection, provides information on blood flow to and distribution within the kidney. The MR signal change observed after administration of contrast agents varied dramatically depending upon the agents injected and the imaging parameters used. Hence a broad range of physiolgic processes may be described using these techniques, i.e. contrast agent enhanced functional MR examinations.

In the field of echography, there is a large interest for the diagnostic potential of textures or speckle patterns encountered in echographic B-scan images. In this work, we present a new approach to this problem. We study the capability and the diagnostic value of a method to extract parameters describing tissue dynamics by tracking myocardial speckle pattern motion during the cardiac cycle. Such speckle motion is shown to be closely related to the myocardial tissue dynamics and therefore should be of diagnostic significance. A model was previously developed to generate typical B-scan images of the myocardium during the cardiac cycle. We appropriately modified the tissue parameters in our model to simulate the deformations happening during cardiac contraction (translation, wall thickening, fiber contraction, rotation etc.). The resulting simulated speckle pattern motion was then studied. An optical flow algorithm developed in our laboratory was successfully used to quantify this motion as a velocity field described by a set of linear equations. The correlation between this velocity field and the myocardial deformation was clearly established from the model for small deformation rates. In practice, the time interval between echographic frames (1/30 second) was expected to be adequate for only small tissue deformations to occur between frames; we could thus expect our method to be successful with real echocardiographic data. Clinical data studied with this procedure have indeed confirmed that velocity field could be obtained from myocardial speckle tracking; this allowed to characterize quantitatively the myocardial dynamics, in particular with respect to local wall thickening and myocardial contraction during the cardiac cycle. Those results indicate a potential for speckle tracking as a diagnostic tool to study myocardial deformation and contractility.

Currently ultrasonic B-scan images combine many acoustic properties and imaging system parameters to form the image. It may, however, be advantageous to image acoustic properties individually or in select groups to obtain a more direct interpretation of the results. This report describes two methods of quantitative ultrasonic imaging which we are pursuing. The two methods are: single-feature parametric imaging and multi-feature tissue-type imaging. Trade-offs among contrast, noise, and spatial resolution using computer simulations and phantom measurements are discussed.

Equipment has been set up for the assessment of observer performance in visual perception studies for testing the efficacy of ultrasound image enhancement techniques developed at the Royal Marsden Hospital. A variant of the Blackwell spatial forced choice experiment was conducted which produced contrast-detail curves of the detectability of uniform discs in a uniform background using a computer controlled experiment and a 512x512 by 8 bit digital video display. A method is described for synthesizing texture with the same statistical properties as B-scan speckle and for incorporating low contrast targets into such images. This model was used in a pilot study to determine approximate thresholds for the detection of speckle modulated discs in a speckle background. The results serve to define the basic limitations of the image display system being used, which is similar to those found in many commercial ultrasound scanners.

An imaging method using the Doppler effect is proposed in this paper. An object is illuminated by a sinusoidal wave from a linearly moving transducer. The frequency components of the reflected wave could be considered to be projections along lines at the corresponding angles. One can reconstruct images by applying the conventional CT (computerized tomography) algorithm to the projections thus obtained. The theory is derived, and it is confirmed by numerical simulation and experiment.

Contrast-Detail (C/D) analysis allows the quantitative determination of an imaging system's ability to display a range of varying-size targets as a function of contrast. Using this technique, a contrast-detail plot is obtained which can, in theory, be used to compare image quality from one imaging system to another. The C/D plot, however, is usually obtained by using data from human observer readings. We have shown earlier(7) that the performance of human observers in the task of threshold detection of simulated lesions embedded in random ultrasound noise is highly inaccurate and non-reproducible for untrained observers. We present an objective, computational method for the determination of the C/D curve for ultrasound images. This method utilizes digital images of the C/D phantom developed at CDRH, and lesion-detection algorithms that simulate the Bayesian approach using the likelihood function for an ideal observer. We present the results of this method, and discuss the relationship to the human observer and to the comparability of image quality between systems.

Ultrasonic diagnosis depends on the information obtained as a result of ultrasonic sonation (irradiation) of the patient. If the exposure is too low, useful information is not obtained; if it were to be too high, harm to the patient could result. Most diagnostic Ultrasound imaging systems use short microsecond pulses as a sonation source and construct the image from the back-scattered signal. There is now growing theoretical and experimental evidence to suggest that these microsecond pulses have temporal peak intensities high enough to produce free radicals through the mechanism of transient cavitation. Within the framework of short pulses, it is not possible to decrease the intensity without sacrificing signal to noise ratio. In this paper we propose the use of a new frequency modulated pulse to resolve the above dilemma. A bandwidth similar to that of a short pulse is maintained, but the total energy of the pulse is now spread over a longer duration with the result that the temporal peak intensity can be reduced significantly. For the purpose of theoretical simulation, we have modeled the back-scattered signal as a convolution of the tissue impulse response with the incident pulse. A necessary post-processing step of pulse compression required to regain the resolution is also outlined. While maintaining the resolution, increase in sensitivity (signal to noise) is demonstrated by a comparison with the response from a conventional short pulse. This increase in sensitivity can be used to reduce the temporal peak intensity of the pulse, thereby reducing the possible bio-effects. Experimental setup to acquire and process the data for imaging is also described.

Decision functions for many common detection/discrimination tasks can be cast in the form of functions which are linear in the image data. This is the case, for example, for the simple matched filter result in the case of detection of a known signal in known location against an additive Gaussian noise background. For other tasks the ideal (maximum likelihood) observer decision rule is higher order in the data. One example of this is detection of a known signal as above but with position unknown. Wagner and Barrett have examined this subject in some detail and have suggested that the quadratic term in such higher order decision tasks may be of primary importance.' We have simulated a variety of position unknown detection tasks in order to calculate the efficiency (vis-a-vis the exact maximum likelihood decision rule) of using the first, second, or higher order terms. We find that the second order term may dominate in some cases, but that for others it is inferior by a factor of 2 or more (especially for low contrast, non-zero mean objects) in comparison with the ideal observer decision function.

Bio-magnetic imaging systems can be used to locate simple sources of magnetic fields in the human body. In this paper we consider characterization of bio-magnetic imaging systems. We formulate a generalization of the point spread function (PSF) and optical transfer function (OTF) for use in this characterization. The generalized PSF and OTF are vector quantities derived using two components factors: the Fourier transformed Maxwell equations the estimator used in converting magnetic fields in electric-current densities An example is chosen from co-planar imaging systems, i.e., systems where the magnetic field measurements and electric currents are all located in parallel planes. Imaging ambiguities of these systems are examined in terms of imaging artifacts.

A new method of fast NMR imaging is proposed which uses oscillating gradients and also collects data on Cartesian coordinates. The method is a heuristic attempt at finding an optimal data collection strategy for NMR imaging. It is described in detail and its advantages and disadvantages compared to other methods are analyzed.

A technique for the measurement of the mean and variance of velocity field in each finite voxel by nuclear magnetic resonance(NMR) is introduced. Theoretical formalism and experimental results are presented. The experimental findings are in good agreement with the theoretical predictions.

A new technique for generating spatially resolved pressure-gradient images of flow fields is introduced.

A significant source of noise in screen-film radiography results from the variation in light output of the screen for absorption of x-rays of equal energy. Two methods are described in the literature for measuring the statistics of the number of light quanta emitted for each absorbed x-ray. The coincidence method registers x-ray events by detecting temporally correlated light quanta. It is limited in its ability to see events having a small number of quanta and can be biased by non-Poisson photomultiplier dark events. The pulse height method uses pulse shaping to produce pulses whose height is proportional to the number of quanta. In this case small pulses may be lost in order to discriminate against photomultiplier dark counts. We present a new method, based on the synchronous detection of a chopped x-ray source, which has the potential to avoid these shortcomings. Analytical methods necessary to remove the effects of unwanted x-ray energy components (such as backscatter radiation from our x-ray fluorescent targets) are discussed which provide stable estimates of the first and second moments of the light emission statistics. The method is then used to obtain a new set of measurements of the light emission statistics (including the mean light output and Swank I factor) for samples of Kodak Lanex x-ray intensifying screens for mean incident x-ray energies of 14.4, 17.8, 27.9, 35.2, 42.1, 49.8, and 59.6 keV.

In part I of this work we described an exact analytical solution for the zero-frequency Wiener spectrum of radiographic screen-film noise, modeling the nonlinear grain response as a 2-quantum threshold. We now extend our results to a full Wiener spectrum description of the image noise. Furthermore, to simulate a wide range of grain quantum sensitivities, we modify our film model to allow for each absorbed light quanta to randomly contribute to the final development of the grain. We find an analytical expression for the circularly symmetric covariance function associated with the random variable denoting the transmittance of each grain. The Wiener spectrum is obtained as the Hankel transform of this covariance function. In this way we calculate the Wiener spectrum for a full range of imaging variables (screen amplification, point spread function, exposure, etc.).

We are developing a model to characterize the spatial-frequency dependence of detective quantum efficiency, DQE(f). The transfer of energy through a fluorescent screen is modelled as a cascade of stochastic processes. We have improved on existing models in three ways: by including the effect of finite phosphor thickness, by including the possibility of secondary quantum sinks especially at high spatial frequencies, and by using poly-energetic x-ray spectra. This paper deals mainly with the effect of the finite phosphor thickness on signal and noise, i.e., the difference in transfer of signal and noise by the phosphor. It has been observed that for film-screen systems the shape of the noise power spectrum (NPS) does not match the give of the modulation transfer function (MTF) squared. Even after correction for film granularity, MTF falls off faster than NPS. This causes a significant reduction in the DQE at high spatial frequencies. We have found that our model can predict MTF and the shape of the quantum noise power spectrum for an ortho-M/Min-R film-screen system. Our model confirms the hypothesis that the finite phosphor thickness results in the noise being transferred more efficiently than the signal at high spatial frequencies. This is in part responsible for the decline in DQE(f) with increasing spatial frequency.

This prospective study compared images obtained using a photo-stimulable imaging plate (IP) to images obtained using a conventional film/screen (FS) combination in a total of 66 patients, 26 of whom were undergoing surgical procedures requiring intra-operative arteriography and 40 of whom were undergoing cholangiography. Exposure factors, number of repeat examinations, and elapsed time for processing each image were recorded. Diagnostic accuracy of the two techniques was assessed objectively and image quality was assessed subjectively. The radiation dose was reduced by 50% using the IP technique. Due to the greater dynamic range of the IP system, no repeat IP examinations were required for either arteriography or cholangiography, while 6 (9%) of the FS studies had to be repeated due to under- or over-exposure. IP studies required very slightly more time for processing than FS studies, an average of 35 seconds per patient. Receiver operating characteristic (ROC) analysis revealed no significant difference in diagnostic accuracy between the two imaging techniques for either the arteriographic or cholangiographic studies. Subjective evaluation also revealed no significant difference in observer preference for IP vs FS studies.

Signal-to-noise measurements have been made on storage phosphor samples from different manufacturers. Variations in quality have been found which lead to differences in performance. Some techniques to enhance phosphor performance are described.

The Toshiba Computed Radiography system (TCR) is one of the first digital x-ray imaging systems, which permits imaging of large area formats up to the chest format of 14" x 17". It uses an imaging plate, based on a storage phosphor, which is read out with the aid of a scanning laser beam after exposure to x-radiation in the patients examination room, making use of photostimulable radioluminescence (PSL). As a truly digital system, it separates the functions of detection and display such that the information displayed to the human observer can be optimized with respect to the requirements of the observer.

Computer simulation of double-exposure, dual-energy mammography is used to quantify (1) the accuracy of adipose/glandular tissue cancellation, and (2) microcalcification SNR across the tissue-cancelled mammogramme, under conditions of non-isotropic photon emission from tungsten anodes. The results show that a non-isotropic property in the dual-energy spectra, combined with aluminium/lucite calibration at the central-axis of x-ray emisssion, introduces tissue-cancellation errors of 2-3% into a mammogramme that covers a 16 degree emission angle. These errors are sufficient to mask or obscure the presence of small calcifications. The tissue-cancelled image also contains variations in SNR (of a 200 micron calcification) ranging from 0% to 15% for thick breasts and from 0% to 25% for thin breasts. These values depend on the spatial location of the calcification within the breast. The results are relevant to the design/choice of tungsten-anode x-ray tubes for dual-energy mammography and also stress the importance of breast positioning.

The application of numerical multiparameter optimization techniques tolthe design of improvqd mammographic imaging systems has been discussed by Jafroudi et al., and by Muntz et al.. The approach taken requires the specification of a reference system for dios/p and imaging performance comparisons. The optimized designs that have been reported are based on a reference system without a grid. Since those designs were developed, it has become common practice to use grids in mammography. Accordingly, the optimization has been repeated with a current state-of-the-art system, including a grid, as the reference. An approximation to the revised optimized design has been assembled and compared to a Thomson-CGR Senographe 500T mammography unit equipped with a grid. When the optimized system is operated under conditions that produce imaging performance comparable to that of the Senographe, the average glandular dose delivered by the optimized system is about 1/3 the dose delivered by the Senographe.

A digital breast imaging system is under development to provide improved detectability of breast cancer. In previous work, the limitations of screen-film mammography were studied using both theoretical and experimental techniques. Important limitations were found in both the acquisition and the display components of imaging. These have been addressed in the design of a scanned-projection digital mammography system. A high resolution x-ray image intensifier (XRII), optically coupled to a self-scanned linear photodiode array, is used to record the image. Pre- and post-patient collimation virtually eliminates scattered radiation and veiling glare of the XRII with only a 20% increase in dose due to penumbra. Geometric magnification of 1.6 times is employed to achieve limiting spatial resolution of 7 1p/mm. For low-contrast objects as small as 0.1 mm in diameter, the digital system is capable of producing images with higher contrast and signal-to-noise ratio than optimally-exposed conventional film-screen mammography systems. Greater latitude is obtainable on the digital system because of its wide dynamic range and linearity. The slit system is limited due to long image acquisition times, and poor quantum efficiency. This motivated our current work on a slot beam digital mammography system which is based on a fiber-optic x-ray detector. Preliminary results of this system will be presented.

The point spread function (PSF) is used to characterize imaging systems. The PSF is usually not measured directly but rather the line spread function (LSF) is measured by scanning across the image of an input slit. One of the well known LSF-PSF conversion formulas is then applied.1 These formulas make the assumption that the length of the input-slit image is great compared to the PSF extent. This assumption is unfortunately unwarranted for one of the most important medical imaging devices: the x-ray image intensifier. The large extent image intensifier's PSF and the limited size of the intensifier's isoplanatic patches combine to make consideration of the finite length of the input slit important. For-mulas for calculating the PSF from a measurement of the finite-length line spread function (FLSF) have been developed for the case of a rotationally symmetric PSF.3 In this presentation we generalize the conversion formulas to cover non-symmetric PSF's.

A prototype system for digital fluorography of the chest is under development in our research laboratory. It consists of a special-procedures x-ray unit upgraded with a 40-cm image intensifier, a Compumotor driven multiple-slit assembly, and MicroVAX II based frame grabber and frame processor. Contrast in an image-intensifier x-ray system is degraded by scattered radiation and veiling glare. They both increase with the diameter of the image-intensifier input phosphor. Grids, air gaps, and paired scanning apertures are not effective against veiling glare. In our prototype system both scattered radiation and veiling glare are rejected using the electronic-collimation principle. A multiple-slit assembly is scanned between the x-ray tube and the patient. Images are recorded and digitized for each position of the slit assembly. The resultant image is reconstructed from a series of digitized slit images using an appropriate computer algorithm. Technologies for manufacturing the multiple-slit assembly and the image-reconstruction algorithms are discussed. The peak-detection algorithm does not require as accurate stepping of the slit assembly as the threshold-and-summation algorithm, but it wastes a portion of the radiation dose. Comparative single- and dual-energy images of a chest phantom are presented. Superior separation of bone and soft-tissue images was achieved when the electronic scanning slit was used because, of more efficient scatter rejection.

We propose a scanning slit device for chest radiography that employs multiple scanning slits and a large area image intensifier. A novel technique for optical collimation of the light signal resulting from scattered radiation is performed at the output of the image intensifier, prior to digitization, with a high resolution CCD camera. Preliminary camera dark current measurements show that cooling of the CCD camera electronics will be necessary. Preliminary experiments indicate that this method of optical collimation is feasible.

Recent breakthroughs in electronic detector technology have allowed digital radiographic images to become competitive with, or superior to, those produced with classical film-screen techniques. A summary of these technologies is given in ref. 1. Our group is involved in the research and development of a recently proposed imaging technology (2,3) based on the kinestatic charge detector (KCD). The modulation transfer function (MTF) of the KCD technique has been discussed in refs. 1 and 4. The low frequency detective quantum efficiency (DO(0)) of several KCD designs has been modeled and a value of approximately 0.75 is expected for future detectors (1,5). In this paper, the noise power spectrum (NPS) and the frequency-dependent DOE (MEM) are discussed. Noise contributions from x-ray quanta (random and structured) and data acquisition electronics are considered and preliminary experimental results are given for a recently installed imaging detector. A brief comparison is made of our theoretical and experimental results with published results for film-screen radiography.

The ongoing development of a new class of liquid cooled rotating anode x-ray tubes capable of high average and high peak power is discussed. Tube performance is characterized by zero wait-times between exposures and essentially no derating from peak instantaneous ratings for any arbitrary exposure sequence. The results of a successful program, which demonstrated proof of principle of the novel heat transfer surface will be described. The design and physics of the heat transfer surface of an experimental tube under construction will be reviewed. Potential benefits of this new tube include higher patient throughput in CT and certain fluoroscopic modalities, and possible longer tube life than conventional designs. New high average power techniques such as slit scanning, energy subtraction, x-ray spectrum optimization and special scatter rejection methods would become more clinically practical.

We describe a new system for high Energy Imaging in Radiotherapy treatment. When using this system, a physician will be able to detect the movements of a patient during a radiotherapy treatment session by taking 18 MeV contrast images of the patient within few seconds. The system's detector is a large area, flat device made of parallel scintillating optical fibres lying on a plane perpendicular to the X-ray beam and located behind the patient. These fibres emit light through their extremities when crossed by high energy particles or photons. This light is collected with photodiodes, whose electrical signal is amplified, sampled and hold in synchronism with the pulsed X-ray beam. Then the signals are digitized by an Analog-to-Digital converter and stored in the memory of a personal computer. The detector, which can rotate around the X-ray beam axis, moves toward another selected position and the data are stored every time. Then a tomographic reconstruction algorithm is run and an image is finally produced a few seconds after its detection. We will discuss the feasability of such a system, using this kind of detectors, and show the first images obtained with this technique.

We have developed a scanned projection x-ray imaging system in which a 1 to 2 cm slot beam of radiation is used to acquire 96 image lines simultaneously, thereby reducing scatter, x-ray tube heat loading and image aquisition time. This system is designed for quantitative angiography, but the TDI technique we describe is equally applicable to other radiographic procedures such as mammography. Our system utilizes an x-ray image intensifier and a charge-coupled device operating in the time-delay-integration mode. The slot beam may be scanned to create an image in a few seconds or remain stationary in order to perform quantitative analysis of arterial flow. In this paper we describe our system, report on our investigations of imaging performance and present examples of radiographic images. We have found that the noise in our charge-coupled device camera may be reduced to about 270 noise equivalent carriers by cooling the detector, giving a dynamic range greater than 2000. The limiting spatial frequency for the system is 1.7 mm-1, restricted primarily by the performance of the x-ray image intensifier.

A research system for digital radiography is described, which is based on a selenium detector with capacitive probe readout. The detector, in which a selenium drum is used as the primary image receptor, is exposed by a scanning fan beam. Scatter reduction and primary transmission by slot-radiography as well as the imaging properties of the selenium detector are discussed. The spatial resolution and the noise behaviour of the detector are analysed. The signal-to-noise ratios expressed in terms of noise equivalent quanta and detective quantum efficiency are calculated and compared with competitive systems.

Two different prototype high resolution digital x-ray radiographic imaging systems employing planar array charge-coupled device detectors have been constructed. Both proto-type imaging units have an ultimate resolution capability of at least 10 line pairs/mm. Each system employs a standard 20 x 30 cm x-ray intensifying phosphor screen as the input receptor. One unit operates as a scanning system utilizing a specially manufactured planar array (244 x 380 elements) operating in a time delay and integration mode (TDI). The other system employs a static array (640 x 1024 15 micron square pixels) operating in "snapshot" read out mode. Our studies utilizing these devices with standard clinical radiation exposures have shown a significant improvement in the visualization of low contrast objects in radiographic phantoms and both units have an ultimate resolution capability of at least 10 line pairs/mm.

We have been developing a focal plane x-ray tomography system that will be capable of creating limited angle tomograms in real-time at rates as high as 60 per second. The intended application of the device would be in coronary angiography, where conventional tomographic systems are too slow to capture sharp images of the beating heart. We first reported on this system development three years ago. Since that time, we have been constructing and testing sub-system components while awaiting the completion of the design, construction and testing of the "Tomotron" x-ray tube, which is being manufactured by the Eimac Division of Varian Corporation. We anticipate that the x-ray tube will be ready for integration into the tomography system this spring. In this report, we would like to describe the overall system design and present some of the performance characteristics of three of the sub-system components: image deflection mechanism, x-ray collimator and anti-scatter grid.

A prototype volume CT system for use in angiography has been constructed and tested initially using a simple phantom. This system consists of a fixed x-ray tube, an image intensifier, coupled to a solid state camera, and a computer-controlled turntable on which phantoms were placed. In order to explore the imaging performance of this system, subtraction projection images, acquired over a few projection angles (20), were digitized and used for single slice reconstruction. These data then were reconstructed using two iterative reconstruction algorithms, specifically designed for vascular structures. The quality of the resultant reconstructed images agrees with the results of previous computer simulation studies and indicates that the performance of our algorithms is superior to standard filtered-backprojection. The results also suggest that an image intensifier-based detector can be used for reconstructing 3-D vascular structures from a limited number of subtraction projections.

The detection and discrimination of known signals in Gaussian noise is a well-understood problem. Many authors have pointed out that this problem leads to the formation of a test statistic that is a linear function of the data [1,2]. Of more clinical relevance is the more difficult problem of detecting a random signal in a noisy background. Barrett, Myers, and Wagner [1] and Wagner and Barrett [2] have shown that the detection of signals with random parameters in general leads to a test statistic that is a nonlinear function of the data. In this paper we will further explore the nonlinear detection strategy of an ideal observer for signals with random parameters. We will show that this strategy is intimately related to the operations the ideal observer would perform when estimating the underlying object in a noisy scene. We will review experimental evidence that shows human observer performance is inferior to the performance of the ideal observer for certain nonlinear tasks, suggesting that the human is unable to utilize nonlinear features in the detection of random objects in these cases.

A computer simulation of pre-image data obtained in a Time-of-Flight Positron Emission Tomogrophy (TOFPET) system was implemented to demonstrate the Sine Transform Image Reconstruction Algorithm (STIRA).

The problem of reconstruction in Positron Emission Tomography (PET) is basically estimating the number of photon pairs emitted from the source. Using the concept of maximum likelihood (ML) algorithm, the problem of reconstruction is reduced to determining an estimate of the emitter density that maximizes the probability of observing the actual detector count data over all possible emitter density distributions. A solution using this type of expectation maximization (EM) algorithm with a fixed grid size is severely handicapped by the slow convergence rate, the large computation time, and the non-uniform correction efficiency of each iteration making the algorithm very sensitive to the image-pattern. An efficient knowledge-based multi-grid reconstruction algorithm based on ML approach is presented to overcome these problems.

Reprojection is the process by which projections are produced from an image such that, if these projections are filtered and backprojected, they yield the original image. Because of the computational expense of reprojection, algorithms that employ this process have never been widely used. A method is presented that enables an unmodified backprojector to be used as a reprojector. Because backprojectors are designed to exploit the parallelism in the backprojection algorithm, the time required to obtain reprojections is significantly reduced. Another method, based on the Fourier Slice Theorem, is presented that enables a general purpose array processor to be used as a high speed reprojector. It is also shown that the parameters of the reprojection algorithm can be adjusted to decrease significantly the time required to perform an application that uses reprojection. Finally, two applications of reprojection in computed tomography are discussed.

Modular scintillation cameras are gamma cameras with relatively small crystal faces, a small number of photomultiplier tubes (PMTs), and independent processing electronics. Our prototypical module has a 10 cm square crystal face, four PMTs, and digital processing electronics. Scintillation event information is transferred to images by mapping digitized PMT response combinations to optimal position estimates of event locations. In our prototype, a look-up table is used to perform this mapping. To encode scintillation event information more effectively, we use nonlinear compression of each of the PMT signals. Also introduced are logarithmic matched filtering and likelihood windowing, two processing techniques that result from exploitations of the Poisson model of the distribution of photopeak events. Logarithmic matched filtering is a method of obtaining estimates of mean detector response functions having greater accuracy than that indicated by the digitization of the PMT responses. Likelihood windowing is the utilization of a likelihood threshold, rather than the familiar energy window, as a discriminant of photopeak and scatter events. We have implemented each of the above on our prototypical module. Performance characteristics of this module include energy resolution of 10% full width at half maximum (FWHM) at 140 keV and spatial resolution of better than 4mm FWHM over 90% of the crystal.

The application of fractal concepts to the analysis of non-linear dynamics and morphology has expanded our insight into many diverse natural phenomena. Fractal mathematics provides new methods of analysis also applicable to biophysical phenomena including the structure and function of systems comprising the human body. The brain, heart and the tracheo-bronchial tree possess characteristics common to fractal objects including: (a) a large degree of heterogeneity, (b) self-similar structures over many size scales, and (c) no well defined (characteristic) scale of measure. The fractal dimension, DF is a measure of the structural complexity. This paper presents an overview of some of the general concepts underlying fractals and their relationship to non-linear dynamics and morphology. Areas of investigation that benefit from the application of these concepts to biological phenomena and modeling are discussed and an algorithm for modeling lung development based on fractal concepts is presented. Structures that are in good agreement with actual morphological data may be generated using simple recursive algorithms and constraints.

A method for evaluating image-recovery algorithms is presented, which is based on the numerical assessment of how well a specified visual task may be performed using the reconstructed images. A Monte Carlo technique is used to simulate the complete imaging process including the generation of scenes appropriate to the desired application, subsequent data taking, image recovery, and performance of the stated task based on the final image. The use of a pseudo-random simulation process permits one to assess the response of an image-recovery algorithm to many different scenes. Nonlinear algorithms are readily evaluated. The usefulness of this method is demonstrated through a study of the algebraic reconstruction technique (ART), which reconstructs images from their projections. In the imaging situation studied, it is found that the use of the nonnegativity constraint in ART can dramatically increase the detectability of objects in some instances, especially when the data consist of a limited number of noiseless projections.

Maximum likelihood algorithms have the potential of producing improved estimates with high accuracy. The computational requirements often become prohibitively expensive, however, because of the slow rates of convergence of these algorithms. This paper discusses three classes of acceleration techniques by which convergence to the maximum likelihood solution can be sped up. Some characteristics of these acceleration methods are examined.

Constrained least-squares techniques have been used to produce a well known class of image restoration algorithms. These techniques typically involve minimizing a linear operator on a vector representation of an image, subject to a constraint. For cases where an equality constraint is appropriate the method of Lagrange multipliers can be used to produce a restored image. In this work a fractal textural model,fractional Brownian motion, is used to represent images of interest. Using a variance fractal dimension estimator a non-linear operator, that represents the squared difference between the fractal dimension of the restored image and an a priori value is minimized, subject to the constraint that the norm of the residual between the restored image and available measurement equal the norm of the additive noise.

Measurement of the power spectra of liver scans reveals that the radiocolloid distribution in the human liver behaves as a fractal object. Analysis of the power spectra suggests that the fractal dimension of the functional units of the liver changes with disease state, and that power spectral slope may be a useful classifier for the presence of disease. Models are proposed that relate the power spectral slope to the fractal dimension of the liver parenchyma.

We propose a technique for automatic, anatomically selective, artifact-free enhancement of digital chest radiographs. Anatomically selective enhancement is motivated by the different enhancement requirements of the lung field and the mediastinum. A recent peak detection algorithm is applied to the image histogram to automatically determine a gray-level threshold between the lung and mediastinum fields. The gray-level threshold facilitates anatomically selective gray-scale modification and unsharp masking. Further, in an attempt to suppress possible white-band artifacts due to unsharp mask-ing at sharp edges, local-contrast adaptivity is incorporated into anatomically selective unsharp masking by designing an anatomy-sensitive emphasis parameter that varied asymmetrically with positive and negative values of the local image contrast.

We have employed physical measures of lung texture in an automated method of detecting and characterizing interstitial lung disease in digital chest radiographs. In addition, by using an analysis of these measures relative to an accumulated data base, we have devised an automated classification method for distinguishing between normal lungs and abnormal lungs with interstitial disease. Our results suggest that this computerized method can be a valuable aid to radiologists in their assessment of interstitial lung infiltrates.

The double-square-box method of tracking, in conjunction with spatial correlation techniques, is employed for accurate determination of the three-dimensional position of the vessel centerline, as well as the vessel size and contrast, from stereoscopic coronary arteriograms. In order to facilitate systematic identification and correlation of the vessel information in multiple images, we present a method with which the vessel segments in the vascular tree can be identified accurately in terms of their position in the hierarchical structure using the size, contrast, direction, and branching relationships of the vessels.

In the diagnosis of arteriosclerosis, radio-opaque dye is injected into the interior of the arteries to make them visible. Because of its increased contrast sensitivity, digital subtraction angiography has the potential for providing diagnostic images of arteries with reduced dye volumes. In the conventional technique, a mask image, acquired before the introduction of the dye, is subtracted from the contrast image, acquired after the dye is introduced, to produce a difference image in which only the dye in the arteries is visible. The usefulness of this technique has been severely limited by the image degradation caused by patient motion during image acquisition. This motion produces artifacts in the difference image that obscure the arteries. One technique for dealing with this problem is to reduce the degradation by means of image registration. The registration is carried out by means of a geometrical transformation of the mask image before subtraction so that it is in registration with the contrast image. This paper describes our technique for determining an optimal transformation. We employ a one-to-one elastic mapping and the Jacobian of that mapping to produce a geometrical image transformation. We choose a parameterized class of such mappings and use a heuristic search algorithm to optimize the parameters to minimize the severity of the motion artifacts. To increase the speed of the optimization process we use a statistical image comparison technique that provides a quick approximate evaluation of each image transformation. We present the experimental results of the application of our registration system to mask-contrast pairs, for images acquired from a specially designed phantom (described in a companion paper), and for clinical images.

In this paper a method to quantitate blood flow from contrast injections into a vessel was evaluated using a straight tube, constant flow phantom. Grey scale values integrated over multiple regions-of-interest (ROI), placed contiguously along a "vessel" segment, were used to generate multiple time-density curves. Temporal shifts in the time-density curves for each ROI were determined using a cross-correlation technique. Regional volume estimates were made using a curve fitting approach. Linear regression of the regional volume and temporal shift estimates was used to estimate flow within the vessel segment. Velocity estimates were made in a similar way. Currently, velocity estimations are significantly better than flow estimations. Errors in the diameter estimates remain as the main cause of flow inaccuracies.

We make a comparison of the performances of various three-dimensional reconstruction algorithms for situations where only few conic projections of a vascular tree are available. This problem is ill-posed and prior information must therefore be used to regularize the solution. We restrict ourselves to methods that are able to handle the sparseness and the non-negativity that caracterize a iodinated vascular structure: the Extreme Value Technique and related methods, and the Algebraic Reconstruction Technique. The results we obtained led us to derive a new method based on a two steps detection-estimation scheme.

Blood vessel delineation on digital subtraction angiograms (DSA) is an essential step in the automatic localization and quantification of blood vessel abnormalities. In this paper, we first describe a general framework for object delineation on arbitrary images. We conclude that generic models and continuous delineation, which mostly implies returning to the image, are important characteristics of this framework. Subsequently, we discuss the knowledge-based system for blood vessel delineation that we are currently developing. The system starts from both a single angiogram and a stereoscopic or an orthogonal pair of angiograms. The use of a stereoscopic or orthogonal pair of images requires that a 3-D reconstruction be performed. We will see that this 3-D reconstruction is not considered as an independent process, separated from the delineation (or interpretation) process. Instead, we integrate both processes and exploit the knowledge obtained from the delineation to improve the reconstruction, and vice versa.

In order to derive accurate values for true tissue radiotracers concentrations from gated positron emission tomography (PET) images of the heart, which are critical for quantifying noninvasively regional myocardial blood flow and metabolism, appropriate corrections for partial volume effect (PVE) and contamination from adjacent anatomical structures are required. We therefore developed an integrated software package for quantitative analysis of tomographic images which provides for such corrections. A semiautomatic edge detection technique outlines and partitions the myocardium into sectors. Myocardial wall thickness is measured on the images perpendicularly to the detected edges and used to correct for PVE. The programs automatically correct for radioactive decay, activity calibration and cross contaminations for both static and dynamic studies. Parameters derived with these programs include tracer concentrations and their changes over time. They are used for calculating regional metabolic rates and can be further displayed as color coded parametric images. The approach was validated for PET imaging in 11 dog experiments. 2D echocardiograms (Echo) were recorded simultaneously to validate the edge detection and wall thickness measurement techniques. After correction for PVE using automatic WT measurement, regional tissue tracer concentrations derived from PET images correlated well with true tissue concentrations as determined by well counting (r=0.98). These preliminary studies indicate that the developed automatic image analysis technique allows accurate and convenient evaluation of cardiac PET images for the measurement of both, regional tracer tissue concentrations as well as regional myocardial function.

The anatomical three-dimensional (3D) medical imaging modalities, such as X-ray CT and MRI, have been well recognized in the diagnostic radiology for several years while the nuclear medicine modalities, such as PET, have just started making a strong impact through functional imaging. Though PET images provide the functional information about the human organs, they are hard to interpret because of the lack of anatomical information. Our objective is to develop a knowledge-based biomedical image analysis system which can interpret the anatomical images (such as CT). The anatomical information thus obtained can then be used in analyzing PET images of the same patient. This will not only help in interpreting PET images but it will also provide a means of studying the correlation between the anatomical and functional imaging. This paper presents the preliminary results of the knowledge based biomedical image analysis system for interpreting CT images of the chest.

This paper describes a prototype system for identifying and characterizing Multiple Sclerosis (MS) lesions in the brain from magnetic resonance (MR) images. The system is designed to obtain an initial segmentation of each cross-sectional image with low level vision methods, and then derive successive refinements of image subregions through a model-driven approach that correlates relevant information from T1 and T2 images and 3-D information from complementary cross-sections when necessary. The system uses a b-spline surface model of the brain that matches the characteristics of the individual's brain. The normal internal structures of the brain are then scaled proportionately before carrying out the successive refinement operations for the detection of the MS lesions. The low level vision and the solid modeling components of the system have been successfully tested on several hundred images from a number of MR patient studies. The first steps of model fitting have been implemented and show promising results.

We present means of interactive definition of anatomic objects in medical images via a description of the image in terms of visually sensible regions. The description is produced by computing structures capturing image geometry and following them through the image simplification of Gaussian blurring. In particular, we suggest that the structure made from intensity "ridge" and "course" curves defined by the locus of intensity level curve vertices, augmented by the pile of internal and external symmetric axes of these level curves, satisfies desirable criteria for a structure on which to base such object definition.

In Imaging the human brain, MRI is commonly used to reveal anatomical structure, while PET is used to reveal tissue function. This paper presents a protocol for correlating data between these two imaging modalities; this correlation can provide in vivo regional measurements of brain function which are essential to our understanding of the human brain. We propose a general protocol to standardize the acquisition and analysis of functional image data. First, MR and PET images are collected to form three-dimensional volumes of structural and functional image data. Second, these volumes of image data are corrected for distortions inherent in each imaging modality. Third, the image volumes are correlated to provide correctly aligned structural and functional images. The functional images are then mapped onto the structural images in both two-dimensional and three-dimensional representations. Finally, morphometric techniques can be used to provide statistical measures of the structure and function of the human brain.

A simple algorithm for geometric registration of tomographic images was developed to be applied to images of the heart obtained by Positron Emission Tomography (PET). Commonly, image misregistrations are not only caused by simple positioning differences like translation and rotation, but also by deformations that cannot be expressed by simple geometric transformations. The algorithm that we developed is a combination of two methods: a "rigid" registration followed by an "elastic" deformation. A variable number of reference points is used to register the images. The first part of this method consists of calculating the best match between the reference points of the two images based on the evaluation of the center of gravity, the axes of inertia and the surface extension. A set of geometric rotation, translation and scaling operations are then applied to the image to match. In the second phase an elastic model is further applied to perfectly match the reference points. An algorithm based on a weighted average of the displacement of every reference point is used. This second phase produces local deformations of the image to match every reference point. A set of validation experiments was performed using digitized images of isolated slices of a dog's heart where several deformations were applied. Further experiments on PET images showed that this method provides adequate automatic 2D registration of the tomographic images. The same approach could potentially be used to match images from different imaging modalities like MRI or CT scans.

Improvements in the accuracy of repositioning patients in medical imaging systems during repeat examinations will allow a more precise measurement of the progress of the disease or treatment. The alignment task is normally carried out using a number of inadequate techniques varying from stereotatic frames, to a rigorous anatomical study of the organs shown in the different views. Present techniques are either unreliable or need total patient collaboration for the surgical implantation of a localising device. The paper describes a knowledge based approach, which will enable optimal matching to be achieved between the two data sets and will be able, at least in the case of MRI images, to provide the appropriate co-ordinates for an optimised new slice angle. An extension of the use of accurate repositioning would be the ability to cross match the different types of information from other imaging systems. The system will eventually be able to quantify the absolute difference between images subject to morphological change and temporal distortion. The method of approach uses an anatomical knowledge base to guide the segmentation of the scene into a number of clearly identified invariant key objects. The matching will proceed by iterating to progressively smaller features. Matching is carried out using symbolic feature spaced descriptions of the objects. Key words: Image matching, high level reasoning, stereotatic frames, MRI.

Mathematical morphology operations allow object identification based on shape and are useful for grouping a cluster of small objects into one object. Because of these capabilities, we have implemented and evaluated this technique for our study of Alzheimer's disease. The microscopic hallmark of Alzheimer's disease is the presence of brain lesions known as neurofibrillary tangles and senile plaques. These lesions have distinct shapes compared to normal brain tissue. Neurofibrillary tangles appear as flame-shaped structures, whereas senile plaques appear as circular clusters of small objects. In order to quantitatively analyze the distribution of these lesions, we have developed and applied the tools of mathematical morphology on the Pixar Image Computer. As a preliminary test of the accuracy of the automatic detection algorithm, a study comparing computer and human detection of senile plaques was performed by evaluating 50 images from 5 different patients. The results of this comparison demonstrates that the computer counts correlate very well with the human counts (correlation coefficient = .81). Now that the basic algorithm has been shown to work, optimization of the software will be performed to improve its speed. Also future improvements such as local adaptive thresholding will be made to the image analysis routine to further improve the systems accuracy.

There is a growing need to study and examine microscope slides in various fields of science. However, this task can be cumbersome and vulnerable to human error. This is especially true when large numbers of slides have to be examined. It is apparent that the existing computer and image processing technology should be utilized to speed up the process of cell examination. In our research, an existing method of image segmentation, called pyramid node linking, has been applied with a few modifications to cell segmentation. In pyramid node linking, a pyramid is constructed by successively reducing the resolution of the original image by factors of two to obtain the first and subsequent levels, until there are four pixels left on the top most level. In a bottom-to-top iterative process, the pixels from level to level are linked using information from the level above, the level below, and from the neighbor pixels on the same level. This results in several trees with roots at one of the upper levels and leaves on the original image. This process results in smooth simply-connected regions with well-defined boundaries on the bottom level. We have applied pyramid node linking to complex images consisting of clusters of rat liver cells grown in culture and damaged to different degrees by exposure to various chemicals. The algorithm has been applied to classify the rat liver cells in three categories: undamaged, slightly damaged, and disintegrated.

Analysis of the large amounts of image data obtainable from very-high-speed scanning laser microscopes places severe demands on computer software and hardware architectures. The automated calculation of features over entire images can provide quantitative data useful to a pathologist who must make a diagnosis. A program that identifies objects of diagnostic interest in an image must utilize a model of the image. An expert system is an effective method for building abstract models of object hierarchies and for utilizing heuristic information. In this paper we discuss a composite approach to image understanding and assessment that utilizes an expert system to control a set of image processing functions for the recognition of various objects in an image.

Automated cervical smear screening is potentially a cheap, rapid method for detecting early cervical cancer and so preventing deaths from this disease. CERVIFIP is a fast scanning machine which classifies microscope images of cervical smears. Although its current false negative error rate is below that of cytotechnicians, its false positive error rate is unacceptably high. New boundary and grey-level texture algorithms are being applied to reduce these rates. In addition a hierarchical rules-based classifier is to be superimposed on the existing statistical classifiers.

Local spatial filters were developed to enhance and extract blood vessels in a digitized color non-fluorescein angiograph. One filter dynamically enhances a black/white digital picture by normalizing the pixels relative to their local neighborhoods, and another filter extracts the candidate blood vessel pixels from a black/white digital picture.

This research program successfully developed a real-time video imaging system (the Imaging Burn Depth Indicator, or IBDI) which can discriminate areas of burn wounds expected to heal in three weeks or less from the day of injury from those areas not expected to heal in that time period. The analysis can be performed on or about the third day post-burn on debrided burn wounds. Early evaluation of burn healing probability is a crucial factor in the decision to tangentially excise the burn wound. The IBDI measures the reflectivity of the burn wound in the red, green, and near infrared wavelength bands, which data correlate with burn healing probability. The instrument uses an algorithm established in an earlier study to translate the optical data into burn healing probabilities. The IBDI produces two types of images: a true-color image of the burn and a false-color image of the burn. The false-color image consists of up to four colors, each of which indicates a distinct range of probability that the area of the burn so colored will heal within 21 days. Over 100 burn wound sites were studied. Burn sites were evaluated on day three post-burn by our instrument and by the attending physician. Of 55 sites considered to be of intermediate depth, the IBDI predicted the healing outcome accurately in 84% of the cases. By comparison, the predictions of burn surgeons supervising the care of these patients were accurate in 62% of the cases.

We have developed a prototype digital imaging colposcope system for use in clinical, research and teaching environments. A goal of this system is to aid in earlier detection of cervical pathology and to assist in more precise site directed biopsies. The system is expected to provide a valuable research and teaching tool, as well as a mechanism for including colposcopic images in proposed medical Picture Archiving and Communications Systems (PACS). The system consists of a charge coupled device (CCD) imager attached to the camera port of a conventional photocolposcope, a microcomputer host and a software package for processing of the resultant digital images. It is envisioned that this basic system could eventually be reconfigured to accommodate advantages to be derived from an integrated expert system. We discuss the hardware and software system, review some of the analysis algorithms used for this application and present preliminary results of clinical use of the system.

We have developed a unique prototype solid state digital imaging system for use in the analysis of one dimensional deoxyribonucleic acid (DNA) sequencing gels. The system can be used to digitize, interpret and store DNA sequence information directly from an autoradiogram. The system consists of a two dimensional charge coupled device (CCD) imager, a multiplexing optical front end for reformatting of the gel image, a microcomputer host and a software package for reconstruction of the gel image and analysis of the DNA sequence.

We have developed a prototype microcomputer based gel reader for the digitization and semi-automatic analysis of two dimensional polyacrylamide electrophoretic gels. The system uses a 2D charge coupled device (CCD) detector and operates as a peripheral to an IBM PC XT/AT type host computer. The system incorporates a comprehensive software package which lends itself to highly structured clinical applications as well as a wide range of image processing functions for use in the research laboratory. The system is undergoing further development in order to arrive at a comprehensive, low cost, early disease screening device. Electrophoretic gel samples of protein components obtained from subject patients are digitized and searched for anomalous features. These features are located using a digital reference library of disease related protein features and a probability of detection of the disease related protein is computed. The system then presents the physician with a report advising him of the results of the analysis. We describe the hardware and software system and present results of a sample gel analysis.

A computer program to simulate the three dimensional impulse response of a convergent focus rectangular tube collimator system has been developed. The simulation is done by a deterministic algorithm that models geometric acceptance of photons, photon penetration of the septa, and single scatter interactions in the collimator walls. Output from the simulator is used as a basis set for the impulse response of a multiple detector SPECT system. The basis set is used by fitting procedures that estimate the impulse response of the Donner SPECT system (12 angles, 12x 128 detector positions per angle) from noisy measurements of a point source. These impulse response estimates are used to determine the weighting matrix of new emission reconstruction algorithms.

Researchers have long realized that radiographic images can be decomposed into two "component images," because there exist two predominant interactions that account for x-ray attenuation in the diapostic range of x-ray energies - Compton scattering and photoelectric absorption. Decomposition into component images is achieved through linear (or higher order) combination of radiographic images recorded using x-ray beams with differing effective energies. The component images usually chosen by researchers are those in which either soft tissue or bone is suppressed. The motivation for creating radiographic images free from either bone or soft tissue image contrast is to simplify the task of interpreting radiographic image information. By removing irrelevant structures unrelated to the likely disease being diagnosed, it is argued, detection accuracy may increase. Correspondingly, the efficiency with which soft tissue nodules (tumors) are detected from chest radiographs may be increased by removing image contrast that results from bony structures such as the ribs, whose presence may "mask" the presence of a tumor. This study focuses on dual energy subtraction techniques that produce images in which bone contrast is suppressed.

A fan beam convolution backprojection algorithm has been developed for use with any spherically-shaped detector. This work represents the first step in the development of a cone beam reconstruction algorithm to be used with a prototype volume CT device that employs a conventional image intensifier as its detector. This algorithm has been tested for a high contrast vessel phantom reconstruction using real x-ray projection data acquired with our device.

This paper describes preliminary performance data of the components of a high resolution x-ray imaging device, under development at the University of Arizona through NIH sponsorship for application in coronary angiography. The system will consist of an external modular x-ray sensor, a proximity focussed image intensifier and six CCD's coupled to the output of the image intensifier via six fiber optic tapers. The tapers are joined at the large ends to form a coplanar fiber optic taper assembly.One taper has been delivered whith distortion of less than 1.2 %.One channel of the camera and display electronics is operational. The limiting resolution of a pilotsystem, consisting of a 40 mm Gen.II image intensifier, the fiber optic taper and a CCD is 3.9 1p/mm, the Nyquist frequency of the CCD magnified by the taper.

A new digital radiographic system (Matrix Laser Rad) employing a filmless photostimulable reusable phosphor (PSP) was evaluated for spatial resolution and high and low contrast object detectability. Receiver operator characteristic curve (ROC) comparisons were made to film images using equal or lower exposure factors. The low and high contrast test objects for the ROC evaluation were respectively a 5/32 inch diameter hemispheric depression filled with mineral oil and 250-500 micrometer diameter silicon carbide granules, both in a 2 inch thick plexiglass block. These object sizes were shown to be at thresh-old of detectability for either the film or PSP system at the exposure factors used. The ROC curves were generated from data obtained by radiologists viewing multiple film radiographs and CRT displays of both the digitized film radiographs and PSP images with the test objects in random locations. There were 90 images generated for each system, 45 with test objects and 45 without. Each observer had 15 seconds at a fixed distance to view each image presented at random at standard view box intensities (140 foot-lamberts) on either a light view box or CRT. The unenhanced filmless PSP radiographs had a reduced detectability threshold for the test objects, however with the use of windowing, level manipulations and appropriate filtering, detectability could be improved for the low contrast objects with the digital system surpassing unenhanced film radiographs. In addition, a dose reduction could be achieved with the PSP filmless system without loss of significant detectability. Enhanced film radiographs demonstrated the highest detectability characteristics. To achieve similar detectability for high contrast objects with the PSP system, a significant increase in object size was necessary.

An analytical solution for the focal spot temperature is developed for the case of an exponential decay of the electron flux in the anode. The influence of volume heating on the surface temperature is dominated by the parameter (w/δ)2 where w is the focal spot width and δ is a mean depth of penetration. For typical operating conditions, the influence of volume heating becomes important when the focal spot width is less than 500 microns.

Comparative studies of the diagnostic capability between digitized and conventional radiographs of the chest were performed by image scientific study and clinical evaluation. The purpose of this study is to confirm the diagnostic capability of digitized image whether it can be used for primary diagnosis in routine works. The results of two studies show good correlation each other. It is suggested that the conventional chest radiographs should be digitized with 100 μm pixel size in 12 bits density resolution at least.

Current digital imaging modalities in the medical field incorporate parallel hardware which is heavily used in the stage of image formation like the CT/MR image reconstruction or in the DSA real time subtraction. In order to make image post-processing as efficient as image acquisition, new software approaches have to be found which take full advantage of the parallel hardware architecture. This paper describes the implementation of a two-dimensional median filter which can serve as an example for the development of such an algorithm. The algorithm is analyzed by viewing it as a complete parallel sort of the k pixel values in the chosen window (kernel) which leads to a generalization to rank order operators and other closely related filters reported in literature. A section about the theoretical base of the algorithm gives hints how to characterize operations suitable for implementations on pipeline processors and the way to find the appropriate algorithms. Finally some results about computation time and usefulness of median filtering in radiographic imaging are given.

As part of our ongoing effort to develop an automated computer scheme for the detection and analysis of microcalcifications in digital mammograms, we have analyzed the physical characteristics of microcalcifications from a data base of 39 clinical mammograms in patients undergoing breast biopsy. A signal-extraction method was developed for determination of the size, contrast, and signal-to-noise ratio (SNR) of each microcalcification from unprocessed mammograms. The average power spectrum of the microcalcifications thus extracted was compared to that of the mammographic background. Based on an analysis of these characteristics, we designed a new type of spatial filter, obtained as the difference between a matched filter and a box-rim filter, that can selectively preserve the frequency content of microcalcifications while suppressing the low-frequency background and high-frequency noise. The SNR of the microcalcifications is thereby enhanced. Signal-extraction tests that make use of the size, contrast, local frequency content, and clustering properties of microcalcifications were employed for further discrimination between true signals and normal mammographic structures or artifacts. In order to evaluate the potential clinical utility of our approach, we applied the program to 20 clinical mammograms that contained subtle clustered microcalcifications. These mammograms were not included in the data base mentioned above. The automated computer detection scheme provided a true-positive cluster detection rate of 90% at a false-positive detection rate of one-half cluster per image. These results demonstrate the feasibility of using computer methods to aid radiologists in screening of mammograms for subtle microcalcifications.

Several edge detection algorithms are examined for their utilities in edge or surface searching in medical images. Algorithms that are under study include those methods developed by Sobel, Roberts, Kirsch, Marr-Hildreth, Haralick, Shen-Castan, and Nalwa-Binford. An example of boundaries detected by applying these algorithms to brain images acquired by various imaging modalities is presented.

We have previously reported the creation of software to perform elastic matching of medical images, for example, to compare an idealized atlas with a set of computer tomography (CT) images. In order to evaluate the performance of this software, we have created a digitized atlas from a young normal brain, using 135 myelin-stained sections at 700 micron spacing. Software was written in C on a Hewlett-Packard workstation to allow the entering and editing of regional anatomical contours. The 2-D contours are stacked to create a 3-D atlas that can be rotated, scaled, and resliced in the three standard imaging planes. For each patient being analyzed, an individualized atlas is created from this idealized atlas by elastically matching the atlas to the patient's CT scans. Matching is achieved in two steps - global registration first, followed by deformation of the atlas to match the contours of the CT brain. This is done iteratively at coarse, medium and high resolution to achieve the best results. During this process, all regional contours are also dis-placed and deformed. We have evaluated how well the computer places these regional contours by having four experts outline several subcortical structures on the CT scans of six patients. Their degree of overlap and the variability in positioning were measured and compared with the placement of the same regions of interest by the computer. In one third of the structures there was no difference between the computer and the experts. In 18% of the regions the computer-defined region was closer to the truth than at least one of the experts. When the computer differed, it was usually by 2-3 mm in both x and y, and frequently the computer-defined region was inscribed within the expert's. This is a preliminary test of the system, using only one set of elastic coefficients, one processing variant, and only subcortical structures. The results are promising and techniques are being implemented to overcome any current deficiencies. A more systematic evaluation of the entire process, exploring all the variables that affect the accuracy of elastic matching and adding cortical regions of interest, will follow in the near future. The elastic matching produces an individualized template of regions that can be super-imposed on the anatomical images. The user may then interactively modify these regions based on the observed anatomy, delete or add regions, and extract data from these user-adjusted overlays. The flexibility inherent in this approach allows data analysis with both standard and non-standard approaches, with regions defined both by the anatomy and by areas of functional activity.

Systems built around TRW's Image Resampling Sequencer (IRS) chip can rapidly pan, zoom, rotate, warp, and filter two-dimensional images, compute 2-D sections through 3-D images, and filter 3-D image databases. These capabilities are useful in a number of medical image processing applications, including tomography. This presentation will illustrate how an image transformation system works, what is does to an image, and therefore what artifacts or limitations to anticipate in image transformation and filtering.

A sequential analysis method for angiograms is implemented and explained. A multi-resolution technique is used to segment the subtracted angiogram. The system is an improved version of our earlier work. A two step process for flow computation and interpretation is explained. First a bi-directional segmentation is applied to the image followed by the computation of the peak arrival time associated with the indicator dilution. Methods are suggested to infer flow information and quantified stenosis from integration of the peak arrival time map with densitometric information.

An efficient algorithm has been developed for the detection of clustered micro-calcifications in digitized film-screen mammograms. The calcification cluster detection algorithm involves a decision-theoretic mechanism which identifies calcifications meeting specific size, density, and clustering criteria. The algorithm first utilizes a pixel testing and region grouping procedure to identify pixels with specific intensity characteristics. A region growing procedure is then applied to identify pixels belonging to contiguous structures. Each single- and multi-pixel structure is subsequently tested against criteria for size, shape, contrast, and intensity to determine if the composite structure matches the expected characteristics of mammographic calcifications. Detected calcifications are then tested to determine if they fall within a cluster meeting specific density constraints. Only calcifications located in such clusters are marked in the processed image. This algorithmic process has been applied to mammograms from 40 clinical cases in which a biopsy was performed based on the presence of suspicious calcifications. The algorithm accurately identified the suspicious calcifications in each situation; in several cases, it also detected other clusters which were so fine as not to be visible in clinical screening.

We are develo1pipg feature-extraction techniques for use in the computer-aided detection of pulmonary nodules in digital chest images. Use of such a computer-aided detection scheme, which would alert radiologists to the locations of suspected lung nodules, is expected to reduce the number of false-negative dieggoses. False-negative diagnoses (i.e., misses) are a current problem in chest radiology with "miss-rates" as high as 30%. This may be due to the camouflaging effect of surrounding anatomic background on the nodule, or to the subjective and varying decision criteria used by radiologists.

An algorithm has been developed to determine the surface of an organ in vivo from the 3-D discrete intensity measurements of radionuclide uptake provided by Single Photon Emission Computed Tomographic (S.P.E.C.T.) data. The surface, which need not be convex, is covered by triangular tiles whose vertices are determined to lie on the organ boundary from examination of a density function along lines normal to the organ surface. This approximation is repeatedly refined by replacing the longest edge, and the two tiles that share it, with two new edges sharing a new vertex near the midpoint of the replaced edge, and four new tiles. Volume is defined as the sum of the products of the area of each tile and the dot product of its outward unit normal and centroi d vector. The tile and vertex data also lend themselves to a computationally straightforward projective display of the tiled surface, viewed and illuminated from arbitrary angles. A cine loop of these projections at equally spaced viewing angles provides an informative representation of organ structure and function.

The decreasing cost of computing power and the introduction of low cost imaging boards justifies the increasing number of applications of digital image processing techniques in the area of biomedicine. There is however a large software gap to be fulfilled, between the application and the equipment. The requirements to bridge this gap are twofold: good knowledge of the hardware provided and its interface to the host computer, and expertise in digital image processing and analysis techniques. A software package incorporating these two requirements was developped using the C programming language, in order to create a user friendly image processing programming environment. The software package can be considered in two different ways: as a data structure adapted to image processing and analysis, which acts as the backbone and the standard of communication for all the software; and as a set of routines implementing the basic algorithms used in image processing and analysis. Hardware dependency is restricted to a single module upon which all hardware calls are based. The data structure that was built has four main features: hierchical, open, object oriented, and object dependent dimensions. Considering the vast amount of memory needed by imaging applications and the memory available in small imaging systems, an effective image memory management scheme was implemented. This software package is being used for more than one and a half years by users with different applications. It proved to be an efficient tool for helping people to get adapted into the system, and for standardizing and exchanging software, yet preserving flexibility allowing for users' specific implementations. The philosophy of the software package is discussed and the data structure that was built is described in detail.

Morphological filters are nonlinear signal transformations that operate on a picture directly in the space domain. Such filters are based on the theory of mathematical morphology as formulated by Matheron [1] and Serra [2,3]. Sternberg [4,5] generalized earlier results to include graytone images. He also introduced the "rolling ball" morphological operator and pointed out that it can be used for edge enhancement. This paper reports on an variation of the rolling ball algorithm. An introduction to some of the concepts used here was given by Herman [6]. The filter being presented here features a "mask" operator (called a "structuring element" in some of the literature) which is a function of the two spatial coordinates x and y. The two basic mathematical operations are called "masked erosion" and "masked dilation". In the case of masked erosion the mask is passed over the input image in a raster pattern. At each position of the mask, the pixel values under the mask are multiplied by the mask pixel values. Then the output pixel value, located at the center position of the mask, is set equal to the minimum of the product of the mask and input values. Similarly, for masked dilation, the output pixel value is the maximum of the product of the input and the mask pixel values. The two basic processes of dilation and erosion can be used to construct the next level of operations the "positive sieve" [4] (also called "open-ing") and the "negative sieve" ("closing"). The positive sieve modifies the peaks in the image, whereas the negative sieve works on image valleys. The positive sieve is implemented by passing the output of the masked erosion step through the masked dilation function. The negative sieve reverses this procedure, using a dilation followed by an erosion. Each such sifting operator is characterized by a "hole size". If one considers a two dimensional image as a three dimensional function or surface over the two coordinates x and y, then a positive sieve will eliminate all peaks of this surface which have a cross section parallel to the x-y plane which is smaller than the hole size. Conversely, a negative sieve will clip all valleys smaller than the hole size. The hole size of a masked sifting operator in any given direction is equal to the size of the mask minus one pixel in that direction. It will be shown that the choice of hole size will select the range of pixel detail sizes which are to be enhanced. The shape of the mask will govern the shape of the enhancement. Finally positive sifting is used to enhance positive-going (peak) features, whereas negative sifting enhances the negative-going (valley) landmarks.

Ever since magnetic resonance (MR) invaded the medical imaging field, it has played an increasingly important role and is even currently being considered for stereotactic guidance of probes in the brain. While MR pictures indeed convey more clinical information than CT, the geometry of MR pictures is, unfortunately, not as accurate as the gemetry of CT pictures. In other words, if a square grid phantom is scanned, then the CT picture will show a square grid, while the MR picture will rather reveal a distorted grid. This distortion is primarily due to small variations in the static magnetic field. This small distortion does not impede radiological diagnosis; however, it is a source of concern if one contemplates utilizing the MR pictures for accurate stereotactic positioning of a probe at a very precise point in the brain. Another area of application where the distortion of the MR picture should be compensated for is the superposition of CT and MR pictures so that both informations could be used for diagnosis or stereotactic purposes. This paper essentially addresses the nonlinear distortion of MR pictures and how it could be compensated for through software manipulation of the MR picture.

Acetabular angle which is a diagnostic parameter of congenital hip dislocation has been measured manually in conventional X-ray film system. Using digital image directly provided from a computed radiography, an automatic measurement system was developed for this parameter. The process of the measurement was completed within a reasonable time, and accurate enough. The system was combined with an image database, so that it would be a measurement tool of PACS.

An algorithm was designed to discriminate tissue types, including pathology, utilizing 3D data sets acquired with Magnetic Resonance. This procedure can be adapted to different segmentation problems and works automatically for the whole data set after an interactive training on one representative slice out of the volume. 3D surface reconstruction with ray tracing methods is used to visualize different soft tissue types, according to the classification procedure to yield optimal anatomical and topographical relationships.

This manuscript describes an independent workstation which consists of a data acquisition and transfer system, a host computer, and a display and record system. The main tasks of the workstation include the collecting and managing of a vast amount of data, creating and processing 2-D and 3-D images, conducting quantitative data analysis, and recording and exchanging information. This workstation not only meets the requirements for routine clinical applications, but it is also used extensively for research purposes. It is stand-alone and works as a physician's workstation; moreover, it can be easily linked into a computer-network and serve as a component of PACS (Picture Archiving and Communication System).

A probabilistic formulation of several statistical models of the a priori source information is presented with maximum entropy analysis on individual source element behavior and Bernoulli counting process on source strength correlation. The maximum likelihood and maximum entropy image restoring of Frieden is shown to be a special case of these models when each source element has a uniform probabilistic distribution with all of them having the same constrained distribution range in treating data as a likelihood constraint. A maximum a posteriori probability analysis of incorporating both the a priori source and data information into account is extensively studied. Iterative imaging algorithms are derived by employing the expectation-maximization technique of Dempster et al. These algorithms are applied to computer generated and experimental radioisotope phantom imaging data. Improved images are obtained, compared to that of standard maximum likelihood algorithms of Shepp et al and Lange et al.

Two different a priori source probabilistic information functions are formulated with estimated probable strengths and variances of source elements. Correspondingly, two solutions for maximizing a posteriori probability with the different a priori source information are presented via Bayes' Law. Iterative imaging algorithms for the solutions are derived by employing the expectation-maximization technique of Demspter et al. These imaging algorithms are applied to computer generated and experimental phantom imaging data and improved images are obtained, compared to that of standard maximum likelihood algorithm.

The determination of percent stenosis of coronary arteries is an important task in medicine. In this paper we discuss three different algorithms which can be used in conjunction with videodensitometry to measure this quantity. These algorithms may be used in subtracting the background under a vessel segment thus eliminating the need for a preinjection mask. Mathematical details of the algorithms and experimental results are presented.

We propose a new method which enables improved spatial resolution of a reconstructed image from a gamma camera emission CT(SPECT). SPECT images are blurred due to the effect of the collimator aperture. The blur effect changes with the geometrical structure of the collimator and with the distance between the rotational center of the gamma camera and the collimator surface. The acquired projection, affected by the collimator aperture, can be assumed to be the convolution of the ideal projection by the shift-variant blur functions. So the measured projection will be represented as a weighted summation of neighboring ideal projections of different angles and positions. From this standpoint, the blur function can be defined as a function of a common spatial frequency in the Fourier domain. Hence the deconvolution of the measured projection was circularly processed in the frequency domain. The effectiveness of the method was proven by simulations involving various aperture angles and the distances between the collimator surface and the rotational center.

A phantom has been designed which simulates the environment of the left ventricle in the chest. The phantom consists of 5 tubes approximating arteries with contrast, layers of acrylic plastic to simulate soft tissue, bones, and a bladder which can be inflated with a dilute contrast solution to simulate the left ventricle. The parts of the phantom can be adjusted to provide known amounts of motion in three dimensions and is used in the evaluation of a three-dimensional motion correction algorithm which is described in detail in a companion paper. Images of the phantom appear to be like those seen in x-ray imaging of the chest. The results using the phantom to test the registration algorithms show that a significant improvement in the subtraction images of the phantom can be achieved in the presence of simulated three-dimensional motion, when compared to conventional subtraction.

An experimental design has been devised and tested for evaluating computerized contrast enhancement techniques designed for the task of improving lung nodule detection accuracy in digital chest radiographs. Under this design, a global contrast enhancement technique has been tested which provided images resulting in improved detection of simulated chest nodules. The work was conducted using images from a Toshiba Computed Radiography System (TCR-201). The images used were made using a 3M chest phantom with simulated nodules of four diameters and two thicknesses. Each image that contained a nodule only had one, and it was in one of only four possible locations. A phantom with location and background information provided to the observer was used to eliminate confounding factors affecting the evaluation of detectability. A relatively simple histogram modification algorithm processed the data, and the resulting images were written to film by the TCR system. A psychophysical study was conducted to compare detection ability with the enhanced images versus standard TCR processed images. In a film-to-film comparison of 48 cases (32 with nodules), when viewing the enhanced images, two out of three observers achieved a distinct improvement in true/positive responses over the ordinary TCR images with no increase in false/positive responses.

Reformatted CT images of the mandible and maxilla are described as a planning aid to the surgical implantation of dental fixtures. Precisely scaled and cross referenced axial, oblique, CT generated panorex, and 3-D images are generated to help indicate where and how critical anatomic structures are positioned. This information guides the oral surgeon to those sites where dental implants have optimal osteotic support and least risk to sensitive neural tissue. Oblique images are generated at 1-2 mm increments along the arch of the mandible (or maxilla). Each oblique is oriented perpendicular to the local arch curvature. The adjoining five CT generated panorex views match the patient's mandibular (or maxilla) arch, with each of the views separated by twice the distance between axial CT slices. All views are mutually cross-referenced to show fine detail of the underlying mandibular (or maxilla) structure. Several exams are illustrated and benefit to subsequent surgery is assessed.

In circular tomosynthesis, object detail at a distance from the plane of interest is blurred according to the zero order Bessel function, whose main lobe defines slice thickness, while its ringing side lobes permit further outlying structures to "leak" through. Using the orthogonality of the Bessel functions, a set of sampling cone opening angles was specified such that the corresponding blurring functions formed an orthogonal basis over a finite distance from the reference plane. This basis set was used to synthesize by a Fourier- Bessel series a window function defining slice thickness with superior side lobe suppression. The window function that concentrates the most "energy" within a finite interval for a fixed number of sampling cones is the circular prolate function. Its application permitted tomosynthesis of 3 mm thick slices at a spatial resolution of 0.5 mm by using only 3 sampling cones with opening angles 1.9, 4.4, and 6.9°, yielding a suppression of the first side lobe of -38 dB, as compared to -8 dB achieved with the Bessel function.

A modular real-time video image processor has been developed which is able to support medical fluoroscopic procedures. These procedures require human interpretation of video images in real-time. Two modules are described: a two-dimensional locally adaptive contrast enhancement unit, and a gaussian temporal averager. Other available modules are listed, but not discussed.

Scatter radiation degrades image contrast as well as quantitative relationships in transmission x-radiography, especially with broad area detectors. Use of an anti-scatter grid and/or air gap eliminates much of the detected scatter radiation, but at the expense of attenuated primary radiation and geometric unsharpness. An alternate method is investigated that can more closely approximate the desired "primary" image either in conjunction with or in absence of the abovementioned techniques. The characterization and parameterization of a scatter point spread function (PSF) for a given imaging geometry (object thickness, field size, focus-object-detector distances) and radiographic technique (photon energy, grid/no grid) allows the removal of the scattered components by deconvolution using inverse filter post-processing methods. Assumptions of a stationary and spatially invariant PSF are made to enable the use of an efficient two-dimensional Fourier transform inverse filtering scheme. In spite of the inherent non-linear attributes of the scattering and image detection processes, a first order linear approximation using a Gaussian form to model the scatter PSF provides a numerically invertable filter kernel that removes scatter and improves image contrast as well as quantitative accuracy.

Quantitative analysis of regional heart wall motion provides a useful method for assessing the severity and the extent of ischemia and infarction in patients with coronary artery diseases. Manual tracing of heart borders for such analysis is time consuming and with high inter- and intra-observer variabilty. To overcome these limitations, an automated technique to extract the heart borders is required. Such a technique based on the Laplacian of Gaussian operator (∇2G) has been implemented. Edge points are located by detecting the zero-crossings after the image is convolved with the ∇2G filter. Contiguous edge points are then chained together to form edge lines. Heuristics based on the morphology of the heart cavity are used to select the edge lines corresponding to the endocardium border.

We present a useful clinical method for relating plane X-ray film coordinates to a head-fixed reference system. It is commercialized as an add-on to one of the most widespread systems for stereotactic neurosurgery. It can be used for stereotactic irradiation of cerebral blood vessel malformations, as well as for planning avascular electrode paths in stereotactic biopsies or conventional neurosurgery. There are no constraints on the relative position of the X-ray tube, patient and film holder. Extensive simulation tests demonstrate the reliability and accuracy for clinical practice.

A system is being developed to test the possibility of doing peripheral, digital subtraction angiography (DSA) with a single contrast injection using a moving gantry system. Given repositioning errors that occur between the mask and contrast-containing images, factors affecting the success of subtractions following image registration have been investigated theoretically and experimentally. For a 1 mm gantry displacement, parallax and geometric image distortion (pin-cushion) both give subtraction errors following registration that are approximately 25% of the error resulting from no registration. Image processing techniques improve the subtractions. The geometric distortion effect is reduced using a piece-wise, 8 parameter unwarping method. Plots of image similarity measures versus pixel shift are well behaved and well fit by a parabola, leading to the development of an iterative, automatic registration algorithm that uses parabolic prediction of the new minimum. The registration algorithm converges quickly (less than 1 second on a MicroVAX) and is relatively immune to the region of interest (ROI) selected.

In order to support screening of chest radiographs on mass survey, a computer-aided diagnostic system that automatically detects abnormality of candidate images using a digital image analysis technique has been developed. Extracting boundary lines of lung fields and examining their shapes allowed various kind of abnormalities to be detected. Correction and expansion were facilitated by describing the system control, image analysis control and judgement of abnormality in the rule type programing language. In the experiments using typical samples of student's radiograms, good results were obtained for the detection of abnormal shape of lung field, cardiac hypertrophy and scoliosis. As for the detection of diaphragmatic abnormality, relatively good results were obtained but further improvements will be necessary.

The use of the discrete cosine transform (DCT) in imaging applications is still not as extensive as it's properties would imply, due to a comparative lack of available hardware for fast computation. Commercially available image processors typically implement the Fourier transform, and most do not have the word size necessary to handle the wide dynamic range required for radiological images with up to 2K x 2K size and 8 or more bits resolution. Efficient and easy to implement algorithms exist for the computation of the DCT. The advent of programmable single chip digital signal processors (DSP's) with fast multiplier-accumulator (MAC) units and large word sizes (24 to 32 bits) allow for hardware solutions that are economical, fast, and flexible. A special module utilizing multiple DSP's for performing the fast cosine transform (FCT) for image compression has been built. The flexibility afforded by DSP programming allows for handling of input data with different formats, dynamical scaling, and any necessary pre- or post-processing operations.

A hardware implementation of the full frame image compression method developed at UCLA has been completed and is undergoing evaluation. Variable compression ratios can be achieved according to user input parameters. Current clinical acceptance studies focus on a compression ratio of 10:1. The compression and reconstruction process for 1K x 1K images each take approximately 8.5 sec. Future upgrades can enhance the performance to 1.1 sec for 1K images and 4.3 sec for 2K images. Peripheral functions incorporated into the compression module include 1K image display and parallel data link to a VAX base PACS system.

This paper describes the evaluation of a new image compression technique as a first example of a successful procedure to evaluate image processing techniques. Forty-eight CT images of the abdomen were compressed and then reconstructed by a method called sub-band coding using vector quantization. The compression ratios used were 16:1 and 19:1. Five radiologists participated in this study: two to select the images and help in determining the appropriate training set for development of the code-book for the algorithm, and three to participate in the actual experiment. Data were taken to determine diagnostic certainty, ratings of the visibility of diagnostically important structures in the image, and accuracy of identifying which images were the originals and which were not. Results from this study demonstrate: (1) The areas under the ROC curve were virtually identical for all three sets of images (AUC's of 0.78, 0.77, 0.78 for original, 16:1, and 19:1 respectively) and these were not statistically different; (2) all diagnostically relevant structures in the compressed images were maintained by the compression technique; and (3) half of the 16:1 compressed images were identified as original images and 3 out of 48 of the 19:1. The authors conclude that the compression technique shows promise for use on CT images.

Image Transforms, such as the S-Transform, provide a hierarchical representation of the image which is attractive as part of a data compression technique in a PACS environment. In this paper, the S-Transform is shown to be a special case of subband coding, thus characterizing the spectral behavior of the transform. The S-Transform decomposes the frequency plane into roughly octave spaced regions. The corresponding "octave" images in the spatial domain can be linearly combined with different weights in order to synthesize an enhanced image. This is a computationally efficient process which provides a great deal of flexibility in the specification of the enhancement MTF characteristic. Examples are shown of enhanced computed radiographs, comparing the results of our technique and unsharp masking.

This paper describes an efficient compression method for medical images. This method can compress images with small bit rates and generate reconstructed images of high quality. Discrete Cosine Transform (DCT) and Block Truncation Coding (BTC) are widely used in various fields. These methods, however, have the following problems. With the DCT method, the reconstructed image quality is quite good, except that the quality of sharp edges in the image is clearly inferior. The BTC method can reconstruct sharp edges, but the reconstruced image is not suitable for medical images. To solve these problems, we have proposed a new hybrid compression method. In this method an original image is divided into sub-images. The high frequency component of each sub-image is evaluated. If the component is small, DCT is performed on the sub-image. When the high frequency component is large, BTC is applied to the sub-image itself, then DCT is performed on the difference between the original and the reconstructed sub-images. In our experiments, this method proved to be effective.

The ACR-NEMA Digital Imaging and Communications Committee established a working group to propose means to incorporate a wide variety of current compression techniques in the framework of the ACR-NEMA Standard for optional use. A draft Standard on compression has been circulated for comment. It proposes methods of allowing the use of public domain techniques, both bit preserving and non-bit preserving, predetermined techniques between devices already aware of the selected algorithm, and specification by the transmitting device of both algorithm and parameters prior to transmitting compression data. This paper will report on details of the proposal and the current status. An example for an allowed compression algorithm, is presented.

The application of direct volume visualization techniques to the presentation of CT data is explored. No surface detection or fitting of geometric primitives is involved. Images are formed by directly shading each data sample and projecting it onto the picture plane. The visualizations in this study are based on a hybrid physical model incorporating aspects of both surfaces and semi-transparent gels. Using a surface model, shading calculations are performed at every voxel with local gradient vectors serving as surface normals. In a separate step, surface classification and enhancement operators are applied to obtain a partial opacity for every voxel. Independence of shading and classification calculations insures an undistorted presentation of 3-D shape. The use of non-binary classification operators insure that small or poorly defined features are not lost. The resulting colors and opacities are merged from back to front along view rays using volumetric compositing, an approximation to the visibility calculations required to render a semi-transparent gel. The technique is simple and fast, yet produces images exhibiting smooth surface silhouettes and few other aliasing artifacts. The use of selective blurring and super-sampling to further improve image quality is also described.

The 3D comprehension of anatomic and computed objects in medical images, presented using shaded display, can be increased by attention to light sources, surface textures, and transparency. Increased 3D cues are provided by real-time interactive modification of viewpoint, object selection, transparency, and clipping planes. Methods for achieving these increases in comprehension will be presented. In particular, the following will be described, with applications from diagnostic CT and MRI and radiotherapy treatment planning: 1) A set of workstation-based tools, using heuristic and interactive approaches, for defining object contours, connecting them into objects, and tiling their surfaces. 2) A renderer that provides more rapid computation of many different presentations of the same view of a scene by keeping a large intermediate file of geometric information about a particular view. Choices of objects to be displayed, and for each object, its color, specularity, and transparency, can be deferred until, and changed after, all the geometric computations. 3) The usefulness of the Tektronix stereo polarizing plate and the kinetic depth effect for adding to the 3D comprehension of objects. 4) The custom-built graphics engine, Pixel-planes, and its use in providing, all in near real time, object selection, variation in object transparency levels, variation in viewpoint, and specification of clipping planes. Also described will be the ability of Pixel-planes to present oblique grey-scale image slices superimposed on the clipping planes as the clipping plane is interactively moved.

3D-display of medical objects derived from cross-sectional images has demonstrated its clinical usefulness in various applications such as surgery planning and recently also in diagnostic radiology. Instead of viewing sequences of images the region of interest can be looked at in its 3D-shape or at least as a cross-sectional image within the anatomical surroundings (fig, 1,2). This new way of viewing is certainly more natural and understandable than the conventional one. If we imagine that the invention of X-rays, CT or MR had not been made yet, would we not aim at an imaging modality which would deliver images as known from anatomy? So having the new 3D-imaging facilities the physicians might act in the future Hore like anatomists (with radiological eyes), A 3D-image, however, if generated from a single parameter (such as a Hounsfield value in CT) does not nearly have the information content of the anatomical reality. In addition, the capability of performing a dissection at the computer screen requires, that the program behind the screen is able to perform the corresponding anatomic segmentation, This means that the data structure on which the 'anatomist program' is working must contain more detailed information on the organs to be displayed,

A 3-D display system which is based on the principle of active stereoparallax, has been developed. It consists of an IRIS-2400 graphics workstation, a Stereographics stereoscopic imaging system employing electro-optical shutters in glasses, and two Polhemus 3-D cursors, one in style form and one in block form. The usefulness of this system in general, and for medical imaging in particular, is discussed. The principle of active stereoparallax implies both stereoscopy and movement parallax. It gives the observer a reliable perception of space and allows him to make systematic distance estimates. Movement parallax is caused by coupling the observer's movement to the resulting shifts of the objects in the image. In our system, this is realized by recording the observer's head movement with the 3-D cursor in block form. It is shown that the interaction between the observer and the environment is important for space perception. The system has been developed in close cooperation with the Faculty of Industrial Design Engineering of the Delft University of Technology.

A broad area of applications for processing and analysis of medical images can be found with images that are not computer generated, such as radiographic or cytological images. Success when processing and analysing these images depends very much on the quality of the image acquisition system. An image acquisition system based on a CCD linear image sensor was built for the acquisition of chest radiographs. CCDs provide a wide dynamic range and very high spatial resolution, and are well suited for high quality image acquisition systems. These abilities are however reduced considering the variations between the output characteristics of different sensing elements. The digitizer that was built has specially designed hardware for real-time correction of these variations. The digitizer is based on a microcomputer that manages the overall system operations (e.g., calibration, image acquisition, scanning control, light control, fault detection, etc...) and programs the real-time correction hardware. The combination of the CCD image sensor and the correction hardware proved to be quite powerful and effective, taking into account the measurements that were made. The paper discusses the use of CCD image sensors and their characteristics, and describes in detail the algorithms and hardware that were used. It finally points out prospective applications for these type of digitizers, namely in the digitization of radiographic images in a PACS environment.

The User Interface Study Group at the University of Arizona is investigating the interaction of Radiologists with digital workstations. Using the Arizona Viewing Console we have conducted an experiment to compare a digital workstation with a particular conventional reading process used for cases from a local Health Maintenance Organization. A model consisting of three distinct phases of activity was developed to describe conventional reading process. From this model software was developed for the Arizona Viewing Console to approximate the process. Radiologists were then video taped reading similar sets of cases at each workstation and the tapes were analyzed for frequency of hand movements and time required for each phase of the process. This study provides a comparison between conventional reading and a digital workstation. This paper describes the reading process, the model and its approximation on the digital workstation, as well as the analysis of the video tapes.

Digital laser film printers are now being used as one means of obtaining hardcopy output of digitally acquired radiographic images. The noise characteristics of these devices are therefore of interest as part of the imaging chain in diagnostic radiology. We report measurements of the noise power spectrum (NPS) of nominally uniform fields written by a digital laser film printer. These measurements were made using conventional long-slit one-dimensional auto-spectral methods, as well as with cross-spectral methods applied to spatially registered images. The resulting spectra are significantly more complex than those generally estimated for uniformly exposed film. This complexity is manifested in the combination of narrow line and continuous spectral components, and in the highly anisotropic nature of the spectra. Simple models, such as the addition of one-dimensional periodic patterns and isotropic noise, do not adequately describe our data. We describe and illustrate by simulation four kinds of noise which might occur in digitally printed images. These can be distinguished in practice by the observed dependence of the NPS on slit and block length. Behavior consistent with all four kinds of noise is observed in our experimental data.

This paper describes the psychophysical evaluation of a prototype electronic viewing console developed by the Toshiba Corporation. The evaluation consisted of two phases. In the first phase, four radiologists viewed 10 pediatric images. Each image was viewed first with no enhancement, second with window and level, and third with pan and zoom. A diagnosis and certainty was recorded for each of these viewings. The areas under the curve (AUC) were 0.80, 0.86, and 0.91 for no enhancement, window and level, and pan and zoom, respectively, compared to an AUC of 0.90 for film. The second phase was a clinical comparison of the capabilities of this electronic system to those of conventional film and computed radiography using cholangiography studies of the biliary system. Thirty-six images were viewed by five radiologists, and free use of enhancement functions was allowed. The areas under the curve for the images seen on the viewing console, conventional film, and computed radiography film images were 0.84, 0.84, and 0.83, respectively. The authors conclude that for examinations of these anatomic regions, pediatric chest and cholangiograms, each of them depicting a variety of pathologic processes, electronic viewing consoles seem to convey the needed information for accurate diagnosis.

A high resolution, operator interactive, medical viewing and analysis system has been developed by Toshiba and Bio-Imaging Research. This system provides many advanced features including high resolution displays, a very large image memory and advanced image processing capability. In particular, the system provides CRT frame buffers capable of update in one frame period, an array processor capable of image processing at operator interactive speeds, and a memory system capable of updating multiple frame buffers at frame rates whilst supporting multiple array processors. The display system provides 1024 x 1536 display resolution at 40Hz frame and 80Hz field rates. In particular, the ability to provide whole or partial update of the screen at the scanning rate is a key feature. This allows multiple viewports or windows in the display buffer with both fixed and cine capability. To support image processing features such as windowing, pan, zoom, minification, filtering, ROI analysis, multiplanar and 3D reconstruction, a high performance CPU is integrated into the system. This CPU is an array processor capable of up to 400 million instructions per second. To support the multiple viewer and array processors' instantaneous high memory bandwidth requirement, an ultra fast memory system is used. This memory system has a bandwidth capability of 400MB/sec and a total capacity of 256MB. This bandwidth is more than adequate to support several high resolution CRT's and also the fast processing unit. This fully integrated approach allows effective real time image processing. The integrated design of viewing system, memory system and array processor are key to the imaging system. It is the intention to describe the architecture of the image system in this paper.

Future radiology departments with DIN/PACS will likely contain many different workstations developed by independent image processing vendors. However, radiologists and other users of the workstations should not be forced to learn more than one interface to DIN/PACS: the same image processing, data base management and networking commands should be usable from any workstation or network node on the system. In addition, programmers of this consistent user interface should not have to concern themselves with the intricacies of each workstation on the system. Rather, they should be able to select functions from a workstation library which manipulates the image display in (as much as possible) a device-independent manner. Previously we have introducted a Multi-process Virtual Image Processor (MUVIP) software architecture for maximizing image processing applications code portability [I]. Applications programs generate device-independent image processing commands, or "metacode," which is passed to a separate process called an Autonomous Virtual Image Processor (AVIP). Thus, in order to transport an applications program to a new image processor, it is necessary only to create a new AVIP corresponding to that processor. In this sense each applications program is "totally portable," since transferring it to a new device does not require either recompilation or relinking of the applications program itself. In this paper, we consider implementations of the MUVIP architecture on two hosts, and its application to three image processors. A test function set was created, and a consistent user interface for those functions was developed for each combination of host and image processor. The image processor used by each program was selectable at run time, and in the case of one program, two image processors could be used together in a cooperative mode, providing more capability than either could achieve alone. The raw interprocess communication overhead of MUVIP was measured, and its impact on the test functions' performance was examined and found to be negligible. Thus, it was demonstrated that the MUVIP architecture is a viable alternative for enhancing the portability of image processing applications programs.

Imaging workstations for radiology would be used by radiologists for a number of hours each day. Such long use demands a good ergonomic design of the workstation in order to avoid user fatigue and frustration. The film-and-viewbox methods presently in use have evolved over the years since radiography became a diagnostic tool. The result of this evolution is that, despite the problems of film, the ergonomics of film reading and reporting is quite mature. This paper will use a somewhat lighthearted look at workstation design using the ideas of well-known architects and designers to illustrate points which should be considered when implementing electronic viewing systems. Examples will also be drawn from non-radiologic environments in which the user is presented with visual information for his or her integration.

In the Department of Radiology at the University of Arizona investigation of the interaction of Radiologists with digital workstations is conducted by the User Interface Study Group. Using the Arizona Viewing Console we have conducted an experiment to compare one digital workstation with reading done at a film alternator (also known as a rotator). A model consisting of distinct phases of activity was developed to describe alternator reading process. From this model software was developed for the Arizona Viewing Console to approximate the process. Radiologists were then video taped reading similar sets of cases at each workstation and the tapes were analyzed for frequency of hand movements and time required for each phase of the process. The information gathered provides a comparison between alternator reading and reading at a digital workstation. This paper describes the reading process, the model and its approximation on the digital workstation, as well as the analysis of the video tapes.

A prototype workstation with the capabilities of both full resolution presentation of four 512 x 512 CT images and reduced presentation of a larger number of images (e.g. two full abdomen CT studies) has been developed. This prototype has been designed to develop an evaluation protocol for the comparison of two workstations (manual and/or electronic) to each other, and to employ this methodology to test a specific navigation strategy for single screen displays showing a large number of images.

A PACS architecture was simulated to quantify its performance. The model consisted of reading stations, acquisition nodes, communication links, a database management system, and a storage system consisting of magnetic and optical disks. Two levels of storage were simulated, a high-speed magnetic disk system for short term storage, and optical disk jukeboxes for long term storage. The communications link was a single bus via which image data were requested and delivered. Real input data to the simulation model were obtained from surveys of radiology procedures (Bowman Gray School of Medicine). From these the following inputs were calculated: - the size of short term storage necessary - the amount of long term storage required - the frequency of access of each store, and - the distribution of the number of films requested per diagnosis. The performance measures obtained were - the mean retrieval time for an image, - mean queue lengths, and - the utilization of each device. Parametric analysis was done for - the bus speed, - the packet size for the communications link, - the record size on the magnetic disk, - compression ratio, - influx of new images, - DBMS time, and - diagnosis think times. Plots give the optimum values for those values of input speed and device performance which are sufficient to achieve subsecond image retrieval times

Using the Performance Analysis Workstation (PAW), a UNIX based modeling and performance analysis tool, the use of film in the Duke University Medical Center (DUMC) Radiology Department has been simulated. The Radiology Department Traffic Model for DUMC simulates the flow of film through the department, and then incorporates the effect of introducing a PACS-type system into present operations. Each Radiology Section is considered separately for queueing of two types of film: old film (from previous exams) and new film (from the present exam). The amount of film in each queue at any time is controlled by controlling hours of operation, service times, delay, and arrival rates. The model also takes into account the use of film in each major radiology area. This gives some idea of the load on a display device in that area as well as the amount of storage needed to adequately handle its daily load if local storage at the display device is desired.

Radiological PACS image sizes and desired retrieval response times demand high-bandwidth communication networks. Local area network technology at speeds higher that 10 Megabits/second (IEEE 802.3) have not achieved standardization nor production volume. Our current PACS experiments are based on a three-level subnet approach using 10 Mb/s Ethernet channels. An Ethernet channel is shown to support image transfers at an average throughput of 3 Mb/s. Preliminary measurements and simulation results suggest that traffic from as many as two-to-three archives can be supported on the same channel.

Diagnostic cycle time is defined as the time between the initiation of a diagnostic test and the start of specific therapy indicated by the findings of the test. Timely delivery of treatment may not only improve the outcome of this therapy but also reduce the cost of treatment through efficient resource utilization. The increasingly limited availability of health care dollars and competition between health care providers, both encourage investigation into methods to reduce the diagnostic cycle time. The transfer of medical information over established communication networks may be improved using state of the art electronic technology. A model was developed to investigate the effects of PACS technology on the diagnostic cycle time.

The purpose of this study was to develop and design a model computerized patient history form that could be used by practicing radiologists during the x-ray reading and interpretation process. We conducted a multi-observer study to determine 1) whether such a form is warranted and can be a feasible part of the reading process; and 2) what the component parts of the form and their order of appearance in the form should be. Twenty board-certified radiologists were tested in their areas of subspecialty; each radiologist read a series of 10 cases, half with and half without clinical histories. Results indicated that all those tested would use such a history if it were provided to them along with the radiographs to be read; that they prefer such a history to be brief but detailed enough to provide more information than the typical requisition does; that they prefer a rapidly accessible history; and that there are preferences about what information appears and the order in which it appears.

The management of the vast amounts of medical images and information generated by today's clinical services is a growing problem. The solution to the problem will increasingly require the use of advanced technologies data storage, image display, communication, and human engineering. The progress of individual technologies has been rapid; however, system integration and user acceptance of digital image management technology have been slow . Critical evaluation of the efficacy of this use of technology is essential for the evolution of picture archiving and communication system (PACS) in radiology.

To successfully replace the alternator as the preferred diagnostic tool, the technology incorporated into a PACS diagnostic workstation must duplicate, if not surpass, the most important or fundamental features of the alternator. Alternatively, incorporating the alternator into a more global view of the operations of the radiology environment as a whole, the workstation may provide capabilities not available in a film-based system, such as sophisticated image processing techniques, communications capabilities, and data management services. This paper reports on systems analysis studies carried out to determine and characterize the fundamental alternator features. A model is discussed and quantized. In broad terms, features included in the model are the amount of information displayable at one time, the time to display another equal amount of information, and the ease in which the change between sets of information is made. An equivalent functional model of an electronic alternator is defined. Feature characteristics are mapped into available technology. This paper will also explore several examples of the additional functionality technically possible in PACS diagnostic workstations.

The past decade has witnessed advances in computer architectures that have provided more speed and performance with memory than ever before. Real time processing for medical applications is approaching reality with these new hardware implementations. It becomes essential to investigate these architectures in light of present and possible future Picture Archiving and Communication Systems (PACS) requirements. We have chosen four possible architectures that represent the breadth of the present multi-processor systems. Actual and simulated timing results for performing a 2-Dimensional Fast Fourier Transform are presented. We then discuss how to evaluate candidate architectures.

Many Radiology Departments are contemplating a move from manual operations to Picture Archiving and Communications System (PACS) based operations. A question remains regarding the true cost-benefit of serious investments in PACS based operations. A methodology has been established by the Research Triangle Institute (RTI) in cooperation with the University of North Carolina (UNC) at Chapel Hill which addresses aspects of the cost-benefit analysis of manual versus digital operations within a Radiology Department. The methodology has evolved as a collection of modeling tools and techniques. Early tools included spreadsheet models of existing activities within the UNC Radiology Department. Second generation tools now include deterministic network flow modeling and discrete event queuing network modeling packages. Complementary queuing models have been developed to separate the modeling of Department workload size and mix from the model of the man/machine system that services the workload. A range of performance measures have been incorporated into the second generation models. Among these measures are workload throughput, man/machine resource utilization, system and subsystem responses times, and resource availability measures.

The management of radiology images and information has always been a difficult task. The difficulty has been growing as the number of imaging modalities within radiology becomes larger. It is generally agreed that the use of advanced computer technology may help the situation (1). Many innovative technologies can now be integrated to address the problem of managing radiology images and information.

A satellite image transmission system has been installed between an orthopedic clinic in suburban Philadelphia and the Hospital of the University of Pennsylvania in central Philadelphia. The two facilities are approximately 15 miles apart. Images are acquired with a laser film digitizer and transmitted over a Ku-Band satellite link. After receipt, the images are archived and sent to a Virtual Imaging VIEW-2000 workstation for display and interpretation. This paper addresses the software on the VIEW-2000, from a designer's point of view. The workstation software handles image retreival, image management and image display. User interaction is provided via a standard alphanumeric keyboard and a three-button mouse. Image manipulation includes rove, zoom, window/level and inverse intensity. Image sequencing logic facilitates radiologist interaction. Various enhancements are considered.

A PACS module for Pediatric Radiology began routine clinical operation in March 1987. The system thus far has been used to conduct daily X-ray rounds as well as for fast patient reviews during all hours of the day. This paper describes the clinical operation of this system including patient registration, image acquisition, image management, and patient case reviews. A PACS system operator is responsible for the daily mainenance of the system. These responsibilities are outlined.

The details of a clinical PACS module for Pediatric radiology are discussed. The system operates as a large queueing network allowing several processes to execute concurrently in a prioritized and coordinated fashion. This paper presents clinical data accumulated since March 1987 on system workload, system performance, and workstation optimization techniques.

As a critical care unit that functions under strict time constraints, the coronary care unit (CCU) requires rapid access to radiological information of the patient. In order to meet this clinical demand, a digital remote viewing system was developed and has been in clinical operation since March 1987. The system delivers the softcopy of chest images to the intensive care unit. During the nine months of continuous clinical operation, the system was evaluated by analyzing the utilization and performance statistics.

The ACR-NEMA standard implies a conceptual schema (the "NEMA schema") for a medical PACS system. We describe a prototype implementation of the NEMA schema using a relational database management system together with a translation facility which provides ACR-NEMA dataset-oriented access to the database.

This paper discusses evolutionary developments for an architecturally integrated and unified Picture Archive and Communication Systems (PACS). It describes an ACR-NEMA implementation, expectations for ACR-NEM modality integration and a vision of the 1990's and beyond.

We have designed and installed ACR-NEMA communications capability on an IBM PC AT personal computer. Our implementation consists of an IBM PC AT bus compatible interface board with a custom device driver running under the MS DOS operating system. The interface board, which occupies one 16 bit bus slot, is capable of sending or receiving single data frames at rates in excess of the 8 MBytes/sec target specified in the standard. A full implementation (including multiple virtual channels) of all communication protocol layers from the physical layer through the session layer are operational in the driver. Sustained transmission rates between two AT computers equipped with this interface have been measured to be 750 KBytes per second for buffer-to-buffer transmission in "data acknowledge" service class. We have identified several difficulties inherent in the ACR/NEMA standard. These and the question of whether the overall philosophy of the standard optimally meets the needs of a real radiology department are discussed.

The optical memory card (OC) is a plastic coated card of standard credit card size with a reflective area on which 2 to 200 megabytes of digital information can be recorded and read by lasers. The OC has the potential for serving as an archiving medium in a digital imaging network and could enhance continuity of care to the wounded soldier in the battlefield. This paper addresses the applicability of the OC to the U.S. Army combat medical care system, identifies two projects that will be investigating its use, describes the OC technology, and outlines its past promises, present status and barriers to be overcome.

As important as the clinical utility of digital radiography is to its acceptance by medical professionals, the performance of the PACS that supports such modalities, as experienced at the diagnostic reporting station, is at least as important. That performance is determined by the network configuration and memory node distribution of the PACS which must be based on the clinical requirements for image communication. To this end accurate film image flow data were accumulated in order to identify and quantify the clinical endpoint flows and volumes that would have to be supported by PACS. Based on those data, a statistical model was developed, and is presented, of the film image flow in a 12-room diagnostic imaging department. Actual image flow data support the authors' contention that, despite current technological limitations, an effective PACS configuration can be developed to provide the same level of image communication service currently provided by the film-based system.

A digital optical disk archive has been running continuously in the Pediatric Radiology Section at UCLA for over one year. During this year both the computerized patient registration system and the optical disk image archive have been accumulating patient and image data. Statistics derived from one year's operation are combined with a pediatric film library survey to summarize system use and to project departmental archive requirements.

Data base archive system (DBAS) for picture archiving and communication systems (PACS) require very large storage systems. The storage systems must be high speed and must be very inteligent in order to handle requests for image storage and retrieval. The DBAS is interfaced to a high speed fiber optic network via ACR-NEMA interfaces. User workstations and imaging equipment represent the sinks and sources of digital image data. The image data to be stored in the DBAS may be as large as 2048 x 2048 x 12 bits, or 50 megabits per image. Image storage and retrieval requests contain a mix of image and text data. The text data represents patient information and doctor's diagnosis. User scenarios include simultaneous image generation and retrieval. Hence, the image data transfer and storage must occur at a high speed in order to achieve response times of less than two seconds per image request. The approaches used to implement a DBAS must contain parallel and intelligent processing to handle the large amount of image data in a real-time mode. In this paper characteristics, elements of image data base system, and the functional description of DBAS architecture components and intelligence are described. The use of expert systems to process the image requests and to locate the information is explored.

Data base archive systems (DBAS) for picture archiving and communications systems (PACS) require very large storage devices for the digital images. The images, as large as 50 Mega-bits, are stored and retrieved simultaneously at the DBAS. The response time to the user for image storage and retrieval must be of the order of two seconds. Consequently, the image storage at the DBAS must be efficient to process the image requests in the required time. This paper describes an image migration algorithm for efficiently storing and moving the images within the DBAS storage devices. The DBAS storage devices are structured in a three level hierarchical architecture. Level 1 is composed of high speed magnetic disks and contains recent images, up to seven days. Level 2 consists of automatic juke box optical disk systems and contains images up to 30 days old. Level 3 is a manual juke box optical disk system and contains storage for images up to 7-10 years. A fourth level is a manual archive for images greater than seven years. The DBAS migration algorithm was developed to migrate images within this three level storage structure. A migration model was developed to represent the image data flow between levels. The model can be used to determine the size of each level at any time and the amount of data flow between levels. The model is driven by image generation requests by the Imaging Equipment and image retrieval requests by the user workstations. The migration algorithm for the DBAS was included in the simulation of the entire PACS system, including the Image Network, DBAS, Imaging Equipment, and User Workstations. This paper presents the result of the simulation and shows the effect of the image migration algorithm on the channel utilization.

The initial aim of videodisc development was to produce a system that records audio/video information on a disc, replicates this disc accurately and inexpensively on plastic and finally, plays the replicas on home television screens by means of a disc player attachment. What made the videodisc so interesting is the potential combination of three properties: very low cost in high-volume duplication, very high information density, rapid acces to any portion of a long recording. Presently, two major disc categories are wel1 established: mass replica and write-one discs. Single headed players are almost exclusively used today. But, in some applications, where mainly access and interactivity are very important, two headed players are either used or presently being considered. Some new problems are surfacing, which are not of that much concern for the single-headed players. One of new problems is a time-based error (TBE) caused by recorded asymmetry, which will by discussed in this paper. Al so, the influence of this phenomenon on the medical images stored on the videodisc will be examined.

In the Netherlands a national PACS development program has been started, supported by-the Dutch Society of Radiology, and funded by the Dutch Department of Health, because of the national character of the project. Three main partners are cooperating in this development: the Utrecht University Hospital (AZU), BAZIS and Philips International (Product Division Medical Systems), with the Delft University of Technology as the main subcontractor of BAZIS. The non-profit foundation BAZIS, developing and supporting the 'ZIS' Hospital Information System (in use in some 30 Dutch hospitals now, together over 15,000 acute beds), initiated its current IMAGIS (IMAGe Information System) projects in 1984. The Dutch PACS project, in which BAZIS is participating, started in 1986. The final goal of IMAGIS is to achieve a PACS which is fully integrated with already existing hospital information systems (HIS), with support for scientific research. Both the development and operation of such a HIS-PACS include many scientific research aspects. The current efforts of BAZIS are concentrated on three main issues: diagnostic image quality evaluation (e.g. effects of data compression); modelling, software simulation and technology assessment of a prototype PACS (both general and detailed aspects); coupling and integration of PACS and HIS (e.g. the BAZIS 'ZIS'). In this paper we will refer to the following intermediate results: the psychophysical software package for Feature Evaluation And System Inspection By Logged Experiments (FEASIBLE); the modelling and simulation software package for Medical Image Representation, Archiving and Communication, Learned by Extensive Simulation (MIRACLES); first results of the coupling experiments. Finally we will outline the future direction of the IMAGIS projects.

A first step has been made towards one of the ultimate goals of the so-called IMAGIS projects: the realization of a working PACS integrated with a Hospital Information System. This HIS-PACS coupling is part Of the Dutch Pacs project, a national PACS development program, supported by the Dutch Society of Radiology and funded by the Dutch government. This development is a cooperative effort of BAZIS (the Dutch Development and Support Group of the Hospital Information System 'ZIS'), the Utrecht University Hospital (AZU) and Philips International (Product Division Medical Systems) since 1986. In order to have a prototype HIS-PACS combination available as soon as possible, the three partners mentioned agreed on using a stepwise approach for carrying out the coupling project. For the first phase of the project we have restricted ourselves to: 1. the coupling of a specific HIS (BAZIS/ZIS) to a specific PACS (Philips/MARCOM) 2. only the in-patients of a particular ward of Internal Medicine 3. only one-way dataflow: from HIS to PACS, using a relatively slow link. This paper describes design criteria, technical solution chosen, message format and experimental set-up.

The advantages to be expected of full-scale PACS implementation are widely described in literature. In the decision to introduce such systems costs however will also play an important part. The benefits to be achieved should outbalance the costs. In this paper the development of a software package for cost assessment is described. The configuration requirements are determined starting from the workload of the radiology department and a PACS model using results of simulation studies. Of course additional user requirements are important input parameters (e.g. number of workstations, screen resolution etc.). Data on costs of the various configuration components and the expected future trends in these costs are used to estimate the costs of the specific PACS as a function of time. Savings to be achieved in the various categories of resources are input (again with their trends) to lead to an overview of the total savings to be expected. The package results in a graph of net annual costs as a function of the moment of PACS introduction. In the package a critique module is foreseen that checks whether the data fed into the system are to a reasonable level in agreement with expert opinions.

The storage of large amounts of digital image data and the communication of the images within a hospital are major problems in the development of a Picture Archiving and Communications System (PACS). Image compression reduces both the problems of storage capacity and of the image transmission rate by approximately a factor 12. It is shown that the use of memory buffers at different hierarchical system levels provides for an additional reduction in the required transmission rate. In order to determine the communication rates and number and size of the buffers, the PACS has been modelled by a system of queues and servers. Queueing models can be analysed by means of computer simulation but also mathematically. Examples of mathematical analysis are given and a simplified model of the department of radiology is evaluated.

The Dutch PACS project is a cooperation of the Utrecht University Hospital, BAZIS Hospital Information Systems group, and Philips. In this paper the impact on the organization of logistic procedures within a diagnostic imaging department will be described, as a part of the Dutch PACS project at the Utrecht University Hospital, the clinical evaluation (as described previously in SPIE proc. #767 and #626). A detailed analysis was carried out regarding image logistics, object flow (patients, images, request forms etc.) and working procedures of all personal involved. Goal is to compare the results of these studies with similar data after the introduction of a complete PACS working procedure for the target ward and to find design parameters for the intrinsic image management system. Furthermore, the data are used to establish the efficiency of the new situation. The results of this study are described in detail per parameter.

The goal in the setup of the clinical evaluation of the PACSystem in the University Hospital Utrecht (UUH) is to realize such a presentation of images, coupled to other patient information, that radiologists and applicants can make diagnoses more flexible, faster and more fruitful. Moreover we try to get insight in the way a PACSystem can solve the problems regarding the contempory way of working with respects to diagnoses and image information; i. e. problems related to archiving and loaning of pictures, which may lead to loss and long searching times, problems with respect to room for archive, the high costs of photographic material and the impossibility of post processing on film. For the evaluation studies the following points are of importance. a. the storage of digitized image information on magnetic disc and optical disc (DOR). Direct digital input is coming from e. g. CT. While conventional films are digitised with a scanner/digitizer (MARSCAN). The image information can be seen on several monitors at the same time (multi access), i.e via MARVIEW in the digital reading room and a viewing station on the clinical ward. b. A coupling between the Hospital Information System (BAZIS) and PACS will be realized. c. The MARVIEW station offers several forms of image manipulation. In principle there is chosen for a complete PACS situation, that is to say that all information and procedures relating to the patients of a specific ward are manipulated up in the PACSystem. This specific ward is a clinical internal department and has fifteen beds. The data of this ward are compared to two comparable wards before and after installation of the PACSystem. Crucial is that in this situation we deal with a real operating clinical ward. To evaluate the effects of the introduction of such a PACSystem on diagnoses and organisation, a number of sub-projects are carried out in joint venture between the University Hospital Utrecht, BAZIS and PHILIPS. Two projects are especially of importance, namely: a. the clinical evaluation of the different system parts of the MARCOM/hard- and software the University Hospital Utrecht. b. the preparation of a simulation software package for general use (BAZIS). In this article only the setup of the clinical evaluation in the University Hospital Utrecht is discussed.

After enthusiastic visions of totally digital imaging departments projected for the end of this century a more realistic approach is needed. In Europe PAC systems are steadily evolving by stepwise upgrading imaging equipment with PACS building blocks. An extension of the ACR/NEMA standard will be presented connecting multiple imaging devices . In a majority of big hospitals the organisation of the imaging department is supported by a computerbased Radiology Information System RIS. There are a few key projects where these departmental systems are linked together with digital imaging modalities, electronic archive and digital viewing stations via a local area network.

Within a PACS environment the integration of images which have been recorded on film is a major requirement. In the past several studies on digital radiology and softcopy viewing were based on images recorded on film. Film is used as the "gold standard" for the assessment of diagnostic image quality. Based on the results of these studies filmdigitizing systems for routine operation have been specified which yield adequate diagnostic quality but not at the expense of high cost and high operational effort. In this paper different methods of film digitization will be compared. A survey of applicational results will be given.

The cumulative logistic distribution function is shown to be a robust mathematical model of film characteristic curves for a variety of radiological imaging tasks. This model allows one to write analytical functions for several important metrices in film sensitometry. In practice, curve-fitting is accomplished by linear regression analysis of the logistic distribution by forming the logit of D/Dm where D is the optical density and Dm denotes the maximum optical density obtainable. This model has been used to characterize films in radioscintigraphy, conventional film/screen combination and a multiformat camera. Linearization of the characteristic curves resulted in correlation coefficients ranging from 0.980 - 0.996.

A system is described that electronically controls the medical photography for a computed tomography (CT) scanner system. Multiple CT exams can be photographed with each image automatically adjusted to a specific gamma table presentation and positioned to any film location within a given film format. Our approach uses a library that can store 24 CT exam photography protocols. Library entries can be added, deleted, or edited. Mixed film formats, multiple image types, and automated annotation capabilities allow all CT exams to be filmed at our clinic cost-effectively and unattended. Using this automated approach to CT exam photography, one full-time equivalent CT technologist has been saved from the operational cost of our center. We outline the film protocol database, illustrate protocol options and by example, show the flexibility of this approach. Features of this system illustrate essential components of any such approach.

The use of a beam splitting device for medical gastro-intestinal fluoroscopy has demonstrated that clinical images obtained with a 100mm photofluorographic camera, and a 1024 X 1024 digital matrix with pulsed progressive readout acquisition techniques, are identical. In addition, it has been found that clinical images can be obtained with digital systems at dose levels lower than those possible with film. The use of pulsed fluoroscopy with intermittent storage of the fluoroscopic image has also been demonstrated to reduce the fluoroscopy part of the examination to very low dose levels, particularly when low repetition rates of about 2 frames per second (fps) are used. The use of digital methods reduces the amount of radiation required and also the heat generated by the x-ray tube. Images can therefore be produced using a very small focal spot on the x-ray tube, which can produce further improvement in the resolution of the clinical images.

The miniature image display is used as a visual directory of images in an examination to facilitate image selection on a display console. However, the miniature images, generated from the patient image database, are redundant information for a picture archiving system. In a compressed picture archiving system, this is neither economic nor efficient without further design of both the display and the compression subsystems. This paper reports some possible image compression algorithms, with proper image decomposition methods, which can be used efficiently for the miniature image predisplay and the full resolution diagnostic image display by sharing the same compressed image database.

Recent developments in microcomputer technology have resulted in the widespread availability of workstations with large memories, high processor speeds, and graphical displays. Because of the large number of applications in computer aided design and document preparation, most of the utilities which have been written for these workstations are designed to display wire-frame figures, shaded graphics models, or text. Such tools are not suited to the special needs of image processing. This paper describes displaytool, a utility which provides a window-based image display environment tailored for image processing. Displaytool supports subplane selection for multidimensional images; a large pixel value range, including negative values; interactive manipulation of colormaps for continuous contrast and brightness adjustment; statistical image summaries with graphical histogram display; interactive image subtraction; and simultaneous display of multiple images with arbitrary overlapping. The program runs under UN/X1 on Sun workstations within the Suntool window environment.

An "electronic viewbox" is described that has been designed to meet the demand for a modestly priced soft-copy display for radiology. Issues associated with spatial resolution, intensity resolution, image magnification, user interface, digital communications and possible applications are discussed.

There have been numerous studies of the the spatial and contrast requirements for digital radiology. In our application we are particularly interested in the overall system performance of an image workstation. As the transcribed report of the radiologist's findings is in most cases the "final product" used by the referring physician, it seems natural to base the performance measure upon the clinical value of this report. Twenty four cases from the teaching files and for which the pathologies had been verified by other means were used in the test. The films were digitized by Dr H.K. Huang of UCLA using their Konica laser-based scanner, and then displayed to the equivalent of 1000 lines of resolution for a 14-by-17 film with to 8 bits of contrast resolution. Five radiologist reported their findings by reading the analog and the digital series of films. For each report a disease vector was calculated which takes into account the disease found and the radiologist's confidence. A disease vector length is then calculated which summarizes in one value the quality of the report. Overall we find that the analog series outperforms the digital series by a small margin.

Videodiscs have introduced new capabilities in electronic storage including the publication of medical image databases. However, in several research programs conducted by the Lister Hill Center, evaluators found that videodiscs mastered in the standard NTSC television format did not deliver sufficient spatial and brightness resolution to reproduce medical images, such as radiographs, accurately enough for diagnostic usage. To explore techniques to upgrade the quality of medical monochrome images, a PC-based X-Ray Imaging System (XRIS) with CD-ROM storage and a standard video display was developed. XRIS was used as a test bed to evaluate digital storage as a means to improve the signal-to-noise ratio of electronically stored radiographs, and investigate the application of windowed imagingdisplay techniques to improve the spatial resolution. The system was designed for compatibility with small computer-controlled retrieval software and low cost NTSC displays. Test results showed that the XRIS prototype delivers improved image quality over the analog videodisc case both in terms of signal-to-noise ratio and spatial resolution.

Clinical experience with a computer program for reconstructing and visualizing three-dimensional (3-D) structures is reported. Applications to the study of soft-tissue and skeletal structures, such as the temporomandibular joint and craniofacial anatomy, using computed tomography (CT) data are described. Several features specific to the computer algorithm are demonstrated and evaluated. These include: (1) manipulation of density windows to selectively visualize bone or soft tissue structures; (2) the efficacy of gradient shading algorithms in revealing fine surface detail; and (3) the rapid generation of cut-away views revealing details of internal structures. Also demonstrated is the importance of high resolution data as input to the 3-D program. The implementation of the program (VoxelView-32) described here, is on a MASSCOMP computer running UNIX. Data were collected with General Electric or Siemens CT scanners and transferred to the MASSCOMP for off-line 3-D recon-struction, via magnetic tape or Ethernet. An interactive graphics facility on the MASSCOMP allows viewing of 2-D slices, subregioning, and selection of lower and upper density thresholds for segmentation. The software then enters a pre-processing phase during which a volume representation of the segmented object (soft tissue or bone) is automatically created. This is followed by a rendering phase during which multiple views of the segmented object are automatically generated. The pre-processing phase typically takes 4 to 8 minutes (although very large datasets may require as much as 30 minutes) and the rendering phase typically takes 1 to 2 minutes for each 3-D view. Volume representation and rendering techniques are used at all stages of the processing, and gradient shading is used for enhanced surface detail.

Artifacts seen on laser digitized radiographs are analyzed and mathematically explained based on the concept of image contrast. Our investigation considered the following determinants: 1)laser spot size, 2)signal processing components, 3)image characteristics, and 4)observer performance. A functional relationship between the sampling interval, laser spot size, and the contrast of the artifact seen on the digitized image is derived. This relationship is verified experimentally using a variable spot size laser source and sampling at interval of 175.0 microns. The problem of digitizing artifact seen on film images obtained with anti-scatter grids is discussed.

Low-cost image processing systems which can provide convenient access to image processing and analysis techniques hold great potential as diagnostic and research tools in medical imaging. At the University of Washington, we have developed a PC-based medium performance image processing system for use as an experimental radiological workstation. The workstation uses a standard IBM PC/AT personal computer augmented with a custom designed image processor implemented on two IBM PC/AT prototyping boards. Features of the system include up to 52 512 x 512 x 8 bit frame buffers (4 on the image processor board and up to 48 in the host computer memory) and a 512 x 512 x 4 bit graphics overlay memory, hardware zoom, pan and scroll, pseudo coloring, and a 60 Hz noninterlaced display. Many image processing and analysis functions are provided in this workstation, and all user requests are supported in an interactive fashion. For example, arithmetic and logical point operations between two 512 x 512 frame buffers require approximately 170 ms, while computationally intensive functions such as an 11 x 11 convolution or a full screen geometric transformation (warping) can be completed in less than 10 seconds. A full screen 2-D Fast Fourier Transform (FFT) and Inverse FFT (IFFT) based on the row-column method can be completed in less than 20 seconds. The developed system can easily be configured into a DIN/PACS workstation or a biological imaging system. Hardware and software details of this workstation as well as user interface functions implemented will be discussed in the paper.

Three dimensional medical imaging systems such as CAT, PET, and MRI are now producing imagery of such complexity that the traditional slice presentations are insufficient for communicating the information to the observer. This paper discusses the use of computer generated holography as a vehicle for presenting this information in the form of holographic stereograms.

Volume reconstruction algorithms that provide a three-dimensional (3D) rendering of a tomographic study are now available but not directly applicable to magnetic resonance (MR) studies. The problem with images in such studies is that the gray ranges of tissues overlap. To overcome it, we have developed an interactive graphics editor (IGE) that allows its user to define subregions within the volume of MR images which comprises such a study. The editor performs a statistical analysis of the volume contained in each subregion. With this information the user can enhance individual subregions with such techniques as user-defined remapping of gray levels, the addition of color, the addition of transparency, and spatial filtering. The editor is implemented on a Pixar Image Computer which uses as a host a Sun 3/180 computer. It has been used by us to create from MR studies three-dimensional images of the brain and knee.

The basic goal of this paper is to present the characteristics of those frame buffers currently used to display images, versus those features which might more ideally suit frame buffers for this purpose. We use current and perceived needs and data based on our experience to identify the salient points which are useful and necessary. Over the last five years we have implemented the same image processing system on eight different frame buffers' (RAMTEK 9050 and 9400, Grinnell GMR-270, Hewlett-Packard 98710 and 98720, Nodecrest Ltd. TVN 256, Lexidata 3400, Pixar Image Computer). This paper extrapolates from these efforts on existing displays to determine what characteristics are desirable in a frame buffer used primarily for multimodality image display. Two case examples are presented: 1. A system developed for high quality computer graphics, as exemplified by the Pixar Image Computer (or in the future the AT&T PXM900 Pixel Machine). 2. A system developed for nuclear medicine and radiation therapy treatment planning as exemplified by the Nodecrest Ltd. TVN256 and TVN512 displays.

Several techniques now well established in computer graphics and graphic animation are combined in this work to develop a realistic presentation of anatomic structures. A rudimentary type of ray tracing is implemented for routine clinical CT exams. In particular, we describe our method in the context of a standardized cross-reference set of multiplanar reformatted CT pictures. In addition, an image coherence technique is briefly outlined that speeds rendering of a series of 3-D views. This paper describes a method by which object surface information from an existing view is used to help predict where ray tracing can begin to search for ray object intersections in the subsequent view. This method is shown to reduce the computational expense of finding ray-object intersections by beginning this search in the proximity of object surfaces. Finally, we have casted shadows in the scene of objects rendered by our method. Several example images illustrate our results.

The display of three-dimensional medical data is becoming more common, but current display hardware and image rendering algorithms do not generally allow real-time interaction with the image by the user. Real-time interactions, such as image rotation, utilize the motion processing capabilities of the human visual system, allowing a better understanding of the structures being imaged. Recent advances in general purpose graphics display equipment could make real-time interaction feasible in a clinical setting. We have evaluated the capabilities of one type of advanced display architecture, the PIXAR1 Image Computer, for real-time interaction while displaying three-dimensional medical data as two-dimensional projections. It was discovered during this investigation that the most suitable algorithms for implementation were based on the rendering of voxel rather than surface data. Two voxel-based techniques, back-to-front and front-to-back rendering produced acceptable, but not real-time, performance. The quality of the images produced was not high, but allowed the determination of an image orientation which could then be used by a later, high-quality rendering technique. Two conclusions were reached: first, the current performance of display hardware may allow acceptable interactive performance and produce high-quality images if a scheme of adaptive refinement is used wherein succesively higher quality images are generated for the user. Second, the correct algorithm to use for fast rendering of volume data is highly dependent upon the architecture of the display processor, and in particular upon the ability of the processor to randomly access image data. If the processor is constrained to sequential or near-sequential access to the voxel data, the choice of algorithms and the utilization of parallel processing is severely limited.

A workstation has been developed to evaluate computed tomographic (CT) image data in 2 and 3 dimensions. The workstation consists of an independent image display station (Independent Viewing and Analysis Station or WAS, International Imaging Systems, Inc., Milpitas, Calif.) and a VAX host computer. The WAS has 1024 X 1024 X 24 bits of image memory plus 4 bits of graphics overlay. An independent VLSI graphics processor and 1024 X 1024 X 4 bit graphics memory, independent of the image memory, are included in the self-contained WAS unit. A local microprocessor host (Motorola 68000 microprocessor) controls the IVAS from directives obtained through a direct memory access channel to the VAX host. This facilitated the creation of an object oriented software enviroment for the IVAS under control of a VAX host program written in the C language. The workstation created has an interactive user interface consisting of a mouse and pull-down menus. The workstation enables loading multiple images, typically 256 x 256 or 512 x 512, into the 1024 X 1024 frame buffer. Once loaded, the images can be manipulated by applying gray scale transforms, editing them and performing 3-D reconstructions from serial sections. Algorithms for three dimensional (3-D) reconstructions were implemented in the VAX/VMS host computer environment and are available on the workstation through special menu functions for handling these reconstructions. The functions interactively combine depth and gradient shading of surfaces to suit specific applications in craniofacial surgical planning or orthopedics. This workstation is user friendly and is very easy to handle. A workstation of this type may become a popular tool for physicians and surgeons in evalution of medical images.

This paper describes the management of CR images and discusses the feasibility and utility of the chest examination with no X-ray film, but simple based on CRT. In addition, possible use of digital images for conferences has been examined as well, in regard to its feasibility and utility.

Picture archiving and communication system (PACS) are total-imaging system which are clinically useful in improving the quality of diagnostic images and facilitating the day-to-day evaluation of patients. The most important consideration for clinicians is the feasibility of producing reliable diagnostic images on the CRT screen of a PACS, which does not employ film. In order to evaluate this point, X-ray films demonstrating various pathological findings were digitized, image-processed, and interpreted on CRT. The results of CRT interpretation were compared with the findings obtained using conventional films alone. As a result of this comparison, it was demonstrated that the CRT images were diagnostically superior to conventional X-ray films or to recent CR images if image processing techniques were optimized to exhibit the findings of various pathological conditions.

The TAAC-1 Application Accelerator is a specialized processor with an embedded full color display subsystem that is installed in a Sun workstation. The flexibility, performance, and graphics capability of this device make it well suited for medical imaging applications. The architecture of the TAAC-1, software tools, and results of 2D and 3D medical image processing and display applications are described.

The Digital Imaging Network (DIN) Project is a collaborative project among numerous components of the Department of Defense, Public Health Service, Veterans Administration, industry, academia, and the MITRE Corporation. The project is evaluating prototype DIN systems (DINS) at Georgetown University (in collaboration with George Washington University) in Washington, DC, and at the University of Washington in Seattle, WA. Results of the project will be used to plan DINS for implementation in fixed and deployable military medical care facilities in the 1990's.

The MITRE Corporation is currently under contract to the U.S. Army Medical Research and Development Command to evaluate Digital Imaging Network Systems (DINS) technology at two university evaluation sites. Equipment developed by Philips Medical Systems will be evaluated at the University of Washington in Seattle and AT&T's CommView system will be evaluated at Georgetown University. While other papers at the 1988 PACS II conference will discuss the details of the systems being evaluated at the university sites, this paper discusses the reasons that the Army is considering the use of DINS in its fixed facilities and in deployable medical system (DEPMEDS) applications. Reasons for converting to DINS technology will be discussed and expected benefits will be presented.

The increased use of digital imaging and data base management technologies has set the stage for the development of comprehensive medical image and information management system for large medical facilities. An image management and communication system (IMACS) can take on a variety of forms depending on specific user needs. In one form or another, the technology of IMACS is a part of radiology's future.

The Department of Radiology at the University of Washington was selected, along with Georgetown and George Washington Universities, to perform a clinical evaluation of available PACS technology. Although planning of our system (to be supplied by Philips Medical Systems) is well along, no equipment has yet been installed. In this paper, we describe an integrated approach to the clinical evaluation of particular PACS systems, which will be applied to the prototype system after it is in clinical use.

The Laser Archive, Review and Communications (LARC) system is designed primarily for the archiving and reviewing of Magnetic Resonance (MR) images. After the images are reconstructed on the Magnetic Resonance imager, they are sent, via a direct parallel connection, to the Laser Archive, for storage on optical disk. By using an Ethernet Local Area Network (LAN), the images can be sent from the archive for display on Remote Viewing Systems (RVS), or can be sent back to the imaging system for display on the main operator's console.

The University of North Carolina at Chapel Hill PACS research effort has been primarily concerned with the conceptualization of the problems and solutions necessary to achieve a viable clinical system. Our approach has emphasized a clear understanding of the operation of the existing department, and our implementation to date has attempted to remain as generic as possible. Our major areas of research have been systems analysis of the radiology operation (modeling), the ACR-NEMA standard, and display/analysis consoles. The program is now moving into a new phase, in which the emphasis is shifting toward the establishment of a clinical "service" in the neuro radiology area.

Kyoto University Hospital is currently developing a prototype PAC system named KIDS (Kyoto univ. hosp. Image Database and communication System). The present goal of the system is to achieve the totally digital CT and MRI unit in the radiological department. Because KIDS is designed as a first step of a long-range plan towards a hospital wide system, it includes all of the basic functions required in realizing the PAC system, such as communication networks, a long term archiving unit, a laser film printer and image workstations. The system concept, architecture and current status are described in this paper. Our early experience and evaluations with the system in a clinical environment are also mentioned.

The Medical Imaging Department at Victoria General Hospital is the first in Canada to implement an integrated multi-modality picture archive and communication system for clinical use. The aim of this paper is to present the current status of the picture archive and communication system components and to describe its function. This system was installed in April of 1987, and upgraded in November of 1987. A picture archive and communication system includes image sources, an image management system, and image display and reporting facilities. The installed image sources (digital radiography, digital fluoroscopy, computed tomography, and digital subtraction angiography) provide digital data for the image management system. The image management system provides facilities for receiving, storing, retrieving, and transmitting images using conventional computers and networks. There are two display stations, a viewing console and an image processing workstation, which provide various image display and manipulation functions. In parallel with the implementation of the picture archive and communication system there are clinical, physical, and economic evaluations being pursued. An initial examination of digital image transfer rates indicate that users will experience similar image availability times as with conventional film imaging. Clinical experience to date with the picture archive and communication system has been limited to that required to evaluate digital imaging as a diagnostic tool, using digital radiography and digital fluoroscopy studies. Computed tomography and digital subtraction angiography have only recently been connected to the picture archive and communication system. Clinical experience with these modalities is limited to several cases, but image fidelity appears to be well above clinically acceptable levels.

A direct link between a clinical pediatric PACS module and a FONAR MRI image network was implemented. The original MR network combines together the MR scanner, a remote viewing station and a central archiving station. The pediatric PACS directly connects to the archiving unit through an Ethernet TCP-IP network adhering to FONAR's protocol. The PACS communication software developed supports the transfer of patient studies and the patient information directly from the MR archive database to the pediatric PACS. In the first phase of our project we developed a package to transfer data between a VAX-111750 and the IBM PC I AT-based MR archive database through the Ethernet network. This system served as a model for PACS-to-modality network communication. Once testing was complete on this research network, the software and network hardware was moved to the clinical pediatric VAX for full PACS integration. In parallel to the direct transmission of digital images to the Pediatric PACS, a broadband communication system in video format was developed for real-time broadcasting of images originating from the MR console to 8 remote viewing stations distributed in the radiology department. These analog viewing stations allow the radiologists to directly monitor patient positioning and to select the scan levels during a patient examination from remote locations in the radiology department. This paper reports (1) the technical details of this implementation, (2) the merits of this network development scheme, and (3) the performance statistics of the network-to-PACS interface.

A local Picture Archiving and Communication System (PACS) has been operational in the Division of Nuclear Medicine at Robert Wood Johnson University Hospital for the past five years. Recently, it has been interfaced to a total PACS which is based on different hardware and software. Using this hybrid system, we describe our initial efforts to facilitate the short and long term archiving of NM studies and the use of combined image displays for correlative image analysis.

A Radiology Operations System (ROS) integrates the function of a PACS, an RIS and an HIS interface by means of unified telecommunications. The essential elements of the ROS are: 1. Picture Archival and Communication System (PACS): This system provides capabilities to acquire, display, manipulate and store images and text data required by the radiology department and its clinical and academic clients. 2. Radiology Information System (RIS): This system supports the text data needs of the department. It specializes in clerical and administrative activities. 3. Hospital Information System (HIS) Interface: This software supports the interface between the radiology department and the users of radiology department information. 4. Network: The network supporting an ROS moves text and image data throughout the department and into the hospital community. A variety of technologies are included in the network and applied as the needs for speed and volume dictate. This paper will examine the ROS concept, showing that the support provided is greater than the sum of its constituent systems. A case study is presented to illustrate the benefits that accrue from the implementation of an ROS.

For PACS to successfully replace the current film-based diagnosis system , there must be certain merits on PACS . None will use any system that may not work at least as smoothly as the current system . PACS is a big , expensive system . So , it must be at least equivalent or compatible to the current system from the performance point of view . We think the most promising way to achieve larger or full PACS is to create a system that has at least equal ability as the current film-based system , and then refine it so that the users really want to manipulate in their daily work . The point we must be aware of is that even if the prototype system is excellent in several points , when it should have defects in one point, this system may not be regarded as superior to the current system. Our way to establish PACS is to create a small system that has at least all-modality capability and high-speed processing power , and do a cooperate research on the real clinical environment and gather useful informations from them . Then we can develop new image processing methods , new mam-machine interface ,and any unforeseen features that are necessary to this kind of system . We have developed TDF-500 system from this viewpoint , and have also installed some of this system into real clinical environments . Overall reputations are good so far , and we are confident that PACS work station must meet the very basic but important features such as all-modality , multi-screen , and easy operation . Based on these informations , we will develop new technologies that must be necessary for future PACS .

In January, 1984, the American College of Radiology (ACR) representing the users of imaging equipment and the National Electrical Manufacturers Association (NEMA) representing the manufacturers of imaging equipment joined forces to create a committee that could solve the compatibility issues surrounding the exchange of digital medical images. This committee, the ACR-NEMA Digital Imaging and Communication Standards Committee was composed of radiologists and experts from industry who addressed the problems involved in interfacing different digital imaging modalities. In just two years, the committee and three of its working groups created an industry standard interface, ACR-NEMA Digital Imaging and Communications Standard, Publication No. 300-1985. The ACR-NEMA interface allows digital medical images and related information to be communicated between different imaging devices, regardless of manufacturer or use of differing image formats. The interface is modeled on the International Standards Organization's Open Systems Interconnection sever-layer reference model. It is believed that the development of the Interface was the first step in the development of standards for Medical Picture Archiving and Communications Systems (PACS). Developing the interface Standard has required intensive technical analysis and examination of the future trends for digital imaging in order to design a model which would not be quickly outmoded. To continue the enhancement and future development of image management systems, various working groups have been created under the direction of the ACR-NEMA Committee.

An image archive has been designed to manage and store radiologic images received from within the main Hospital and a from a suburban orthopedic clinic. Images are stored on both magnetic as well as optical media. Prior comparison examinations are combined with the current examination to generate a 'viewing folder' that is sent to the display station for primary diagnosis. An 'archive-manager' controls the database managment, periodic optical disk backup and 'viewing-folder' generation. Images are converted into the ACR/NEMA message format before being written to the optical disk. The software design of the 'archive-manager' and its associated modules is presented. Enhancements to the system are discussed.

A satellite image transmission system has been installed between an orthopedic clinic in suburban Philadelphia and the Hospital of the University of Pennsylvania in central Philadelphia. The two facilities are approximately 15 miles apart. Images at the clinic are acquired with a laser scanning film digitizer at a 200 micron resolution. The resulting data (1680x2048x12 bits) are compressed using a non-destructive algorithm and transmitted via a Ku-Band satellite link to the hospital. The effective transmission rate is approximately one Mbit/second. Upon receipt, the images are restored before being archived and displayed on the existing Medical Image Management System in the Radiology Department. This paper describes the flow of image data and presents an overview of the technical aspects of the satellite transmission system. System parameters are tabulated and enhancements are also discussed.

This paper outlines the storage requirements for Picture Archiving and Communications Systems (PACS) and strategies for meeting those requirements. A radiology archive, either film based or PACS, follows a layered model, with each layer characterized by a required access time and typical retention time. For this paper, a three layer model is considered. In the first layer are images from newly acquired exams awaiting interpretation, and images from recently interpreted exams which may be needed for reference as part of continuing patient management. In the middle layer are images from exams which, while not currently active, are likely to be needed for either the diagnosis of new exams or for reference. In the final layer are images which are not active and are least likely to be required to support diagnosis or patient management. In a film based environment these layers correspond to the active file room or reading areas, the main (2 year) file room, and the inactive file room. In this model, the PACS will be defined in terms of a hierarchy of 1 week, 1 or 2 years, and anything older. Image generation rates have been gathered from surveys of nine hospitals. These data have been used to derive "large" (600-700 bed) and "small" (300-500 bed) target environments which generate the capacity requirements for the layers. These, coupled with the access time requirements, are used to identify data storage technology with each layer.