Acoustic nuclear magnetic resonance (ANMR) has been shown to be an effective tool for characterizing the nuclear spin system in solids. In conventional NMR, the transitions between nuclear spin energy levels are induced by the interaction of an RF magnetic field with the nuclear magnetic dipole moment. In ANMR the energy exchange between the acoustic wave and the nuclear spin system can take place through the periodic perturbation of either the magnetic dipole-dipole interaction distance or the electric field gradient, the latter acting in turn upon the nuclear electric quadrupole moment. We should expect to gain different medical knowledge about a sample from these two NMR techniques as a result of their distinct interaction mechanisms. Little is known about the potential for using acoustic nuclear magnetic resonance in biological tissues. We are currently addressing many fundamental questions concerning the possible ANMR interaction mechanisms in tissues and their possible uses for medical diagnosis.

The current research, potential applications, and utilization of 19F-NMR techniques in medicine and biomedical studies are reviewed. The wide ranging possibilities involving spectroscopy and/or imaging include investigations of: metabolic processes using fluorine labeled substrates; enzymatic activity, reaction mechanisms and biomolecular binding; membrane and cell systems; perfluorocarbon compound utility; biodistribution and retention of fluorinated materials including anesthetics; and, gas phase applications.

Perfluorocarbon compounds (PFC's), well known in industry and of late as synthetic oxygen carriers, have a wide range of significant applications in diagnostic imaging. Their enhancement effect is detectable by ultrasound and magnetic resonance and if radiopaque, such as perfluoroctylbromide (PFOB), by standard radiography and computed tomography (CT). We have utilized PFOB as a CT contrast agent to enhance the blood pool, and as both a CT and an ultrasound contrast agent to enhance the liver, spleen, abscesses, infarctions, and tumors or any tissue where inflammatory cells can be found. PFC's, except for the echogenic enhancement of the vascular space on their first pass to the lung, do not enhance the blood pool on ultrasound. Otherwise, ultrasound applications are similar to those observed for CT. Fluosol, which was available for human trials, is not radiopaque and therefore served as an ultrasound contrast agent. In a preliminary clinical trial, Fluosol produced tumor enhancement in man at 1.6g/kg allowing the visualization of previously missed lesions and liver and spleen enhancement at 2.4g/kg allowing the visualization of previously missed non-enhancing lesions. Perfluorocarbon toxicity seems to be related to the constituents of the emulsion rather than the perfluorocarbon itself. Improvements in the emulsifier and emulsification technology has yielded stable emulsions at high concentrations and low toxicity.

The techniques of statistical pattern recognition are implemented to determine the best combination of tissue characterization parameters for maximizing the diagnostic accuracy of a given task. In this paper, we considered combinations of four ultrasonic tissue parameters to discriminate between normal liver and chronic hepatitis. The separation between normal and diseased samples was made by application of the Bayes test for minimum risk which minimizes the error rate for classifying tissue states while including the prior probability for the presence of disease and the cost of misclassification. Large differences in classification performance of various tissue parameter combinations were demonstrated by ROC analysis. The power of additional features to classify tissue states, even those derived from other imaging modalities, can be compared directly in this manner.

Advances in neurobiology today are as important as the advances in atomic physics at the turn of the century and molecular genetics in the 1950's. Positron-emission tomography is participating in these advances by making it possible for the first time to measure the chemistry of the living human brain in health and disease and to relate the changes at the molecular level to the functioning of the human mind. The amount of data generated requires modern data processing, display, and archiving capabilities. To achieve maximum benefit from the PET imaging and the derived quantitative measurements, the data must be combined with information, usually of a structural nature, from other imaging modalities, chiefly computed tomography and magnetic resonance imaging.

This study presents preliminary results of dual energy imaging techniques in film digital radiography applicable for clinical use. The studies were carried out by single x-ray exposure as well as double exposure technique. The single exposure technique was accomplished by placing two sets of screen/film combinations in a single cassette with an energy separation filter between them. The two optimal energies defined as the largest energy separation between the high and the low energy image were selected by a computer simulation followed by an experimental verification. A humanoid and a wedge-shaped plastic phantom were used to test the image quality. The images on the films were digitized by a high resolution laser scanner. The quality of the separated images though slightly inferior to the double exposure technique, was acceptable for clinical application.

Principal components analysis, PCA, is basically a data reduction technique. PCA has been used in several problems in diagnostic radiology: processing radioisotope brain scans (Ref.1), automatic alignment of radionuclide images (Ref. 2), processing MRI images (Ref. 3,4), analyzing first-pass cardiac studies (Ref. 5) correcting for attenuation in bone mineral measurements (Ref. 6) and in dual energy x-ray imaging (Ref. 6,7). This paper will progress as follows; a brief introduction to the mathematics of PCA will be followed by two brief examples of how PCA has been used in the literature. Finally my own experience with PCA in dual-energy x-ray imaging will be given.

A dual energy imaging technique was developed in a digital fluoroscopic unit and a method for calculating the thickness of bone and effective soft tissue was developed using the image data obtained from tissue equivalent phantom material. The thicknesses were then combined by using their densities and a single weighting factor was obtained at each pixel location and was used for modifying the attenuation and scattering property of the Cobalt 60 gamma radiation, for heterogeneity correction in clinical dosimetry. A sector integration algorithm was developed and used for dose calculation in an irregularly shaped radiation field used for treatment of lung cancer. The calculated doses after heterogeneity correction were compared with measured values in phantom using film densitometry. The performance parameters of the imaging system such as the dynamic range, linearity, spectral property and the geometrical distortion were evaluated and discussed,

The contrast transfer function (CTF), noise power spectrum (NPS), noise equivalent quanta (NEQ), and detective quantum efficiency (DQE) of a high-resolution screen-film combination have been measured for exposure to a 30 kVp x-ray spectrum. In addition to these overall system characteristics, selected component sources of noise in the screen-film combination have been determined experimentally. These data are interpreted in terms of a simple model which is used to predict screen-film NPS, NEQ, and DQE. Reasonable agreement between model predictions and experimental measurements has been found.

The measurement of the Wiener spectrum associated with screen-film noise has beep widely reported in recent years, and digital data obtained by microdensitometer slit-scanning is now almost a standard art. However, physical autocorrelation techniques, described and developed for other types of imagery, might have powerful advantages in some respects. An analysis is carried out of the fundamental statistics associated with the two approaches, and some attractive features associated with physical autocorrelation are illustrated.

A method is described for computer simulation of radiographic screen-film images. This method is based on a previously published model of the screen-film imaging process.l The x-ray transmittance of a test object is sampled at a pitch of 50 μm by scanning a high-resolution, low-noise direct-exposure radiograph. This transmittance is then used, along with the x-ray exposure incident upon the object, to determine the expected number of quanta per pixel incident upon the screen. The random nature of x-ray arrival and absorption, x-ray quantum to light photon conversion, and photon absorption by the film is simulated by appropriate random number generation. Standard FFT techniques are used for computing the effects of scattering. Finally, the computed film density for each pixel is produced on a high-resolution, low-noise output film by a scanning printer. The simulation allows independent specification of x-ray exposure, x-ray quantum absorption, light conversion statistics, light scattering, and film characteristics (sensitometry and gran-ularity). Each of these parameters is independently measured for radiographic systems of interest. The simulator is tested by comparing actual radiographic images with simulated images resulting from the independently measured parameters. Images are also shown illustrating the effects of changes in these parameters on image quality. Finally, comparison is made with a "perfect" imaging system where information content is only limited by the finite number of x-rays.

Together with the Modulation Transfer Function(MIT)the Detective Quantum Effiency (DQE) forms the physical base for the description of the information transfer of image intensifiers. The measurement of the DQE relies on the determination of the signal to noise ratios in the input and output information flows. The most difficult part of the measurement is the measurement of the temporal behaviour of the combination of measuring equipment and testspecimen. The temporal characteristics of this combination determine the sampling time that is used in computing the signal to noise ratios from DC and RMS data. If the range of X-ray image intensifiers under consideration is restricted to the family of CsI inputscreen and P20 phosphor outputscreen type image intensifiers, then it is possible to design a rather simple measuring procedure, that can be used as a standard for comparing DQE measuring results.

A particular system for computed radiography uses a photostimulable phosphor which is scanned by a laser that reads and also erases the storage medium. The process of reading and erasing is called destructive scanning, and a model of its effects is presented here. Model predictions of system dc gain vs. scanning exposure for the case of raster scanning with a Gaussian beam are compared with measured data.

A model for light scattering and absorption effects is applied to a system for computed radiography in which a turbid photostimulable phosphor is scanned with a laser. The spatial spreading of the laser light and the escape probability of the stimulated light are important in determining system response, and models for their effects are presented. Model calculations of system MTF are compared with measured data.

A technique for tomosynthesis utilizing a computed radiographic system is described in this paper. The process involves taking a series of radiographic images of an object at various angles using a linear motion tomographic unit and a scanning laser-stimulated luminescence plate as the image receptor. The digital image data obtained with the imaging plate system are transferred to a computer for tomographic reconstruction. The reconstruction algorithm utilizes the knowledge of the geometric scan data of the tomographic unit, and it shifts each angular projection in such a way that only one plane is in register when they are added together. This technique was evaluated using an anatomical phantom. The preliminary results offer tomographic images that are comparable to those obtained by the conventional film technique.

Low angle x-ray scattering at diagnostic energies in narrow beam geometry is due to coherent (Rayleigh) and incoherent (Compton) scattering. It has been found that single coherent scatter dominates below 10 deg. Interference effects with coherent scatter leads to diffraction patterns which differ from material to material. A technique, analogous to conventional CT, allows the reconstruction of the 2D distribution of the x-ray diffraction properties within an object slice, as demonstrated recently? Use of the bremsstrahlung spectrum of an x-ray tube permits short measuring times, but causes a significant energy broadening of the diffraction curves, thus deteriorating the maximum contrast obtainable by diffraction imaging. With energy resolved photon counting of the scattered x-ray quanta this broadening can be corrected, yielding an image contrast approaching that of a monochromatic x-ray source.

A total PACS will be inevitable for radiology practice within several years. To achieve a total PACS for radiology, a satisfactory digital radiographic unit is required, because approximately 65% of digital data for PACS comes from digital radiographs. There are several possibilities for producing digital radiographs, and 3 - 4 companies have been marketing digital radiographic devices. Some data regarding the digital radiographic units on the market are compared. It will aid in assessing the current status and availability of this aspect of development, as well as providing a summary of further development of digital radiographic technology.

The goal of developing an on-line electronic digital radiographic (EDR) system to replace conventional film-screen radiography (FSR) is important for at least two reasons. First, theoretical arguments show that EDR can have improved diagnostic quality, reduced patient dose and faster image accessibility than FSR. Secondly, the availability of EDR systems will remove the final impediment to the realization of the PACS concept inasmuch as FSR is the only major nonelectronic imaging modality left in the modern radiology department. The Kinestatic Charge Detector (KCD) has properties which make it a candidate for an on-line EDR systems-10. The KCD is a strip detector with high spatial resolution in two dimensions. However, mechanically and electronically, it operates like a one-dimensional detector. Thus, it can effectively scan on the order of 64 to 128 parallel x-ray lines simultaneously but with a 64 to 128-fold reduction in the number of actual detector cells and electronic channels. Moreover, this can be done at quantum detection efficiencies approaching unity. In this paper, theoretical calculations and experimental measurements of the performance parameters of a KCD are presented. Some of the particular parameters discussed include spatial, contrast, and temporal resolution.

Apart from providing the inherent benefits of digital imaging, a digital chest system should preferably be superior to large-size film with respect to diagnostic quality. Systems, demonstrated until now, tend to fail in that respect. Here we describe a system -under development- that combines the advantages of film (e.g. high resolution), area detectors (wide sensor dynamic range) and slit detectors (efficient reduction of scattered radiation, high contrast resolution and instantaneous image availibility). Thereby the system is low-dose, compact and operating at standard X-ray tube loading.

Suppression of the low energy component of diagnostic x-ray spectra is achieved in most cases by using aluminum filtration. The use of heavier elements alone or in conjunction with aluminum has been suggested to improve image contrast and reduce radiation dose. Previous studies have attempted to determine the optimum x-ray filter on the basis of the total attenuation coefficients alone. In this study we characterize x-ray filters by accounting for their absorption and scattering properties. Our results show that the standard aluminum filters produce scatter radiation which can account for 10-15% of all scatter. This effect is especially significant in radiographic procedures such as mammography, where scattered photons contribute to a reduction in radiographic contrast and lesion detectability. Heavier filters such as lead can reduce this scatter component by as much as 80%. Coherent scattering at an angle from 0 to 15° is responsible for the major portion of the scatter from aluminum filters at small scatter angles. This is demonstrated by the ratio of the scatter cross-section in a small volume element in the forward direction to the total cross-section for both coherent and incoherent scattering. We have concluded that an optimum filter can be selected on the basis of its ratio of the photoelectric to scatter components of attenuation. This criterion may also be extended to tissue compensating filters used in radiography for exposure equalization and computed tomography filters.

Capacitive sensing of x-ray generated charge images on amorphous selenium layers represents an attractive technique for the electronic recording of projection radiographs. In order to evaluate the practical limits of image quality with such a selenium detector we have set up a laboratory system that allows the recording of small size (70 x 70 mm2) radiographic images. We have measured MTF and noise power spectra and have calculated from these data noise equivalent quanta (NEQ) and detective quantum efficiency (DQE) as a function of dose and spatial frequency. On the basis of these quantities the imaging performance of the system is discussed.

In this paper we describe a rotating dynamic phantom which allows quality control of hardware and software for gated gamma camera systems in nuclear medicine. The phantom not only allows simulation of a gated heart study but also testing of the response of the whole system to time frequencies.

The gray level dynamic range of clinical MR transaxial head images from four manufacturers were studied by examination of image bit planes for the presence of spatial information or noise. Both calculated and measured voxel signal-to-noise ratios were determined. Also, the power spectrum of image bit planes were produced to demonstrate that the noise of lower bit planes is spectrally white and contain no information. These observations point to a six and at most seven bit gray level dynamic range for the images studied.

Relaxation times T1 and T2 and proton density NM) are intrinsic tissue properties which dictate in part the appearance of magnetic resonance images. In this work methods are presented for estimating these quantities using high speed fitting algorithms. A technique is described which uses a digital video processor for generating a computed T1 image from two acquired images in approximately one second. Comparable results are possible with T2 fitting as well.

This paper presents a computer simulation applicable to any magnetic resonance imaging method. The simulation solves the phenomenological Bloch equation numerically for time-dependent gradient fields, applying pulse rotation and time-evolution matrix operators in the rotating frame to each element of a digital phantom. Application of these operators yields. the global magnetization, calculated for each sampling time to produce the discrete Free Induction Decay (FID). A collection of these FIDs are then inverse Fast Fourier Transformed into the desired spatial image for display on a video monitor using an image processor. Since the simulation utilizes a general matrix operator approach to pulse app-lication and FID sampling, it may be used to simulate any MR imaging method employing var-ious pulsing sequences, gradient profiles and sampling intervals. This model is used to emulate a 2DFT phase-encoded spin-echo sequence and also very fast MR imaging using small angle excitation. A discussion of the computational problems involved and the simulations efficacy are included.

Human observers behave as if they have two sources of intrinsic variability, commonly referred to as internal noise. One component (here referred to as "constant") is independent of the external noise level but does depend on mean display luminance. The other component (referred to as "induced") is directly proportional to the external noise level and dominates when the display noise is easily visible. The induced internal noise is predicted by two models - one based on intrinsic signal parameter jitter and the other on a zone of indecision. Spectral density is the appropriate measure for internal noise.

We investigated the effect of quantization on the signal contrast, noise, and signal-to-noise ratio at three stages of the digital imaging chain, namely, acquisition, processing and display. Histogram analysis was performed for simple objects of constant magnitude superimposed on a uniform noisy background. After quantization, physical quantities, such as the signal contrast, noise and SNR, will vary depending on the fractional displacement of the quantization bin relative to the analog values. These physical quantities on average can provide information on the effectivenesss of a given A/D converter. As the rms value of the analog noise decreases, the various physical quantities become more sensitive to the effect of quantization. On average, the rms noise value after quantization will gradually increase as the quantization increment increases. The average SNR after quantization decreases initially as the quantization increment increases and then it misleadingly increases relative to the SNR prior to quantization. As quantization or further bit reduction becomes coarser, the noise Wiener spectrum will on average increase, and this increase will become relatively larger when the original noise data are obtained with higher x-ray exposures or larger sampling apertures. Image processing techniques allow for the potential increase in the number of bits in the image. However, the conventional CRT monitor is capable of only displaying 256 grey levels. At the display stage, though, observers may be able to detect signals having a contrast of less than a single display grey level due to the capability of the noise to carry the signal. However, in some cases, the potential of various processing techniques may be limited by the display device.

Observers rated the likelihood that two circular features on a CT image differed in size, or the likelihood that a ring was present on a CT image. The ring features were produced by subtracting the image profiles of the two disks that had to be distinguished. Since a physical cross-correlator used the same formal calculations on the image pixel values in both tasks, the signal-to-noise ratio (SNR) for the size discrimination task could be predicted from the values of SNR in the ring discrimination task. The 15 sets of images in both tasks studied three levels of feature contrast for discrimination with the relative difference in the disks' diameters adjusted to vary observers' performance over a measureable range. Indices of performance accuracy, as estimated from observers' ROC curves, were well predicted by assuming that they were proportional to the calculated values of Task SNR. The fitted constants of proportionality were similiar in both the size-discrimination (0.470) and ring-detection (0.556) tasks. Thus the human. observer's processing of image information in both the size-discrimination and ring-detection tasks was similar to, but less efficient than, that of the physical cross-correlator.

Assessment of image quality is an important and long-standing problem. To the user of a medical imaging system, it is important to be able to determine the expected performance of the system in various clinical applications. For the system designer, it is important to know how alteration of various design parameters influences performance.

With the large number of images that will be viewed simultaneously in a medical picture archiving and communication system (PACS) system in the diagnosis of a particular patient, image by image interactive contrast enhancement, at present by intensity windowing, becomes unacceptably time-consuming. Furthermore, windowing has disadvantages of being non-reproducible and providing adequate contrast primarily in selected image regions. The method of adaptive histogram equalization (ahe) appears to provide a solution to these problems. It is reproducible, automatic, and simultaneously provides contrast in all image regions. After summarizing the basic method, this paper will 1) describe a new contrast limited form of ahe that appears to allow its application to a wide variety of medical images, 2) present a VLSI machine design that will allow the calculation of ahe in a fraction of a second per megapixel, and 3) report the results of a study demonstrating that for chest CT images, ahe provides no measurable loss of diagnostic performance compared to the now standard windowing.

For obvious reasons people in the field claim that measured data do not agree with what they perceive. Scientists reply by saying that their data are "true". Are they? Since images are made to be looked at, a request for data meaningful for what is perceived, is not foolish. We show that, when noise is characterized by standard density fluctuation figures, a good correlation with noise perception by the naked eye on a large size radiograph is obtained in applying microdensitometric scanning with a 400 micron aperture. For other viewing conditions the aperture size has to be adapted.

Chest radiographs recorded on DuPont WDR (wide dynamic range) film are used as input for digitization, image processing and display by an HCM graphic arts scanner interfaced to a mini-computer. The imaging chain is evaluated in terms of the MTFs of its components and its response to film density. Some advantages of using a very wide-latitude, constant-gamma film are discussed. A subjectively preferred unsharp masking algorithm is described and clinical examples are shown.

The motion of the contracting heart has made it impossible to study coronary arteries with venous injections of contrast using digital subtraction angiography (DSA), even with cardiac gating. Furthermore, for intravenous injections, the images are statistically very poor due to the small size of the vessels and dilution of the contrast media, rendering the images diagnostically useless. A technique based on polynomial transformations has been implemented and is being evaluated which permits removal of motion between a pair of images acquired during mask-mode DSA. This technique is capable of handling three-dimensional motion and is based on applications of techniques of fluid dynamics, statistical sampling theory, and artificial intelligence. A series of phantom images, exhibiting three dimensional motion, are shown which demonstrate the ability of the technique to remove motion.

The degree to which beam hardening due to iodine, and x-ray and light scatter within the imaging chain, cause a nonlinear videodensitometric response of DSA grey level to iodine areal density were studied. Beam hardening effects were found to be secondary to the effects of scatter. Local scatter estimates could reliably be made, using an array of small lead beam-stops, which provided estimates of the scatter contribution regionally within an image. When the local scatter levels were subtracted from component images prior to DSA logarithmic subtraction, the DSA grey-scale response was made approximately linear with iodine areal density. Convolution filtering was studied as a method for estimating the scatter over an entire image based on beam-stop measurements made at a few points within the image. Simple, large-size (.0100x100 pixels) convolution kernels were capable of reproducing the scatter distribution within a chest image with an rms percentage error of <10% over a 36 cm field of view. The method should be implementable using a single lead beam-stop placed over a highly transmissive region in a patient.

Image processing algorithms and photographic techniques have been developed to allow objective, reproducible quantification of facial skin wrinkles, age spots, pores, and other visible skin features. The methods have been used to determine the effects of environmen-tal solar exposure on facial skin aging.

In order to design improved spatial compounding ultrasound scanning systems it is necessary to determine the correlation of speckle patterns as a function of aperture translation. We have conducted experiments measuring the speckle correlation as a function of lateral aperture translation for linear phased array pulse-echo ultrasonic imaging systems. Results are presented for variable frequency, range, transducer length, focus error, and reflecting material. Tests were conducted on two commercially available and one research imaging system. The measured rates of correlation coefficient decrease are independent of frequency, reflecting material, and target range when the target is in the focal zone. The experimentally determined correlations are used to derive the optimal spatial separation of images for speckle reduction, and the benefits of spatial compounding for a fixed aperture system discussed.

The use of subtraction radiography in dentistry is impeded by the necessity to couple physically the x-ray source, the patient and the film, in order to achieve a reproducible projection geometry. This need can he obviated by the ability to synthesize arbitrary projection images from a basis set of projections bearing a known geometric relationship to each other. Implementation of this method requires knowledge of the projection angle of the desired projection image relative to the basis set. This investigation explores the feasibility of using the gray-level standard deviations in corresponding subtraction images as similarity measures, in order to determine retrospectively the projection angle of a radiograph of interest with respect to the set of basis projections. An iterative coordinate estimation procedure is developed incorporating this technique, and its accuracy is evaluated using radiographs obtained from dry skull specimens.

Tomographic reconstruction using limited projection data requires the incorporation of all available a priori information in the reconstruction process. The difficulty in using much of the available information, including the projection data, is the lack of a mathematical model for inaccurate or vague data. The use of the fuzzy set theory gives the flexibility to model and incorporate such information in the reconstruction process.

In a digital diagnostic imaging department, the majority of operations for handling and processing of images can be grouped into a small set of basic operations, such as image data buffering and storage, image processing and analysis, image display, image data transmission and image data compression. These operations occur in almost all nodes of the diagnostic imaging communications network of the department. An image processor architecture was developed in which each of these functions has been mapped into hardware and software modules. The modular approach has advantages in terms of economics, service, expandability and upgradeability. The architectural design is based on the principles of hierarchical functionality, distributed and parallel processing and aims at real time response. Parallel processing and real time response is facilitated in part by a dual bus system: a VME control bus and a high speed image data bus, consisting of 8 independent parallel 16-bit busses, capable of handling combined up to 144 MBytes/sec. The presented image processor is versatile enough to meet the video rate processing needs of digital subtraction angiography, the large pixel matrix processing requirements of static projection radiography, or the broad range of manipulation and display needs of a multi-modality diagnostic work station. Several hardware modules are described in detail. For illustrating the capabilities of the image processor, processed 2000 x 2000 pixel computed radiographs are shown and estimated computation times for executing the processing opera-tions are presented.

We have developed a double-square-box region-of-search algorithm for semi-automated tracking of the vascular tree. Starting from a user-specified point, the algorithm accurately and automatically tracks the connected vessels in the entire vascular tree. In this study, we used the tracking algorithm in conjunction with the iterative deconvolution technique to obtain accurate vessel sizes and contrasts along the vessels being tracked. The tracking algorithm used the vessel size information to generate square-box regions of search for subsequent tracking points. Both the vessel size and the vessel contrast were monitored to determine the point at which tracking of a particular vessel would be terminated. The vessel sizes and contrasts along the vascular tree will be useful for further quantitative analysis of the vessels. At present, the algorithm accurately tracks the vascular tree into regions in which the ratio between the vessel contrast and the root mean square (RMS) noise is approximately 3 for vessel sizes of 0.5 mm or more.

Although the need to quantify the anatomical dimensions of coronary arterial atherosclerotic lesions is well recognized, standard analysis using 35 mm cine film precludes this technology during interventional procedures. The advent of digital radiographic (DR) imaging has caused speculation as to the applicability of electronic imaging to quantitative coronary arteriography. We report here the results of an experimental comparison of the accuracy of automatically determined geometric diameter (GD) and videodensitometric relative cross-sectional area (RA) using a radiographic arterial phantom. The phantom consists of a plexiglass plate with seven arterial models with "lesions" which vary from 0.5 mm to 3.0 mm diameter and proximal/distal lumena varying from 1.0 to 5.0 mm diameter. DR images at 512 x 512 resolution were preprocessed to correct for pincushion distortion and then analyzed using a quantification program which uses automatic edge detection to calculate GD and RA. Regression analysis of actual vs measured GD and RA yields excellent results. GD measurements show an overestimation of lumena with diameters less than 1.0 mm. RA, however, is not affected by this problem and yields accurate estimation down to the the 1% range. Geometric diameter percent stenosis as a result is overestimated in very small lumena, but RA percent stenosis yields accurate results over the range from 20% to 99%. As the program runs in less than 30 seconds, we conclude that DR imaging enables immediate quantitative coronary arteriography.

A scheme is proposed for discriminating between patients with Alzheimer's disease and age-matched normals using regional cerebral blood flow data measured by the noninvasive Xe-133 inhalation technique. Regional blood flows which are known to be decreased in Alzheimer's disease but are known to overlap to some degree with flows from normals are usually interpreted subjectively. In the scheme presented these flow values are used to form cumulative distributions for each subject group and for each detector. Pairs of distributions for homologous detectors are compared and the blood flow value at which this difference is the greatest is identified as the cutoff value. The 32 pairs of distributions give rise to 32 cutoff values. For each individual and detector the flow values are compared to their respective cutoffs. Those flow values which exceed their cutoff are assigned a 1 and those that are less than the cutoff a 0. For each subject and hemisphere these binary values are cummed where the sums for each hemisphere range in value from 0 to 16. A new cutoff for this sum is chosen and the sums for each patient are compared to this value. In the two groups sited the cutoff was set at a sum of 12 for each hemisphere. The majority of the normals had hemispheric sums of 14 or greater. The patients with Alzheimer's disease had sums that were equally distributed over the whole range of possible sums. This result indicated that the classification scheme was unlikely to classify a normal as an abnormal. However, there was a significant likelihood that an abnormal could be classified as a normal. These two qualities are defined as the sensitivity and specificity respectively. The test was sensitive (90%) but less specific (70%). The results of this classification scheme compared favorably with the subjective interpretation of experienced readers.

We have conducted a study to assess the effects of digitization and unsharp-mask filtering on the ability of observers to detect subtle microcalcifications in mammograms. Thirty-two conventional screen-film mammograms were selected from patient files by two experienced mammographers. Twelve of the mammograms contained a suspicious cluster of microcalcifications in patients who subsequently underwent biopsy. Twenty of the mammograms were normal cases which were initially interpreted as being free of clustered microcalcifications and did not demonstrate such on careful review. The mammograms were digitized with a high-quality Fuji image processing/simulation system. The system consists of two drum scanners with which an original radiograph can be digitized, processed by a minicomputer, and reconstituted on film. In this study, we employed a sampling aperture of 0.1 mm X 0.1 mm and a sampling distance of 0.1 mm. The density range from 0.2 to 2.75 was digitized to 1024 grey levels per pixel. The digitized images were printed on a single emulsion film with a display aperture having the same size as the sampling aperture. The system was carefully calibrated so that the density and contrast of a digitized image were closely matched to those of the original radiograph. Initially, we evaluated the effects of the weighting factor and the mask size of a unsharp-mask filter on the appearance of mammograms for various types of breasts. Subjective visual comparisons suggested that a mask size of 91 X 91 pixels (9.1 mm X 9.1 mm) enhances the visibility of microcalcifications without excessively increasing the high-frequency noise. Further, a density-dependent weighting factor that increases linearly from 1.5 to 3.0 in the density range of 0.2 to 2.5 enhances the contrast of microcalcifications without introducing many potentially confusing artifacts in the low-density areas. An unsharp-mask filter with these parameters was used to process the digitized mammograms. We conducted observer performance experiments to evaluate the detectability of micro-calcifications in three sets of mammograms: the original film images, unprocessed digitized images, and unsharp-masked images. Each set included the same 20 normal cases and 12 abnormal cases. A total of 5 board-certified radiologists and 4 senior radiology residents participated as observers. In the first experiment, the detectability of microcalcifications was measured for the original, unprocessed digitized, and unsharp-masked images. Each observer read all 96 films in one session with the cases arranged in a different random order. A maximum of 15 seconds was allowed to read each image. To facilitate receiver operating character-istic (ROC) analysis, each observer ranked his/her observation regarding the presence or absence of a cluster of 3 or more microcalcifications on a 5-point confidence rating scale (1=definitely no microcalcifications, 2=probably no microcalcifications; 3=microcalcifi-cations possibly present; 4=microcalcifications probably present; 5=microcalcifications definitely present). The observer identified the location of the suspected microcalci-fications when the confidence rating was 2 or greater. In the second experiment, we evaluated whether reading the unsharp-masked image and the unprocessed digitized image side by side for each case would reduce false-positive detection rates for microcalcifications and thus improve overall performance. The observer was again allowed a maximum of 15 seconds to read each pair of images and was instructed to use the unsharp-masked image for primary reading and the unprocessed digitized image for reference. The experimental setting and procedures were otherwise the same as those for the first experiment.

To enhance our capabilities for observing fluorescent stains in biological systems, we are developing a low cost imaging system based around an IBM AT microcomputer and a commercial image capture board compatible with a standard RS-170 format video camera. The image is digitized in real time with 256 grey levels, while being displayed and also stored in memory. The software allows for interactive processing of the data, such as histogram equalization or pseudocolor enhancement of the display. The entire image, or a quadrant thereof, can be averaged over time to improve the signal to noise ratio. Images may be stored to disk for later use or comparison. The camera may be selected for better response in the UV or near IR. Combined with signal averaging, this increases the sensitivity relative to that of the human eye, while still allowing for the fluorescence distribution on either the surface or internal cytoskeletal structure to be observed.

In recent years, a large number of applications have been developed for image processing systems in the area of biological imaging. We have already finished the development of a dedicated microcomputer-based image processing and analysis system for quantitative microscopy. The system's primary function has been to facilitate and ultimately automate quantitative image analysis tasks such as the measurement of cellular DNA contents. We have recognized from this development experience, and interaction with system users, biologists and technicians, that the increasingly widespread use of image processing systems, and the development and application of new techniques for utilizing the capabilities of such systems, would generate a need for some kind of inexpensive general purpose image acquisition and processing system specially tailored for the needs of the medical community. We are currently engaged in the development and testing of hardware and software for a fairly high-performance image processing computer system based on a popular personal computer. In this paper, we describe the design and development of this system. Biological image processing computer systems have now reached a level of hardware and software refinement where they could become convenient image analysis tools for biologists. The development of a general purpose image processing system for quantitative image analysis that is inexpensive, flexible, and easy-to-use represents a significant step towards making the microscopic digital image processing techniques more widely applicable not only in a research environment as a biologist's workstation, but also in clinical environments as a diagnostic tool.

Based upon a twenty year experience of inquiry into, and clinical application of, imaging technology - the current fluoroscopic environment is described. This realistic appraisal identifies key issues and problems which must be addressed so that the patient and health care system can take maximum benefit from the emerging "digital" technology. A gradual, logical implementation is preferred to the alternatives which may be either no action at all, or - a further explosion of powerful, untried, expensive technology.

We have developed a high resolution digital subtraction angiography (DSA) system which has a 1024 x 1024 pixel matrix incorporating a high resolution TV camera. The resolution of a DSA system is usually restricted by a TV camera so far. A diode-operating implegnated-cathode SATICON (DIS) imaging tube was selected and the TV camera was operated in slow scan mode with 1125 scan line numbers to improve the resolution of the TV camera. This system was free from degradation in resolution due to 512 X 512 pixel matrix. Diagnostic accuracy was improved by using the high resolution DSA system.

We discuss the applicability of intensified x-ray television systems for general digital radiography and the requirements necessary for physician acceptance. Television systems for videofluorography when limited to conventional fluoroscopic exposure rates (25uR/s to x-ray intensifier), with particular application to the gastro-intestinal system, all suffer from three problems which tend to degrade the image: (a) lack of resolution, (b) noise, and (c) patient movement. The system to be described in this paper addresses each of these problems. Resolution is that provided by the use of a 1024 x 1024 pixel frame store combined with a 1024 line video camera and a 10"/6" x-ray image intensifier. Problems of noise and sensitivity to patient movement are overcome by using a short but intense burst of radiation to produce the latent image, which is then read off the video camera in a progressive fashion and placed in the digital store. Hard copy is produced from a high resolution multiformat camera, or a high resolution digital laser camera. It is intended that this PPR system will replace the 100mm spot film camera in present use, and will provide information in digital form for further processing and eventual digital archiving.

A significant obstacle to the use of local area networks in Nuclear Medicine has been the high cost of computer systems capable of digitizing the analog outputs of conventional gamma cameras. A PC-based Nuclear Image Acquisition Station has been designed using readily available components that permits acquisition, display and transfer of nuclear images. Processing functions, including camera uniformity corrections, image rotation and edge enhancement and other operators Elre available locally. With appropriate file format manipulation, images may alternatively be transferred to a conventional Nuclear Medicine computer for processing and archival storage. Hardware and software costs required to implement these functions on an existing micro are less than $4000.

We have developed an innovative new technique which allows the construction of arbitrarily large mosaics of arrays of CCD detectors with the complete absence of gaps between the mosaic components. Further, because the elements of the mosaic are subject to parallel processing,very fast digitizing can be accomplished, even at high resolutions.

Medical images acquired and stored digitally continue to pose a major problem in the area of picture archiving and transmission. The need for accurate reproduction of such images, which constitute patient medical records, and the medico-legal problems of possible loss of information has led us to examine the suitability of data compression schemes for several different medical image modalities. We have examined both reversible coding and irreversible coding as methods of image for-matting and reproduction. In reversible coding we have tested run-length coding and arithmetic coding on image bit planes. In irreversible coding, we have studied transform coding, linear predictive coding, and block truncation coding and their effects on image quality versus compression ratio in several image modalities. In transform coding, we have applied discrete Fourier coding, discrete cosine coding, discrete sine transform, and Walsh-Hadamard transform to images in which a subset of the transformed coefficients were retained and quantized. In linear predictive coding, we used a fixed level quantizer. In the case of block truncation coding, the first and second moments were retained. Results of all types of irreversible coding for data compression were unsatisfactory in terms of reproduction of the original image. Run-length coding was useful on several bit planes of an image but not on others. Arithmetic coding was found to be completely reversible and resulted in up to 2 to 1 compression ratio.

This paper addresses the problem of data compression of medical imagery such as X-rays, Computer Tomography, Magnetic Resonance, Nuclear Medicine and Ultrasound. The Discrete Cosine Transform (DCT) has been extensively studied for image data compression, and good compression has been obtained without unduly sacrificing image quality. Vector Quantization has only recently been applied to image data compression, but shows promise of outperforming more traditional transform coding methods, especially at high compression. Vector Quantization is quite well suited for those applications where the images to be processed are very much alike, or can be grouped into a small number of classifications. These and similar studies continue to suffer from the lack of a uniformly agreed upon measure of image quality. This is also exacerbated by the large variety of electronic displays and viewing conditions.

An analytic expression for the point spread function (PSF) of multiformat video cameras is presented. Experimental techniques are presented which provide a means of evaluating PSF parameters, enabling the characterization of a specific camera. Results indicate that multiformat cameras suffer from long range optical scattering effects, resulting in a low frequency drop in the MTF.

Image display systems have traditionally been grouped into two different classes of equipment. Simple frame buffers, long used to provide a means of refreshing a monitor, have rarely provided any real computational ability or choices for interfacing. Image processors, the workhorses of image analysis, have had to share memory access with hardware for refreshing a monitor. International Imaging Systems has developed a line of low cost image analysis workstations that offload the monitor refreshing responsibilities from a remote image processor to a local frame buffer and allow the user access to interactive image display and analysis tools. The MIDAS (Monochromatic Image Display and Analysis Station) and IVAS (Image Viewing and Analysis Station) devices provide 1024x1024 pixel frame buffers, embedded MC68000 controllers, graphics controllers with overlay memory, look-up tables for radiometric transformations, plus interfaces to a variety of interactive devices and external buses. The devices can be clustered locally to allow the transfer of imagery between stations, peripheral devices (such as parallel transfer drives), and image processors. External bus adaptors (eg ACR/NEMA) can be added to provide compatability with other manufacturer's equipment in an integrated systems environment.

During our design of a workstation for diagnostic radiology, and having studied other designs being produced either commercially or institutionally, it became evident that a number of questions existed regarding the number and physical arrangement of CT images which a radiologist likes to have during interpretation. We wished to answer the questions about number and physical arrangement, and at the same time, determine how much difference there was concerning these parameters among the subspecialty sections. We approached this problem by using different numbers and arrangements of CT scan images presented to the radiologist on film. This paper will discuss the findings of this study, and explain some of the reasons given by the radiologists for their answers. The implications for workstation designs will also be discussed.

The Clinical Radiology Imaging System (CRIS) developed in the UCLA Image Processing Laboratory (IPL) is designed to allow radiologists, clinicians, and technologists to review and manipulate radiological images in digital form on a multiple viewing station (MVS). The system is composed of a multiple viewing station located at a clinical site and a centralized computer system consisting of a VAX-11/750 minicomputer, a Gould/DeAnza IP8500 image processor, and optical and magnetic disk storage. A broadband network allows video and digital communication between the remote clinical site and the IPL research laboratory. The station allows the real-time presentation of six 512x512x8 bit images from any combination of CT, MRI, DF, ultrasound, and digitized radiographs. The user interacts with the system by way of menus, icons, and a trackball. The CRIS system has been implemented in the Pediatric Radiology Section of the UCLA Medical Center. This paper describes the hardware and software architecture of the system and some early clinical experience.

We compared the diagnostic decision making accuracy of readers using a digital viewing station (DVS) with their accuracy using a conventional film alternator. The measure of accuracy was Az, the area under the ROC (Receiver Operating Characteristic). Five readers viewed 100 images of which 50 contained exactly one simulated nodule randomly located in the lungfield of a lung-chest phantom, and none of the other 50 contained any nodule. Each of the 100 images was seen in random order in each of three conditions: Conventional film (FLM), low-resolution DVS (LRD) 512 x 512 pixels, and high-resolution DVS (HRD) 1024 x 1024 pixels. ROC tests were based on a 6-point rating scale. Analysis of variance showed that there was no effect of readers on A, and a highly significant effect ofz viewing condition on A . Orthogonal comparisons showed that mean Azs of both FLM and HRD (1024) were significantly greater than mean A of LRD (512). However, mean Azs of film and HRD (1024) were not statistically different, a result which has important implications for the designing of diagnostic viewing facilities.

We have developed a prototype display console with which we can investigate issues related to the display of medical images. Our system is not a clinical tool, but rather a research tool that allows us to study how radiologists carry out the image reading procedure. For this study we used chest CT scans which are made up of 22 to 34 scan slices. We were able to simultaneously display up to 32 scan slices at two different intensity windows, i.e. 64 images on one screen. We used two more screens as working screens where the radiologist could display four or sixteen images per screen. Our goal was to determine how the radiologists use the mutlti-image, multiple screen format. The results of our study showed that the radiologists use different strategies in viewing images, and that a display system must allow for those variations. The display of a complete image set, even though the images were shown at much less than full-sampling, was very useful as an overview of the patient study. Simultaneous display of images at different intensity windows was also very useful. The slow times (20 secs) for access to images was a drawback of our system.

The Imatron C-100 Cine-CT TM scanner is a multi-slice high speed Computed Tomography (CT) scanner that produces a pair of anatomically contiguous slices in 50 milliseconds. The scanner operates in several modes. In flow mode, the scanner images up to 8 anatomically contiguous slices in 224 milliseconds without moving the patient. In cine mode, the scanner acquires data at a rate of 34 images/second. In both of these modes, a typical run generates 80 images in just a few seconds. Most patient studies involve one or more cine runs and one or more flow runs. Thus, the C-100 often produces an order of magnitude more images per patient than any other CT scanner. The large amount of data involved in each study requires rapid, easy to use analysis software and efficient data archiving. The C-100 achieves fast scan times by eliminating all moving parts. It generates a moving x-ray fan by scanning a highly focused electron beam along semi-circular tungsten targets that partially surround the patient. The scanner acquires data with a solid-state detector system, converts it to digital form, and sends it via fiber optic cables to a 32 Mbyte dual-ported high speed bulk memory. An array processor and back-projector reconstruct the images, which are stored on a dual-ported 1.37 Gbyte hard disk system. The scanner incorporates two workstations, each containing its own graphic display system. The work-stations communicate with each other through the dual-ported disks. The system stores images for long-term archive on magnetic tape, multi-format film, videotape, or removable optical disks. The C-100 provides interactive image analysis software that includes cine display, func-tional imaging, time-density analysis for flow measurements, off-axis reformatting, cardiac wall motion analysis, and image subtraction. Data management software includes file selection, merging, deletion, archiving, and retrieval.

Two ways of displaying an object represented in linear octtree form are presented. The object can have inherited attributes (such as gray-level, color or texture), each of which is represented by a field in the corresponding linear octtree node. The first method evaluates the surface of the object and, optionally, also all the attribute-to-attribute interfaces; a special field (BLOCKBIT) identifying the presence or absence of a node of the same size and/or attribute in each principal direction, enables us to solve, in most cases, the visibility problem so that only those octant-faces which are finally visible are indeed painted on the screen. The second method displays the object directly node by node, so that three faces are always displayed in a back-to-front order. The first method enables one to display the external surface as well as all attribute-to-attribute interfaces by applying the border algorithm only once; the second method enables one to display an object from external as well as internal view points, thus simulating a clipping capability.

Tomographic studies are used in a comparative and complementary fashion to show bone and soft tissue anatomy (CT, MR) and disease as well as physiological activity (PET, SPECT). The problem is to take several sets of imaging data, which are inherently tridimensional and correlate the information provided. The usual approach is for the physician to mentally correlate the information provided by a set of two dimensional images. Different scaling factors, orien-tation of the acquired images and patient positioning may difficult the accurate identification of the same regions of interest, specially for critical applications such as radiation therapy or surgery planning. In this paper we analyze a method to correlate sets of tomographic data based on geometrical transformations between corresponding surfaces defined in each imaging modality. The outlines of structures defined in one modality can then be viewed in the context of another or a common plane of view can be defined for the different structures for simultaneous viewing. Also 3-dimensional depictions of surfaces defined in various studies can be viewed in a composite image.

It is generally believed that medical 3D imaging augmented by computer graphics is useful in pre-operative surgical planning. Most of the existing hardware and software systems cater to only the requirements of visualizing 3D structures. This is perhaps adequate to meet the needs of pre-operative examination. However, simulation of complex osteotomy and corrective reconstruction procedures requires advanced tools for manipulating and quantitating 3D structures. This paper examines, with examples from craniofacial surgery, what new tools are needed, and outlines approaches to realize some of them.

Although existing 3D displays of CT and MR data aid in visualizing multiple slices of data, commercial systems lack the ability to illustrate changes in the solid displays due to proposed surgical procedures. Such realistic preoperative planning and the design of custom implants would decrease time in surgery, and thus reduce the cost and medical risk to the patient. The CEMAX-1000, a medical imaging console which generates 3D shaded surface displays from directed contours, features the ability to edit these contours to illustrate changes in soft tissue or bone images. A 2D slice edit facilitates modifications in the surface of the displayed object by allowing the surgeon to alter the shape of existing contours or to specify new contours. Working with the 3D display the surgeon defines new objects as subsets of the original and then manipulates the objects to show effects of relative movement proposed in surgery. This 3D edit capability operates interactively on the solid image, after which a new set of contours can be computed and stored in the object oriented data base. The editing capabilities of the CEMAX-1000, features of the data organization relevant to editing, and clinical applications of these features will be presented in this paper.

Biomedical workstations of the future will incorporate three-dimensional interactive capabilities which provide real-time response to most common operator requests. Such systems will find application in many areas of medicine including clinical diagnosis, surgical and radiation therapy planning, biomedical research based on functional imaging, and medical education. This paper considers the requirements of these future systems in terms of image quality, performance, and the interactive environment, and examines the relationship of workstation capabilities to specific medical applications. We describe a prototype physician's workstation that we have designed and built to meet many of these requirements (using conventional graphics technology in conjunction with a custom real-time 3-D processor), and give an account of the remaining issues and challenges that future designers of such systems will have to address.

The growing digital nature of radiology images led to a recognition that compatibility of communication between imaging, display and data storage devices of different modalities and different manufacturers is necessary. The ACR-NEMA Digital Imaging and Communications Standard Committee was formed to develop a communications standard for radiological images. This standard includes the overall structure of a communication message and the protocols for bi-directional communication using end-to-end connections. The evolution and rationale of the ACR-NEMA Digital Imaging and Communication Standard are described. An overview is provided and sane practical implementation considerations are discussed. PACS will became reality only if the medical community accepts and implements the ACR-NEMA Standard.

The ACR-NEMA Standard provides mechanisms by which a considerable amount of non-image information can be transmitted along with an image. Much of this information is similar to data currently managed by radiology information systems. However, the Standard does not provide methods by which the accuracy and consistency of these data can be maintained at a level that makes them usable for such important functions as billing or statistical analysis. A strategy by which extended versions of existing radiology information systems can be implemented with an ACR-NEMA-compatible PACS to provide data accuracy and consistency is discussed. The Standard also provides a mechanism to transmit the location and orientation of planar cross-sectional images. This will simplify obtaining measurements such as those necessary for stereotoxic techniques in CT, MRI or PET. However, the mechanism is not adequate to deal with the geometrical complexities of projection images. This creates an awkward situation for localizer images in CT and MRI. Extending the Standard to deal with projection images would not only remove this defect but also encourage the development of new quantitative applications in digital radiography.

ASR-NEMA standards will play a pivotal role in the difficult transition between today's imaging department and the completely PACS-based department of the future. This paper explores the practical implications of ACR-NEMA standards in PACS systems by focusing on the critical issues of compatibility, evolution, flexibility, and practical implementation. It is shown how a typical radiology department can develop a viable PACS strategy for improving its effectiveness through use of ACR-NEMA standards.

The ACR-NEMA Digital Imaging and Communications Standard specifies layered protocols for exchange of images and commands between two devices. The protocol specifies that the devices communicate as peers. To connect imaging devices to an open network, network interface equipment must support the protocols from physical to transport layers. These protocols have been implemented in network interface units. The implementation of specific protocol requirements will be described, including the virtual channels and end-to-end connection service. Integration with a network protocol will be discussed, and early performance results will be presented.

The ACR-NEMA Digital Imaging and Communications Standards Committee has published its Standard for an interface to medical imaging equipment. Various groups are now working on implementing this Standard, and the need for an exchange medium other than the electrical one of the Standard was forseen as necessary for software development. Working Group V was formed to examine this issue, and proposed to work first on a magnetic tape standard as magnetic tape drives are present on many imaging devices. The message format of the Standard was felt to be readily adaptable to magnetic tape. The physical specification has been chosen to conform to the ANSI standards for unrecorded and recorded magnetic tape, The logical specification is being developed with an attempt to adhere to the ANSI Standard for Magnetic Tape Labels and File Structure (1). This paper will discuss the reasons for developing a new format instead of using existing image interchange standards, explain the need for, and problems raised by tape directories, and explain the tape Standard as it currently exists.

A high speed fiber optic network for the transmission of digital images has been under development for the last three years at our Hospital. This network utilizes a ring architecture with token passing contention handling. Radiographs are digitized with a high resolution camera. Images can be viewed at either high or low resolution. The software for this four node Medical Image Management System (MIMS) is now complete and is undergoing trial runs. Clinical tests begin on March 1, 1986. This paper will focus on the philosophy, evolution and the present state of the interfaces that exist between the system and the physician. Care has been taken to develop an interface that is fast, powerful and error-free. Though simple to use, it presents the physician with a number of powerful options to manipulate the image to facilitate effective interpretation. An effort has been made to incorporate those functions that are useful to the physician. We tried to avoid cluttering the user menu with an array of less-used options.

Picture archiving and communication systems (PACS) have so far been studied primarily as a tool to address the problems of electronic radiology. Certainly the trends toward digital imaging within radiology provide strong ecomomic and medical incentives for the development of medical picture networks, but we believe that the applications for picture networks extend far beyond the field of radiology. Architects, engineers, biologists - practicioners of virtually every academic and commercial pursuit - deal with picture information every day, and increasingly these pictures are finding their way into digital form. We believe that industries and universities of the future will utilize sophisticated workstations serving a variety of scientific and commercial needs, and that these workstations will be linked by wideband networks capable of supporting not only text but high-resolution picture transmission as well.

This paper provides a short overview of the project experience obtained with image storage on optical disk. Further details will be given elsewhere. On the basis of the considerations mentioned at the beginning of the Paper, however, it can already be concluded that such an image storage system will play an important role in future routine work. Storage and communication tasks within the radiological department will be supported effectively by such a system. However, for a long time to come, communication with referral departments without the use of film as the information carrier will remain a prospect only.

The construction of a medical image communication network among hospitals in the Lansing, Michigan area demonstrates new opportunities for expanded relationships and services among hospitals, out patient clinics and associated professional groups. Routine clinical operation of the network commenced in May 1985. Three municipal CATV systems provide the data communication channels for rapid transfer of digital images to remote locations for viewing and archiving. Currently, reconstructed compressed computed tomography images are automatically transmitted in a few seconds from community hospitals to a centralized interpretation service. Broadband network equipment was designed and assembled that interconnects commercial CT scanners and display systems. The use of leased CATV bandwidth (3 MHz) is key to the economic and operational feasibility of the project. Both operating and equipment costs will be presented. Fluctuations in cable signal amplitudes and noise content, and a variety data error patterns occur on such a shared channel resource having thousands of attached users; this requires careful design attention. Various channel characteristics and performance experience with such a complex cable network will be reviewed. The clinical users have adapted surprisingly well to the new patterns of activity and communications.

"Help me communicate more quickly and more effectively with referring clinicians". This request was the driving force behind the installation of the AT&T CommView System at Duke. The CommView System is a type of Digital Image Management System and Picture Archival Communication System whose chief purpose is to deliver interpreted diagnostic images to referring clinicians and attending physicians. The system acquires electronic images from modalities in a diagnostic imaging facility, stores these images in computer managed patient files and distributes these on demand over fiber optic cable to Display Consoles. The CommView System was designed at AT&T Bell Labs; it uses fiber optic ribbon cable between buildings fused to multistrand lightguide building cables to distribute images, typically around a medical center or campus at data transfer rates of 40 Mbps. This paper gives the rationale used in designing a start-up network and placing the initial equipment for a field trial of the AT&T CommView System in the Radiology Department of Duke University Medical Center.

The communications network is the backbone of a Picture Archiving and Communications System (PACS). A working communications network also facilitates research into the different aspects of a prototype PACS. In our previous PACS communications paper [Thompson82], we outlined the design criteria for a medical communications system. These same criteria are in effect except that we have replaced the analog transmission of images by digital transmission. To meet the communications needs of our prototype PACS configuration, the University of North Carolina (UNC) at Chapel Hill chose to develop a rudimentary implementation of the ACR-NEMA standard to allow communication between acquisition devices, host computers, consoles and other medical imaging equipment. This paper reflects on the decision to use digital transmission of images and existing digital modalities. We describe the implementation being carried out at UNC, and how the communications network interrelates with the other ongoing UNC PACS research efforts.

For a PACS to successfully replace film, accuracy and speed of diagnoses made with a PACS must equal or exceed those made with the traditional film system. Therefore, proper attention to human factors is critical in designing the means by which information is selected by and displayed to the user, in providing good image quality and response time, and in physically configuring the system for user comfort and rapid maintenance. This paper discusses human factors requirements for PACS in general, and some of the particular decisions made in planning the current CommView' system and the future evolution of the system.

The strong relationship between a Picture Archiving System (PACS) and the Radiology Information System (RIS) is clear to all. Some general considerations about the structure of an automated linkage between RIS and PACS and its initial implementation are discussed in the following chapters.

This paper will discuss the functional requirements for interfacing Picture Archiving and Communications Systems (PACS) to today's radiology information systems (RIS). It is clear from the operational modeling study done at UNC, that a fully functional PACS will require the use of RIS data. For example, patient ID, scheduling, and information tracking are usually part of a RIS and must be available to the PACS. Similarly, the image tracking and index capabilities of the PACS will be used by the RIS. As part of our modeling data flow analysis research program, we have identified a number of these interactions. The interface points will be identified and discussed with emphasis on the operational impact of such an interface and efficiencies likely to be provided.

The point-to-point fiber optic links have been proven to have better performance than the copper based media for long haul telephony application. Data rates in the gigabits/sec range and repeaterless operations over several hundred kilometers have been frequently reported. However, their use in a bus topology is severely limited by high tapping and splitting losses of optical power. This paper will review the fundamental limitations of the fiber optic links in the LAN environment and address the criteria to select a suitable topology for a fiber optic based LAN. It will also discuss how the shortcomings of the bus topology can be circumvented in the centralized star-like wiring topology. This centralized wiring topology is sufficiently flexible and well suited to design an architecture for the fiber optic based LAN.

Software development is a crucial issue for any PACS implementation. A significant percentage of development time, development cost, and engineering complexity will be committed to the software components. Product performance and user acceptance are highly software dependent. The required software development effort must be accurately estimated for proper resource allocation. Common sources of error in judging effort requirements are underestimating software complexity, device controller firmware problems, personnel education, system documentation inadequacies, and difficulties in implementing resilient error handling. Four person-years were required for the network application and imaging device interface software for network node computers designed and implemented by the authors. As in any complex engineering project, a detailed functional description is a necessary starting point. From the functional description, a formal software design method should be followed to produce a written architectural and data structures definition. With the advent of fast 32 bit microprocessors and continually falling memory costs, there is no reason to consider programming in assembly language. The increased programmer productivity and self documenting qualities of high-level languages surely outweigh the penalties in speed and memory size. Besides a high-level language, a complete set of software development tools such as editors, automatic program version control systems, library maintenance utilities, and document preparation utilities is a necessity. The UNIX operating system provides these facilities as may other possible candidates for development systems. As an example, the software design of a UNIX-based network node computer will be presented in detail, including descriptions of the comprising process modules, interprocess communication and synchronization, device interfacing, and memory requirements.

The architecture of a Picture Archiving and Communication System should be flexible, modular, and be easily upgradable to meet the ever growing communication needs of radiology information. The architecture of one such system being developed at AT&T Bell Laboratories is described in this paper. A fairly detailed presentation of the hardware as well as the software architecture is given. The architectural partitioning of the system into several subsystems to provide commonality among the various components of the product is described. The mapping of the common architectural subsystems across different components of the product is shown. The paper concludes with a description of the product evolution.

This paper seeks to describe work in progress towards the development of a very-high-speed (80 - 200 Mbits/sec) Local Area Network (LAN) for the support of a distributed image processing system. The current practice in most medical, scientific and industrial image acquisition and processing equipment is to use complete, stand-alone image processors dedicated to a particular application. Such a system is not very expandable or flexible and is not cost-effective. A better approach would be to put together a distributed system from individual subsystems (digitizers, preprocessors, array processors, disks, high-resolution displays, etc.) and to link them all together with a high-capacity LAN, capable of the transfer of complete images at near-real-time rates. This would result in a flexible system, reconfigurable and upgradable by simply adding more powerful devices to the existing installation, and also ensure a standard, unified interface between different components. Some possible architectures for the network are discussed. Other factors affecting the design are also considered, based on typical medical imaging systems and needs. A parallel is drawn between the proposed system and the Digital Imaging Network for the Picture Archiving and Communications System (DIN/PACS) currently under development.

A prototype archiving system manufactured by the 3M Corporation has been in place at the University of North Carolina for approximately 12 months. The system was installed as a result of a collaboration between 3M and UNC, with 3M seeking testing of their system, and UNC realizing the need for an archiving system as an essential part of their PACS test-bed facilities. System hardware includes appropriate network and disk interface devices as well as media for both short and long term storage of images and their associated information. The system software includes those procedures necessary to communicate with the network interface elements(NIEs) as well as those procedures necessary to interpret the ACR-NEMA header blocks and to store the images. A subset of the total ACR-NEMA header is parsed and stored in a relational database system. The entire header is stored on disk with the completed study. Interactive programs have been developed that allow radiologists to easily retrieve information about the archived images and to send the full images to a viewing console. Initial experience with the system has consisted primarily of hardware and software debugging. Although the system is ACR-NEMA compatable, further objective and subjective assessments of system performance is awaiting the connection of compatable consoles and acquisition devices to the network.

One of the major bottlenecks in realizing a full size PACS is the image storage component. Further technological developments are necessary before sufficient capacity can be realized. It can be expected that a multi-layered storage structure will be necessary where large capacity with relatively long access times will be offered at the lowest level. The higher levels will show decreasing access times combined with decreasing capacity. Both in the judgement of images acquired in the diagnostic study and in the work of the clinician historical images of the patient concerned may be needed. If these images had to be retrieved from the lowest storage level at the moment they are needed, the service time would definitely be unacceptable. So a strategy is necessary for the anticipation of the need of access to images. In this paper a first attempt is made to identify relevant parameters to be used in such a look ahead algorithm for the activation of images to a next level. A simple algorithm is suggested; the need for further study is emphasized.

A prototype PACS system is being designed which will connect an existing nuclear medicine mini NM-PACS with an imaging facility in the medical school, a PET and experimental medical imaging facility, and research imaging facilities in the Radiology and Nuclear Medicine Department of the Free University in Brussels ( VUB, Jette), Belgium. The local nuclear medicine NM-PACS uses SOPHA (Informatek) equipment with parallel DMA connections. The PET and experimental medical imaging facility is based on Apollo workstations. The medical school imaging network connects a VAX 11/785, a VICOM and several other computers, and is physically located next to the hospital. A local radiology Rad-PACS will be based on an Ethernet link. The "global PACS" described in this paper will connect all image processing facilities. Initially, 4 Apollo workstations with a token passing ring network (12 Mbit/sec), a 500 Mbyte storage, and 1kx1k displays will be used. The following subparts of the "global PACS" are novel: a pictorial-menu driven user-interface software for image display, processing and management is being developed to facilitate the use of the PACS hardware . The image data base is distributed over the different areas of the hospital and university. The system will also provide the distributed computing power (parallel processing and acquisition of images) needed for a new HIDAC PET camera being constructed at the university. This PACS project is part of a large scale European effort (EuroPacs) to develop PACS technology.

PACS literature to date has emphasized the needs of diagnostic imaging; however, the ability to acquire, manipulate, and display data derived from multiple imaging modalities is also vital in the practice of radiation oncology and radiation therapy planning (RTP). Radiographic or scintigraphic images for RTP must include specific spatial calibration data, as well as data relating image acquisition to anatomic localization within the patient. The digital nature of PACS images and displays allows the radiation oncologist to interactively assist in evaluating whether or not near-by structures are tumor-free. The radiation oncologist may also need to review nonradiographic diagnostic images (e.g., endoscopic images or pathology tissue specimens). Finally, it must be possible to take data such as isodose lines and superimpose them onto images relating the proposed therapy field to patient anatomy. Not only would this be useful for the radiation oncologist, but it would also provide information currently not easily available to the diagnostician and useful in subsequent diagnostic efforts. The three-dimensional (volumetric) data creation for RTP is not currently widespread because of the difficulties in converting images into a coherent, reliable and registered data set; this is the unique contribution of PACS. Software must be developed to permit creation of volumetric models based on data derived from both planar images and various tomographic modalities, including calibration and localizaton data for accurate image registration and scaling. This will permit positive definition of tumor volume by diagnosticians and the radiation oncologists as an initial portion of the therapy planning process. As a part of the underlying data structure for such systems, there must be some uniformity of image format between modalities and vendors; this has been adequately addressed by the Digital Imaging and Communications Interface Standard recently adopted by the American College of Radiology and the National Electrical Manufacturers' Association (ACR-NEMA). In addition, such standardization efforts must also incorporate the necessary calibration and coordinate data. This paper will examine some of the unique requirements for PACS (and PACS workstations)optimized for RTP. The assumption is made here that these are not independent, self-sufficient devices; rather, they are subsystems of a PACS network, capable of sharing certain resources.

A prototype PACS is currently being developed for the Medical Imaging Department of our institution. The main functionalities required by the medical staff and the resulting system design are presented. Then textual and image data that are maintained by the system, storage organization and archiving strategy, as well as the user-interface to access data and functions are described. It allows the reader to understand currently available functions and potential use of the system in a research hospital environment.

We are developing an in-house Picture Archiving and Communication System (PACS) targeted to the centralized storage of images acquired from our computer-assisted imaging modalities and to the display of these images on multi-modality viewing stations. The central image database is distributed between two system modules with image processing capabilities located at the viewing stations. The first module of the image database is an image filing subsystem used for the storage and retrieval of complete image files. In our initial work, this module consists of a Data General MV/6000 computer system with 1.16 Gbytes of on-line disk storage. However, more efficient dedicated filing systems may be substituted for this general purpose computer in future revisions. The second module is an image file indexing subsystem which has been implemented on a DEC PDP-11/44 computer and is tightly integrated with our MARS II (ADAC Laboratories) Radiology Information System. These two image database modules communicate via low-speed, serial communications lines. This report focuses on our developmental work on the image file indexing subsystem and its communications protocol with the image filing subsystem. The image file indexing subsystem automatically inserts image file locators for studies referenced by patient when an image is transmitted from the acquisition device (eg, CT) to the image filing subsystem. It also locates image files at the request of a viewing station user based on patient name, physician name, or study type for either read or unread studies. Other capabilities include cross-referencing patient attributes with the MARS II system, deleting studies based on predefined criteria, storing requests for hard copy, storing user selected image processing attributes for individual images, flagging the archival status of an image, and complete managerial functions. This central-ized image filing and indexing system comprises what will become the central element of our PACS.

The growth of large databases always leads to problems of data storage. Applications with multigigabyte requirements such as pictorial or document storage are particularly difficult. A variety of technological solutions have been developed that rely on high density optical storage for rapid access to large volumes of data. A major frustration for those of us engaged in research on application-oriented computer systems has been the difficulty in obtaining and integrating these specialized mass storage devices. This paper details (1) the integration of a Hitachi 301 digital optical disk storage device into a VAX minicomputer system, (2) the software system to support the optical disk, (3) the performance of the hardware/software system, and (4) the integration of optical disk operation into an application in radiological imaging. The software system is designed under the VMS operating system in FORTRAN 77 to facilitate simple integration of optical disk functions into application packages.

A diagnostic radiology department can be viewed as an information processor. The inputs to this processor are orders for studies, clinical histories, and requests for consultation. These inputs are used in conjunction with the database of prior studies to produce the required outputs -- images and diagnostic reports. Automating information transport in this environment will reduce costs and speed up the flow of information, to the benefit of the hospital, radiologists, attending physicians, and patients. This paper describes a "model" radiology department in terms of its internal and external communications. Using this model we present requirements for a system to manage radiological image and text data - a Radiology Operations System.

Multiple benefits for PACS development can be expected to derive from careful assessment or modeling of current system operations. Our approach to modeling the radiology department operations at UNC-CH is based on two assumptions: 1) a department consists of multiple, operationally similar modules and 2) individual modules can be sufficiently characterized by use of a single template. Our procedures and conclusions in developing the modality portion of this template are described. Future modeling work is directed towards characterizing interactions within radiology and between radiology and other clinics which should, we believe, complete the characterization for a single module of the radiology department.

Crucial in the planning of a medical image archiving and communication system is the knowledge of the logistics of diagnostic images in and between departments. On the prerequisite that all requested images should be available at the referring department, knowing the average area (in cm2) of used film per kind of examination, as well as the necessary estimated spatial sampling frequency, an estimate is generated of the necessary intermediate storage buffers and the network capacity to the different departments.

Electronics storage systems, image display stations, protocols, and networks capable of managing a large amount of data must be developed in order for picture archival and communications systems to become a reality. These systems must be designed based on information about procedure volumes and information about the magnitude of data transfer rates needed at any given time. Since significant fluctuations can be expected with respect to the demands during a 24 hour period it is important to generate a traffic model that can predict these fluctuations. In addition to recognizing the demands at one institution, it should be noted that different institutions may have widely differing needs. Data has been collected from three differing radiological environments and an investigation with regards to traffic profile and fluctuations has been performed.

As Radiology becomes more invested in direct digital imaging techniques, the potential for moving these images throughout the department, interpreting them directly in digital mode and archiving them in computer form is a topic of high current interest. A fundamental consideration is the amount of digital data to be handled. Even the low and medium resolution images now handled in digital mode require immense amounts of digital storage space. The first quantification of the amount of digital data was by Dwyer, et al, in a report concerning the workload in a 614-bed hospital. Their assumptions and calculations are reviewed and applied to the workload data from a 1082-bed hospital. Storage requirements for PET and MRI workload are calculated, and an estimate of digital radiography data is presented. The digitization of plain film radiographs will virtually increase the storage requirements by a factor of 10.

Within the Dutch BAZIS cooperation of 25 major hospitals the IMAGIS (Image Information System) project was started in 1984 to realize a working PACS (Picture Archiving and Communication System) prototype in the next few years. Such a prototype should be clinically evaluated primarily at the Utrecht University Hospital. Simulation has been chosen as a major tool for the set-up and study of a future medical PACS in Dutch hospitals. A pilot study conducted by us thusfar has indicated the usefulness of simulation for our purposes, and clearly indicated the need for information on the present use of images within both the radiological department and the hospital as a whole. We use two ways to collect these data. The first way is through a rough (written) inquiry to all participating hospitals. This will immediately be followed by a more detailed questionnaire to the most involved participants of the BAZIS organization. The second way of collecting crucial data is through a thorough observation of the image handling processes at the Radiology Departments of the Leiden and Utrecht University Hospitals. As may be expected, there will be quite some differences between university (teaching) hospitals and general hospitals, which we will take into account for the Dutch situation. The simulation model that is being extended now will be based upon our inquiry and observation results. The model will include as much components as feasible. Among these will be control layers, interfaces to the existing HIS, a multi-level archive, various numbers and types of imaging workstations, connected by different networks for images, graphics and text, each with specific requirements concerning media, speed and topology. Our final aim is twofold: the study of imaging procedures in existing hospitals and hospitals under construction, and the simulation and the realization of a feasible complete IMAGIS, evolving from a number of successive prototyping iteration cycles.

Estimates have been made for the amount of digital storage capacity necessary and the data creation rates for the nuclear medicine portion of an all digital radiology department. After installation, the number of computer studies performed, image files created and bytes archived for each month increased with time. There was a high correspondence between the number of computer studies performed and (1) the number of new image files created (r=0.95) and (2) the total number of pytes archived (r=0.82). The average number of bytes created per work-day was 5.6 x 10⁶ Experience with our nuclear medicine system seems to indicate that when planning a picture archiving and communications system (PACS) installation (1) the amount of necessary online storage may increase and continue to increase after installation and (2) our estimate of nuclear medicine digital images creation rates agree with previous estimates l.

The nation's hospitals have long used distributed data processing as a means of reducing operational costs and providing timely service. Radiology Departments are now also taking advantage of these facilities in order to decrease the cost of producing and archiving radiological images. A typical medium scale hospital consumes large quantities of silver oxide film which, along with attendant labor costs, is expensive compared with costs for digital image processing technology now available. Using this technology large image files can be stored and retrieved through local area networks that can also support the transaction traffic essential in a hospital environment. The evolving systems are called Picture Archiving and Control Systems (PACS). PACS will include radiology imaging equipment, distributed and central image archiving facilities, and significant numbers of user work-stations and graphics display nodes. The devices will be interconnected by high speed local area networks capable of distributing information ranging from simple control messages to large image files of several megabytes in a fashion offering most users a response time of several seconds. This paper will illustrate the PACS system concept, present a queuing model approach to analyzing PACS performance, and discuss results acquired for a variety of parametric samples. The IBM Research Queuing Package (RESQ) has been used for the exercise and will be discussed sufficiently for the reader to appreciate its capability. RESQ simulation results indicate that system response times will be more dependent on the internal architecture and programs of the workstation than on the speed of the transmission media.

This paper presents a simulation of a distributed imaging system used for storing and retrieving digitally formatted medical images. The simulation model includes a one segment Ethernet local area network (LAN) as its communication system. This segment connects three image input devices, an image processor, an active storage device, an archival storage device, and one or more display devices (workstations). The simulated model is driven by a medical imaging workload similar to a hospital radiology department, and consists of a mix of user activities such as display, edit, manipulate and browse. These operations are performed interactively at a workstation. Performance measures such as response times, channel utilization and number of collisions were tabulated and are presented in graphs for different numbers of workstations connected to a system with different channel capacities and different image sizes. Conclusions are drawn with regard to necessary channel capacity and the viability of Ethernet for PACS.

The videodisc offers an excellent medium for handling and presenting visual information. It has attributes which make it applicable for situations in which video cassettes, 35mm slides, and microfilm were previously used. Its capabilities for random frame access and large image volume have spawned its utilization in, and the vigorous growth of, interactive video (IVD) and computer-assisted instruction (CAI). Until recently the creation of videodiscs was confined to production studios to which program materials had to be sent. Now available is a videodisc recorder (VDR) that can be used on site and has excellent quality.

We present here a concise survey of the major themes running through the contemporary literature on the assessment of diagnostic imaging systems. These include the concept of the ROC curve, the ideal observer, the ideal observer signal-to-noise ratio constructed from the laboratory measurements, and complications arising from the application of the ideal observer approach to clinical diagnostic systems and expert systems.

These transcripts were edited slightly for clarity. The presenters did not have an opportunity to review the text because of the press of publication. Although the verbatim transcripts are not accompanied by the slides and thus are somewhat difficult to follow, the thoughts of the individual presenters on many key issues warrant this publication.