The ideal observer signal-to-noise ratio has been derived from statistical decision theory for all of the major medical imaging modalities. This yields an absolute scale for image performance assessment and instrumentation design and optimization. Applications include: the functional dependence of detectable detail size on exposure or imaging time; a framework for comparing data acquisition techniques, e.g., Fourier methods vs reconstruction from projections in NMR imaging; calculations of realizable limits, e.g., the limiting gain of time-of-flight PET scanning. Measurements on human observers show that they can come close to ideal performance, except when the noise has negative correlations as in images reconstructed from projections. In this latter case they suffer a small but significant penalty.

The ideal observer for the task of object localization in the presence of correlated noise is implemented by means of the minimum chi-squared method, which is equivalent to the maximum likelihood method when the noise is additive and normally distributed. The prewhitening approach is explored in which the noise in the data is made uncorrelated by filtering the data with the filter S-1/2, where S is the noise power spectrum. The location of the object is then found by fitting a model of the object to the prewhitened data. A measure of the goodness of fit is proposed that is based upon the serial correlation in the prewhitened residuals, the difference between the data and the fit. These concepts are demonstrated by applying them to simulated data that possess known noise characteristics. It is shown that the accuracies predicted for the ideal observer can be achieved by the prewhitening technique.

Medical image production and display has reached the stage where decisions about what is in the image, and where, are limited solely by statistical noise. One would like to know how well humans use image information and what steps might be taken to improve decision performance. A number of experiments demonstrate that human decision-making for simple tasks is very close to the theoretical limit. Under best conditions, humans require about more SNR than the ideal bserver to operate at a given decision accuracy.

The problem of determining the inherent contrast resolution of an x-ray imaging device (CT or projection x-ray) has not thus far been satisfactorily solved. The contrast signal-to-noise ratio depends on the geometry and composition of the object being imaged, the x-ray beam spectrum and the detector spectral efficiency (signal) as well as a host of factors which contribute to the quantum and non-quantum statistical fluctuation in the image (noise). Because contrast depends on machine spectral factors as well as phantom atomic number differences, measurements made using any single low contrast phantom cannot yield the inherent contrast resolution but only the contrast resolution of the system for phantoms of that geometry and composition. We have overcome this problem by developing and using two phantoms to measure inherent contrast resolution: in one phantom the contrast-generating structures differ from the background material only in electron density, while in the second phantom the contrast is generated purely by a difference in effective atomic number of the materials. Measurements made using these phantoms allow the determination of the inherent contrast resolution of the device for objects of any composition. In this paper, descriptions of the construction and composition of the electron density and atomic number phantoms are given. Computer simulations and preliminary measurements of the inherent contrast resolution of clinical x-ray imaging systems are presented. From these, one can develop optimum methods of operating x-ray imaging devices.

We have been considering the question of calculating, conveniently but with precision, significant basic image quality indicators for mammography. As a result of this work, a number of questions have been addressed regarding both the philosophy and the practicability of such calculations. One way to make a prediction is to define a phantom, determine a primary photon fluence, obtain the scatter fluence from empirical correlations, define a target of interest and make a straightforward calculation. The other extreme is to define a phantom and target and do a complete Monte-Carlo calculation. The first is too approximate for making close calls between various system configurations. The second, although it can be very useful, tends to gather in one place the weaknesses of both experiments and theory. On one hand the Monte-Carlo calculations are unwieldy like experiments; on the other, like any theory they can only represent the physics that is included. It is always necessary to verify the calculation by experiments. For example, scatter radiation in mammography has a simple energy dependency but its angular distribution is quite complicated due to interference phenomena and also quite important. The first approach mentioned above would certainly miss the complication, as experiments have in the past. The Monte-Carlo calculations also miss the complication because, for practical reasons, their angular resolution is generally too low and in addition the appropriate physics is not included. We will describe our approach to the modelling of mammography. It is a combination of prediction techniques, based on variables chosen to be convenient for describing the physics, and experiments to verify uncertainties in the modelling. We will also discuss the point of greatest vulnerability of either calculations or experiments - the phantom. Our thesis is that all possible phantoms and targets can and should be treated in generalized combinations, which can be used to raise the level of usefulness of either experiments or predictions.

We use Digital Subtraction Radiography (DSR) as a vehicle to clarify, simplify and correct some contemporary notions of the physics and psychophysics of radiological image detection.

X-ray imaging utilizing broad x-ray spectra has been studied. The analysis suggests a generalized view of the concept of DQE in a way that makes the efficiency dependent on the imaging task. The conventially defined DQE is shown to be equivalent to the more general DQE task when the task is differentiatinebetween radiation levels without any differences between signal and background spectra. A formulation of the theory and some examples concerning intensifying screens are presented. For some screens the difference between imaging task dependent DQEtask and the conventional DQEN may be appreciable, larger than 30%, when imaging targets with strong energy dependence in their contrast, such as bone or iodine.

A microwave-induced thermoelastic tissue imaging system is suggested as a new and promising imaging modality. It appears to possess some unique features that may allow it to become as useful as these other methods and to permit non-invasive tissue imaging of tissue characteristics which are not identifiable by other techniques. It uses nonionizing radiation and relies on a beam of impinging microwave to launch an acoustic wavefront into tissue. There is a direct relation between the pattern of absorbed microwave pulses and the induced thermoelastic pressure waves in biological tissues. Moreover, regions of differing permittivity would exhibit differential absorption. This thermoelastic wave of pressure would propagate through the tissue and be detected by a two-dimensional array of piezoelectric transducers positioned on the body surface to give an image of the intervening tissue structure. Signals from the outputs of this transducer array are then amplified and band-limited at a signal conditioning stage. A computer-controlled data acquisition system samples and converts them to digital form for further processing. A hybrid parralel/serial design of dividing the array into segments and collecting data from each segment sequentially is used. Image processing algorithms are applied to digitized pictures for enhancing the images. The processed two-dimensional image is displayed on a color monitor. An example showing the image of a human hand model illustrates the potential usefulness of microwave-induced thermoelastic tissue imaging.

A digital camera system has been constructed for obtaining reflectance images of the fundus of the eye with monochromatic light. Images at wavelengths in the visible and near infrared regions of the spectrum are recorded by a charge-coupled device array and transferred to a computer. A variety of image processing operations are performed to restore the pictures, correct for distortions in the image formation process, and extract new and diagnostically useful information. The steps involved in calibrating the system to permit quantitative measurement of fundus reflectance are discussed. Three clinically important applications of such a quantitative system are addressed: the characterization of changes in the optic nerve arising from glaucoma, the diagnosis of choroidal melanoma through spectral signatures, and the early detection and improved management of diabetic retinopathy by measurement of retinal tissue oxygen saturation.

The IS-2000 permits image acquisition with a color TV camera, analysis in digital format, and archival storage in analog format on an optical disk. The IS-2000 was designed from a system model of a consulting ophthalmic practice. We describe the model and the resulting system, pointing out similarities to (and differences from) radiology.

Original techniques for real time pattern recognition and feature analysis from standard video signals have been developed. These techniques have been applied to the monitoring of eye movements and pupillary size during visual field and electrophysio-logical examinations in routine ophtalmological practice. The basic features of the resulting instrument are : 1- the use of low-cost hardware, i.e. standard video equipment and LSI circuitry. 2- the measurement of eye, orientation from the position of the bright pupil relative to the corneal reflection. 3- "real time" processing and high data throughout of 50 samples per second, allowing pupillary and oculomotor reflex analysis. 4- specialized hardware and software permitting an adjustment free feature identifica-tion and analysis directly from video signals. Severe perturbations of the ocular video images can be handled by the system, including partial occlusions of the pupil with eye lids or eye lashes, fluctuations of amplitude levels and parasite light reflections.

With today's increased focus on the cost effective delivery of health care, we are seeing great emphasis placed on the systems approach to cost containment. One area in which this approach has direct application is the collection, storage, retrieval and display of diagnostic information (both images and text). The evolution of relevant technologies has now advanced sufficiently to consider their applications in image storage, communication and display. This paper will discuss the state of technology and other considerations that must be examined in the design of an integrated Multi-Modality Imaging and Communications System (MMICS). Specifically, we will explore the information requirements, both image and text, the data storage and retrieval issues, the communications networking requirements, the image display issues and the need for an interface into the Hospital Information System (HIS). Another recent trend in medical imaging is a trend towards shared resourses. A group of hospitals and/or clinics come together to share a CT or MR facility. Or a group of hospitals, each of which has a CT with which to take data, share certain specialists who may be located at different member institutions. This resource sharing creates the need to connect hospital MMICS's together and perhaps also allow other users, such as referring physicians or clinics located outside of the hospital, to have access to parts of the system.

Current systems for the production of medical images and current development trends give a basis of experience for the design of a digital PACS including images and demographic data. Such a PACS must contain software and hardware concepts which permit the medical requirements, as presently understood, to be realized. As part of its research Siemens is designing and evaluating a hybrid network configuration which allows extensive flexibility and growth potential despite current limitations in available network bandwidth and storage capacity. As demand for digital data expands, additional installations can be added to the system. The modular concept permits incorporation of technological advances with minimal difficulty. The system allows different digital imaging modalities to communicate with a central data storage and processing system. Data display facilities both with and without manipulation capability are realized using high speed multi image storage devices. The human interface is designed to be ergono-metric, interactive, and user-friendly. Standardized, commercially available hardware has been included wherever possible to provide economical worldwide acceptance. Estimates of digital data per unit time under different conditions are presented and compared to the specifications of software and hardware elements both currently available and envisaged in the near future. Potential limitations of the design, as well as possible solutions incorporating expected technological developments, are discussed.

The system is comprised of the following: 1. A magnetic digital video disk recorder which can be used for applications in on-line digital imaging systems and off-line fast access image storage and retrieval buffers. 2. An optical disk recorder and a magnetic tape system for archiving purposes. 3. A display workstation which would enable user access to the images and data. The system design permits location of display station and the storage subsystems in a conventional standalone mode or a geographically dispersed configuration using a local area net-work.

This paper describes the Multimodality Acquisition and Review System, (MARS), as an operational digital radiographic image management system. The presentation emphasizes the unifying effect that a digital radiograph management system has on a radiological department and hospital.

A growing interest in the need for the consolidation of medical imaging information obtained from a wide variety of imaging systems has prompted the development of a number of system architectures designed to bring coherence to medical imaging efforts. One major stumbling block to the successful implementation of these systems is the inability to efficiently and effectively interface the various image acquisition devices to a unifying PACS network. In an effort to reduce this difficulty, a proposed Imaging Equipment Interface Standard has been developed which meets the following goals: 1. Applicability across the spectrum of image sources. 2. Absence of manufacturer specificity. 3. Simplicity and ease of implementation. 4. Functional design for future capability. The standard includes complete and specific details on all aspects required for implemen-tation within an imaging system including image format, electrical characteristics, interface protocol, and provisions for miscellaneous patient-specific data.

A model is developed for the overall DQE of an x-ray screen-film system, and this is related to the component DQE's of the screen and film. The model characterizes the screen in terms of its x-ray absorption efficiency, MTF, and the statistics associated with the conversion of absorbed x-rays to emitted light photons. This model establishes that the film need be characterized solely by its DQE with respect to the emitted light photons. We show that the only explicit exposure dependence of the screen-film DQE results from the film DQE, while the spatial-frequency dependence is dominated by the screen MTF. However, depending on the screen gain, a significant exposure/spatial-frequency domain may exist where the overall DQE is independent of both of these factors and depends mainly on x-ray absorption efficiency.

A range of model DQE calculations illustrates the detailed influence of the most important screen-film parameters. The screen is characterized by typical values for the primary x-ray absorption, spread function associated with secondary photons, and statistics associated with the gain of the conversion process. An existing film DQE model is used to explore the effect of film parameters on screen-film performance. Examples are given of the conditions under which screen-film DQE becomes independent of both film DQE and screen-film MTF. The influence of all these factors on system performance is discussed.

The maximum-entropy method (MEM) for computing noise spectra has been extended to accommodate two-dimensional microdensitometric scans of uniformly exposed radiographs. Based on computer-simulated short scans, the MEM requires one-tenth the data of the conventional FFT method, for the same statistical stability. In particular, MEM estimates of very low-frequency noise power have proved to be substantially more stable than FFT estimates. MEM spectra also have the advantage of being smoother than the FFT method. For measured data, we found the results to be quite sensitive to the nature of the detrending algorithm which was applied to the raw data.

The signal-to-noise ratio associated with recording Q quanta may be defined in terms of a lesser number of noise-equivalent quanta (NEQ). Measurements are presented here for a screen-film system as used to record x-rays, and the results are expressed in three-dimen-sional form by mapping out the NEQ surface as a function of exposure and spatial frequency.

The speed of radiographic screen film systems may strongly vary according to the ambient temperature. Temperature dependence of the conversion efficiency of the phosphors of the screen is hereby a major parameter. Data are presented for several commercial screen film systems. The phenomenon is of ter overlooked in normalisation sheets for sensitometric characteristics evaluation.

The x-ray to light conversion efficiencies of several commercially available screens were determined over a temperature range of 0 - 50°C using 80 kVp x-rays generated from a diagnostic fluoroscopic x-ray unit. The efficiencies of the rare-earth screens investigated (LaBrO:Tb & La2C2S/Gd202S:Tb) varied by less than 5% over this temperature range whereas that of the conventional screen (BaPb604) dropped to approximately 60% of the value at 00C. This slight temperature dependence for the terbium activated phosphors is ascribed to the weak coupling of the host crystal lattice with the inner electron shells involved in the luminescence transitions.

Multiparameter optimizations have been carried out to study the effects of technological advances on the patient doses required to maintain a given image quality in mammography. The assumed advances include; improvement in the power loading limits of the tube focal spot, increased absorption efficiency for a given detector resolution, detector system gain increases and changes in the exposure time limitations that result from patient motion. The optimization permits system geometry, kVp of the examination, filtration, detector resolution, focal spot size and grid characteristics to vary simultaneously and self-consistently subject to image quality as well as technological constraints. In this study the technological constraints are systematically varied so that the effect of technological innovations can be measured by the resulting dose reductions relative to a baseline optimized system.

Objective interpretation of space and time dependent data as encountered in nuclear ventriculography necessitates drastic data reduction, leading to cardiac function quantifiers. Since the original data contain counting fluctuations in addition to clinical information, it seems legitimate to ask how the derived quantities are affected by the counting noise. This question is addressed here for the case of noise induced systematic errors in parameters based on activity-extrema. A computer model of the observed time-activity curves (TACs) is constructed using a truncated Fourier representation of the noise-free TACs and additive, gaussian noise. In its context a quantitative study of the noise-dependence leads to parametrizations of the systematic effects. A strategy for neutralizing the systematic errors, based on these results, is proposed and tested. It is found to be effective in both simulated and clinical environments.

A single energy technique for measuring and analyzing bone mineral content has been implemented on a Siemens (Somatom-2) head scanner. The measurement is based upon a reference phantom that contains various concentrations of potassium phosphate that is exposed simultaneously with the patient. The method facilitates the analysis of the current status of the patient compared to controls as well as the results of follow-up examinations.

An approach to automate the contour detection of left ventricular images is described. The method is based upon the expectation window principle and was implemented for ventriculograms taken with cineangiography. The results obtained show that the technique can provide high quality contours with only moderate operator interaction.

This study investigates which filters are preferred by radiologists for diagnosing particular pathologies found in digitally acquired renal studies. Four types of filters are applied to twenty digitally acquired renal arteriograms demonstrating a wide distribution of pathologies. Three radiologists, who knew the pathology of the patients, examined representative unfiltered and filtered images from each study and then judged whether the filtered image helped demonstrate the pathology. The Butterworth filter was the most versatile filter, the statistical difference filter was most useful for separating overlapping vessels in arterial phase images and the two edge enhancement filters were least useful, only aiding diagnosis 20%. Overall, filtering was thought by the radiology panel to improve diagnosis for only 65% of the ten abnormal cases in this study.

Noise power spectrum analysis was performed on scintillation camera images. The analysis was performed on a set of field floods, uniform exposures over the field of view, in such a way that stochastic and nonstochastic effects for the set were separable. Stochastic noise for the digitally acquired images was found to approach the limit of Poisson count statistics. The nonstochastic component of the calculations dominated the results at the lower spatial frequencies. The noise power spectrum calculations retained information about the patterns of the noise in real-space images. Examples of digitization, photomultiplier tube mottle and edge-packing artifacts are presented.

In order to provide a PAC system which will appeal to, and be used by our Radiology staff, we have surveyed their attitudes about digital imaging in general and viewing sta-tions in particular. We anticipate installing a number of these stations in the full PACS, and we want these stations to provide improvement over the way we currently handle digital images. For these reasons, we felt it necessary to ask the radiologists who will be using the viewing systems what features and functions they would like to have available. The faculty, fellows, and residents in our Radiology department were asked to complete a questionnaire about digital imaging and workstations for viewing. They were asked to comment on existing viewing systems (i.e. CT consoles , nuclear imaging systems, and DIVA consoles). This paper discusses the results of the survey along with our ideas on how best we can implement the functions needed and requested.

Liquid Crystal Light Modulators (LCLMs), primarily display devices, may act as remarkably flexible image operators. Flexibility may be enhanced using dedicated digital drivers which provide real-time changes in the power supply, safely limiting the energy delivered. On the other hand fixed oscillators optimize display modes. Discussed are challenges in utilizing LCLMs as composite operators with varying sequences of operations accessible in real-time.

As more and more radiological imaging modalities become dependent on digital technology there is a need to bring together digital images for diagnostic comparison. Currently this multimodality display is done by transferring images to film and then displaying them on a light box. In this paper we discuss the hardware and software architecture of a prototype multiple 1024x1024 digital image viewing station. This station is based on the Multimodality Acquisition and Review System (MARS) developed by Gould Incorporated. We present details on MARS hardware and software architecture and discuss our preliminary experience with the system using a sample radiological image data base which we have assembled.

Digital real-time storage of image sequences has been increasingly in demand for medical imaging applications. This paper describes the incorporation of such a real-time digital video storage subsystem into an image processor architecture. The system can store up to 2250 frames of 512 x 512 x 8 bits resolution. The acquired or processed real-time images can be displayed at slow, normal, or fast motion speeds. Alternatively, each frame can be randomly accessed. The disk controller contains an on-board microprocessor to off-load controlling and sequencing tasks from the system. Examined here are general system architecture, disk storage and controller design, and the role of real-time image sequence storage and retrieval in medical imaging and other applications.

Preliminary study with a prototype 2048 X 2048 pixel format DR (Digital Radiography) system is reported. This DR system uses a novel proximity type X-ray image intensifier (PT-XRII) which has a 18" X 1" sensitive area to scan a field of 18" X 18" in 1.0 second. It is also capable of a 2X zoom, which scans a field of 9" X 9" in 0.5 second into the 2048 X 2048 pixel format. Images from this prototype system are compared with that of the conventional screen-film system.

A new x-ray imaging system is being developed for angiographic imaging and blood flow measurments. The system makes use of an x-ray image intensifier optically coupled to a multi-element solid-state photo-detector (Reticon diode array RL1024S). We describe the design of the system and its use in two modes of operation. The first mode is used to produce low noise images by imaging only small regions of interest at a time, thereby reducing noise from the detection of scattered radiation, and by making use of the very large dynamic range (8000:1) of the solid-state detector. The second mode of operation is used to obtain information on the flow of an injected bolus of contrast material through a region of interest. This is accomplished by determining the mean concentration of dye in each 0.33 mm section along the length of an arterial phantom as a function of time over a period of several seconds while the bolus passes through. This data is used to determine areas of flow separation and turbulence due to the presence of simulated stenoses ranging in size from 10% to 84% cross-sectional area reduction. Information on the flow conditions through an artery may be useful for chararterising stenoses or for investigating the conditions associated with the development of atherosclerotic disease.

Electrostatic imaging techniques provide the means to record x-ray images without the use of film. The conventional screen-film photon receptor is replaced by a charged selenium-oxide plate. After exposure, the plate is scanned with multiple electrometer probes forming a 1024x1024x12 bit digital image from the latent electrostatic image. This imaging modality represents an early entry into the era of true digital radiography, evolving toward a filmless diagnostic imaging facility. This paper discusses the architecture of an experimental charged selenium plate projection digital x-ray system. As well as some preliminary results of the characteristics of a amorphous selenium plate photon receptor exposed to visible light.

The concept of selective exposure radiography encompasses those techniques which spatially modulate the incident x-ray field to produce a more uniform exit field arising from the patient. The resulting reduction in the dynamic range of the exposure field offers several advantages. In conventional radiography, selective exposure techniques allow all areas of the image to be placed in the linear portion of the film characteristic curve so that local contrast is maximized. With video systems, the reduction in dynamic range minimizes the impact of electronic noise behind the least transmissive regions of the patient. With both electronic and photographic detectors, selective exposure radiography is characterized by uniform quantum statistics and uniform scatter across the image. Several selective exposure techniques currently are being investigated. They include compensating filters placed manually in the x-ray field as well as fan-beam geometries in which the x-ray tube output is modulated with a feedback circuit to maintain constant exposure to the image receptor. At the University of Wisconsin, we have been investigating a digital system which uses an initial low-dose patient image to design an attenuator with transmission complimentary to that of the patient. The attenuator is fabricated for each patient and is positioned automatically in the x-ray beam prior to the acquisition of the final compensated image. The possible applications of this device include chest radiography, coronary angiography, subtraction angiography, and accurate digital videodensitometry.

This work reports the measured dependence of the DSA image's iodine contrast on kVp, patient thickness, X-ray scatter, and DSA system's veiling glare. It also reports the measured patient skin dose at a constant X-ray image intensifier (XRII) entrance dose for the various kVp and the patient thickness settings. It is found that, for any given patient skin dose, the DSA image quality is most optimal in the 80 - 100 kVp range. However, the DSA image quality is markedly reduced by increasing patient thickness and by the combined effect of X-ray scatter and the DSA system's veiling glare. Clinical examples indeed show that 80 - 100 kVp operation yields very satisfactory images for all range of patient thickness and for both intravenous and intra-arterial DSA examinations.

An approximate analysis of the solid anode heating process is presented for cyclic operation. The result is a composite of several boundary value problems that leads to a three parameter scheme for estimating the maximum surface temperature.

A numerical analysis of the steady-state temperature profile in a thin, liquid-cooled rotating x-ray anode is presented. The numerical solution accounts for two-dimensional conduction and the possibility of a non-uniform electron flux. Use of the numerical solution yields important information about temperatures and temperature gradients in the anode.

A new class of internally liquid cooled rotating anode x-ray tubes capable of high average and high peak power is proposed for medical applications. The principles of this new tube design are discussed and estimated performance levels are compared to levels attainable with conventional solid rotating anode medical x-ray tubes and with existing liquid cooled rotating anode x-ray tubes. Performance improvements of the proposed tube include essentially unlimited anode Heat Unit loading, and anode cooling rates, peak and continuous, that are more than an order of magnitude greater than is currently available. The new tube would enable present high average power demanding procedures such as CT and digital and conventional vascular angiography to be performed more efficiently and with higher patient throughput. New imaging techniques requiring even higher average power levels such as energy subtraction, slit scanning, x-ray spectrum optimization and special scatter rejection methods would become more clinically practical.

A system has been developed for performing longitudinal tomography via reconstruction of a small number (28) of digitally recorded images. This technique, 'digital tomosynthesis', allows the user to select any number of desired tomographic planes in post processing, from a single patient exposure lasting a few seconds. Results will be shown from a short clinical trial at UC Irvine with the prototype system.

In the previous paper we presented clinical results from a digital tomosynthesis x-ray machine which can produce a tomographic slice through the human body in a plane parallel to the sensitive surface of the image intensifier. It stores as raw data separate 512 by 512 pixel images taken from 28 discrete angles. For very large image matrices and faster acquisition times it becomes necessary to acquire images continuously as the x-ray source is moved. This results in images which are convolved with a replica of the source motion, and so deconvolution techniques are required to access the underlying image information. This paper discusses how this can be achieved and describes the results of our first steps towards this goal. It is shown that, although the deconvolution problem is three-dimensional, it can be reduced to an independent two-dimensional computation for each tomographic level of interest. It is also suggested that similar techniques could be used to reduce clutter from unwanted planes after synthesizing images from multiple discrete angles.

An adaptive image filtering scheme is presented for edge and contrast enhancement. This filtering technique is based on a sliding window, which moves across the image in both vertical and horizontal directions. At each step, the density value of the center pixel in the window is transformed to a new value via the filtering equation. This algorithm is superior to that of unsharp masking, a technique reported in the past few years, in the sense that it does not uniformly enhance the high frequency noise components of the image for the sake of edge enhancement. Since the parameters of this filtering algorithm are adaptively adjusted according to the statistics of the local area (window), there are no user (operator) interactions required. The relation between the size of the sliding window, the degree of enhancement, and the stability of the filter are shown in applications concerning visualization of digital angiograms and chest radiographs.

Clinical images obtained with an experimental digital chest system represent a unique image data base in that a large number of normal and abnormal chest images in digital format have been made available without the need to digitize films. The above image data base was used to measure certain characteristics of this class of images. Entropy and grey level histogram data and frequency spectrum statistical data were compiled for selected regions in the chest for the purpose of designing an effective image compression algorithm. By segmenting images into regions of high and low entropy, the former can be compressed in either a reversible or irreversible manner, while the latter are synthesized from knowledge of relatively few parameters in the process of displaying the image. The effect of compression ratio on image quality is assessed in a qualitative non-diagnostic manner, and it is found that significant compression can be achieved without objectionable degration of image information content with irreversible coding.

The increased volume of diagnostic images which are electronic in nature has resulted in a need to improve the time consuming and costly manual means of handling these images. An electronic image viewing station for the M/NET PACS system has been developed by MEDINET, INC. The main design focus of the M/NET viewing station is its easy-to-use applications software. Its hardware and software design is presented in this paper.

The evolution of the multiformat camera for medical diagnostic imaging has been extremely rapid. Over a period of little more than ten years, the camera has changed from a relatively crude device for recording images from an oscilloscope screen, to highly automated systems capable of a large number of image sizes, with maintenance of consistent photographic results due to automatic compensation for a number of image variables. The camera design presents many interesting problems, made up of mechanical, electronic, and optical components. This has resulted in the development of manufacturers who have specialized very successfully in the production of this imaging component.

A diffraction limited laser based multiformat camera for medical imaging having a contrast transfer function (CTF) value of 90% at 2048 pixels per image width x 2048 pixels per image height and a CTF value of 37% at 4096 pixels per image width x 4096 pixels per image height has been designed and built. This unit has an 800 step gray scale recording capability. Digital interfacing and system operation are discussed. A discussion of the optical path components is presented, including choice of laser, acousto-optic modulator, film, use of a diffraction limited zoom lens, and an apodization filter. Optical resolution, scanner accuracy and image 'banding' are also discussed.

With the introduction of digital medical imaging modalities and picture archiving and communications systems (PACS) the necessary and desirable capabilities of a film recorder for hard-copy output will vary from the specifications of the current video-based recorder. Image data can be presented to the film recorder in a digital rather than analog video format. The resulting film image is no longer subject to degradation due to nonlinearities in the video portion of the recorder. Spatial resolution is also greatly enhanced (4096 x 4096 pixels). The inherent data processing capabilities of the digital film recorder allow a number of additional functions to be performed internally. Features of the system to be discussed include automatic calibration, automatic failure diagnosis, data compression and expansion, data error detection and correction, gamma and exposure compensation, alphanumeric and graphic frame identification, graphic image recording (EKG, EEG, EMG), color capability, data format conversion, multi-format multi-modality sequences, print spooling from multiple image sources and direct connection to a PACS network.

Conventional multiformat imaging devices using photographic film have been the prime modality for producing hard-copy images from stored digital information or displayed CRT images. As the primary utility of these images develops from that of diagnosis to one of illustrating a diagnosis (either for a referring physician or the patient's chart), some compromise in image quality can be envisioned if the reward is either decreased cost or increased convenience. To this end we have investigated the physician acceptance of a dry silver copying system.

The conventional x-ray image intensifier (XRII) has developed from an aid for fluoroscopy, where resolution is poor and noise high, through XRII photofluorography, where high resolution is needed and noise is moderate, to digital subtraction angiography, where resolution is moderate, noise must be very low and imaging capability must be pushed to its fundamental limits. These limits are not well understood for XRII. A better knowledge of the fundamental properties of the XRII will enable us to understand and determine these limits. We have measured detective quantum efficiency (DQE), spatial modulation transfer function (MTF), spatial noise spectra and temporal frequency response using monoenergetic x-rays derived from x-ray fluorescence. The results can be accounted for by an analysis based on the physical structure and operation of the XRII. Detailed measurements on every tube are unnecessary provided that the salient parameters (thickness, separation, and radii of curvature of the CsI phosphor layer, substrate and input window, MTF and temporal frequency response) are given by the manufacturer. With this fundamental knowledge it becomes possible to understand the limits of existing XRII and to explore the possibility of improving the images by digital enhancement, by better choice of incident x-ray spectra and by modifications of the XRII themselves. This will be illustrated by application to the case of digital subtraction angiography.

This paper describes measurements of the modulation transfer function (MTF), noise power spectrum (NPS), and detective quantum efficiency (DQE) for an x-ray image intensifier based radiography system. The experimental results are compared with the prediction of a theoretical model.

It has been observed that image intensifier contrast ratio measurements taken on clinically installed systems always produce lower values than reported by the manufacturer. One common method of measuring the contrast ratio in the field uses the film recording camera which comes with the system, while the manufacturers always measure the image intensifier isolated from other system components. Evidence is presented showing that the difference between the two measurements is attributable to contrast losses in the optics. Also, evidence is presented suggesting that there is no observable difference in contrast presentation for image tubes with contrast ratios larger than 15. Contrast ratio measurement on clinically installed systems are recommended for performance monitoring purposes only.

The ability to achieve usable high contrast images from a digital subtraction system is largely dependent upon the performance of the TV camera. The radiologist's desire for a greater dynamic range and even higher resolution will impose even more stringent requirements on the camera in the very near future. By proper application of electrical and magnetic excitation to the imaging tube, much higher imaging performance can be obtained in digital subtraction systems. This paper discusses the critical parameters and how to optimize imaging performance.

Advances in the proximity type X-ray image intensifier (PT-XRII) are described. The principal advances are the development of a two-stage PT-XRII and many new applications of this device.

Physical signals acquired, quantized and processed in imaging systems are inherently noisy. Considering the source signal and noise statistics when designing a quantizing and processing system leads to a number of general and practical results. The theory and application of these techniques leads to an optimal bit-efficient quantization format with a number of interesting properties. Applying these can have beneficial effects on hardware costs arithmetic processing and data compression.

We studied the effect of image processing with Metz filters and matched filters on the detection of simulated low-contrast square objects superimposed on radiographic mottle. The signal-to-noise ratios (SNRs) of original and processed images were calculated based on the perceived statistical decision theory model by taking into account the internal noise of a human observer's eye-brain system. Threshold contrasts for objects of various sizes were predicted by assuming a threshold SNR of 3.8 which was determined previously for a 50% correct detection in 18 alternative forced-choice experiments. The relative performance of various image processing techniques was also evaluated experimentally with a contrast-detail diagram method. The simulated images were generated by a high-quality digital image processing and simulation system. The digitized images were Fourier-trans-formed, filtered, inversely Fourier-transformed, and/or contrast-enhanced to produce the processed images. The contrast-detail curves of the original or processed images were obtained by averaging the results of four image samples and twelve observers. Both the theoretical prediction and the C-D experiment demonstrated an improvement in detectabilities of the simple test objects over those of the original images. However, the observers seemed to under-read the filtered images in the sense that the improvement in obser-ver performance was slightly less than the prediction. This is probably caused by the changes in appearance of the object and the noise texture in the filtered images. The usefulness and limitations of the Metz filters and matched filters in comparison with other image processing techniques are discussed.

A prototype video based medical image processing system is evaluated. The system is an adaptation of the Measuronics LMS II technology which is currently used for remote sensing applications. Initial investigations have identified many potential applications of the system in medical image processing. In addition to enhancement of a single radiograph, this system is capable of image subtraction (temporal and energy) and image superposition. This new system may provide an economical means to increase diagnostic information using conventional radiographic techniques and equipment.

The ability to perform post-processing on diagnostic images is becomming more and more important especially as an effort to extract additional information to aid diagnosis. However, it is often the case that researchers need to implement their own image analysis which is beyond the ability of manufacturer-supplied software. This analysis can be done by either directly programming the clinical device or performing the analysis on a centralized image processing system. This paper presents an example where advanced image analysis was performed on a centralized system. Digital fluorographic (DF) blood flow analysis was done after DF images were transferred from the Angiographic Laboratory to the Image Processing Laboratory. The advantages and disadvantages of the use of such a centralized processing scheme are discussed in the light of the DF example.

We have been investigating diagnostic knowledge models which assist in the automatic classification of medical images by combining information extracted from each image with knowledge specific to that class of images. In a more general sense we are trying to integrate verbal and pictorial descriptions of disease via representations of knowledge, study automatic hypothesis generation as related to clinical medicine, evolve new mathematical image measures while integrating them into the total diagnostic process, and investigate ways to augment the knowledge of the physician. Specifically, we have constructed an artificial intelligence knowledge model using the technique of a production system blending pictorial and verbal knowledge about the respective CT scan and patient history. It is an attempt to tie together different sources of knowledge representation, picture feature extraction and hypothesis generation. Our knowledge reasoning and representation system (KRRS) works with data at the conscious reasoning level of the practicing physician while at the visual perceptional level we are building another production system, the picture parameter extractor (PPE). This paper describes KRRS and its relationship to PPE.

Some aspects of an electro-optical system for axial tomography are described. An estimation is made of noise-level and geometrical resolution of the reconstruction system. It appears, that even for an ideal detector, the analog reconstruction system does not limit the density resolution.

Fourier Aperture (FA) consists of two sinusoidal gratings in contact with each other. One possible mode of operation, both gratings are rotated about the same axis through equal angles in opposite directions. A clear difference moire fringes is shown. The moire frequency depends on the angle between two gratings. This aperture of moire fringes is used as a variable frequency grating. It is able to scan the object spectrum in frequency domain. The summation filtered back projection alqorithm with three different apodizing functions were introduced in the reconstruction. 99mTc gamma ray source and 1" diameter NaI scintillation detector was used in the experiment. The thyroid phantom images show the spatial resolution of FA agree with the theoretical prediction. Three-D images capability is also discussed. The matrix inversion algorithm was used in the computer simulation of 3-D images. The depth resolution of FA is limited by the restricted view angle of the detector. Two plane objects with noise free in different depth were reconstructed.

A time varying coded aperture of containing single sinusoidal grating which is translated between the radiative source plane and the detector plane is discussed. To the detector, the aperture behaves as a variable frequency grating that scans the objects over a two dimensional frequency domain by changing its orientation and position. 1-131 isotope was used as the gamma ray source in the experiment. We used summation filtered back projection algorithm to reconstruct the image. The disc and annulus was reconstructed successfully. Because of this single grating aperture has simple geometry and easier data acquisition compared with Fourier aperture. We could use it as a gamma ray imaging device in nuclear medicine. The tomographic image capabilities of this system are discussed also.

Single Grating Aperture (SAG), behaves as a variable frequency grating, records the object spectrum directly in Fourier domain where the quantum noise is introduced in this region originally. This is in contrast to a stationary coded aperture in which the noise appears in space domain in the shadow casting images. The noise properties of SGA imaging system with summation filtered back projection reconstruction algorithm has been derived. The calculation on SNR shows that this imaging system gives the image performance directly proportional to the intensity distribution of the object. The computer simulations of point and uniform disc object show this conclusion correct.

A new digital gamma camera system, Scinticor, TM has been developed which incorporates advances in 1) scintillation crystal design, 2) high speed digital signal processing electronics, 3) high sensitivity collimation, 4) real time data corrections, and 5) fiber optic digital data transmission. Hardware implementation is described and technical performance standards are presented.