The University of Arizona Department of Radiology first considered establishing a photoelectronic radiology department in 1973. It seemed clear that the technology had progressed far enough for us to investigate the possibility of total film replacement.' Data from the space program in particular indicated at that time that sophisticated television images over 1000 x 1000 lines were approaching the detail seen on the traditional x-ray film. This technology has been known over many years of research and development as "photoelectronic imaging devices (PEID) ."14 However, at that time film replacement was out of the question. What was not out of the question was the consideration of using a subtraction technique, "digital video subtraction angiography." To this end, we, and independently the University of Wisconsin,314 proceeded to develop this technology.5'6 Our intravenous video subtraction images in patients started in our research laboratory in 1977 and in March of 1980 we opened a biplane special procedures room dedicated only to photoelectronic imaging (no film).7'8 Digital video subtraction angiography has been successful and is described in much greater detail in these Proceedings by other authors. Current efforts are under way toward total replacement of film. This is an immense problem, one that will require a much greater sophistication of computers, storage devices, system analysis, and cooperation from both the radiologist and the clinician.9'10 In a theoretical study we converted our 65,000 procedures-per-year department to complete photoelectronic imaging (no film) and estimated that we would save approximately five million dollars over ten years.15 Extrapolating this to the entire United States would result in a conservative estimate of saving one billion dollars per year. Not included in these mathematics are cost-effective savings of the physicians' time and effort.

Digital subtraction angiography(DSA)is compared to five other noninvasive imaging methods with respect to physical attributes and medical applications. 1) Digital subtraction angiography measures flow channel (vessel) anatomy and vascular leaks in regions where signals from under and overlying vascular pools do not conflict in strength with the vessel or tissue of interest. 2) X-ray computed tomography, in principle, can separate the under and overlying signals, yet presently it is limited in speed, axial coverage, and computational burden for tasks DSA can efficiently perform. Possible exceptions are the dynamic spatial reconstructor (DSR) of Mayo Clinic and the system under construction at the University of California, San Francisco. 3) Heavy ion imaging measures electron density and is less sensitive to injected contrast than x-ray imaging which has the advantage of the photoelectric effect. A unique attribute of heavy ion imaging is its potential for treatment planning and the fact that beam hardening is not a physical problem. 4) Ultrasound detects surfaces, bulk tissue characteristics, and blood velocity. Doppler ultrasound competes with DSA in some regions of the body and generally involves less equipment and patient procedures. Ultrasound vessel imaging and range-gated Doppler have limitations due to sound absorption by atheromatous tissue and available imaging windows. 5) Emission tomography measures receptor site distribution, metabolism, permeability, and tissue perfusion. Resolution is limited to 7mm full width half maximum (FWHM) in the near future, and extraction of metabolic and perfusion information usually requires kinetic analyses with statistically poor data. The ability of positron tomography to measure metabolism (sugar, fatty acid, and oxygen utilization) and the ability to measure tissue perfusion with single photon tomography (17 mm FWHM) or PET (7 mm FWHM) using non-cyclotron produced radionuclides are the major unique features of emission tomography. 6) Nuclear magnetic resonance procedures measure the concentration of some nuclei (e.g., 1H, 23Na, 32P) as well as their chemical state and the local physical-chemical environment of the resolution volume. Velocity and diffusion are also potential measurements. Two unique capabilities of contemporary interest are the ability to image the spatial distribu-tion of relaxation parameters which give information about the local tissue characteristics, and the ability of NMR spectroscopy to sample (not image) the energy state of phosphorous in selected regions of the body. A third attribute of importance is that possible tissue heating seems to be the only hazard and this can be controlled.

Digital x-ray systems used thus far for non-film subtraction angiography may be divided into two main categories, those using image intensified video fluoroscopy apparatus and those using scanned linear detectors. On each type of system energy subtraction and time subtraction have been investigated. And, even if we limit ourselves to subtraction angiography, several types of energy and time subtraction algorithms have been tried.

Using the standard format of an observer detection study, the comparative performance of digital radiography using a CT localization device and film screen radiography in the detection of pulmonary nodules was evaluated. The experiment was designed to test the operation of these systems in a simulated clinical environment by using an anthropomorphic chest phantom.

Tomosynthesis, an imaging mode related to classical tomography, is designed to produce arbitrarily many tomograms through processing of an appropriate set of radiographs. This paper examines the applicability of digital radiography technology to this technique. In this approach, component radiographs are recorded using efficient detectors and stored directly onto disk memory whereupon tomograms are synthesized under flexible software control and displayed on a CRT. Digital image processing is easily performed to enhance these images. Experiments were conducted by operating a third generation CT scanner in the scanned projection mode. This paper gives an analysis of tomography using traditional geometries and extends it to deal with the fan-beam geometry. Experimental results are also shown.

The concepts of using the fluoroscopic image for producing tomograms and digitized representations of these images is based on the concepts of the single scan radiographic system and the use of a video disk first introduced by Baker and Miller.1 These authors showed that by blanking the video camera during the x-ray exposure and then introducing a delay period before allowing the electron beam to sweep the camera tube target a considerable increase in signal-to-noise ratio could be achieved. Building on this concept Lasser, et a1,2 achieved true fluoroscopic longitudinal tomography. This was done by placing a negative voltage on the camera target during the long period (l, 2 sec) required for the x-ray tube-image receptor motion to be completed. Excellent tomographic images having slice thicknesses of less than 2 mm were produced at exposure levels of 25-60 mR. Images were recorded for several motion patterns found on standard clinical tomographic equipment. These were; linear, circular, and hypocycloidal.

The intense synchrotron radiation produced at electron storage rings provides a new source of X-rays highly suited to iodine K-edge digital subtraction angiography. The high intensity and small angular divergence permit the radiation to be monochromatized by Bragg diffraction and made available in beams of small vertical size, of arbitrary horizontal width, and of tunable energy. The use of such beams provides maximum sensitivity to intra-arterial iodine and virtually eliminates image contrast due to non-vascular body structures. The sensitivity of this method to iodine offers the prospect of visualizing arteries by peripheral venous injection.

Experiments and calculations are used to list the properties of integrating X-ray detectors important in digital applications. This list includes costs, detective quantum efficiency, non-linear effects, signal to noise ratio, and noise equivalent number of photons. Indicated are the critical areas, where improvements are still possible. Naturally all future improvements on the detector side will relieve constraints for the X-ray source and will improve scanner versatility and image quality.

A prototype high purity Germanium detector array designed for use in the pulse height analysis mode as a discrete photon energy detector for medical imaging was fabricated and tested. Results obtained as a function of the photon input rate, energy resolution, and techniques for correcting for pulse overlap at high count rates will be discussed. The ability to detect small. quantities of contrast agents, and the discrimination of photoelectric and Compton images is greatly enhanced by utilizing this detection technique, in theory, and the experimental results obtained from a 1x5 cooled array will be presented.

Conventional radiographs (digitized or not) do not provide information about the depths of details and structures because they are two-dimensional projections of three-dimensional bodies. Taking advantage of the finite size of the X-ray source and the divergent nature of the X-ray beam, a radiograph can be processed by two-dimensional digital filtering techniques, so that the image of a particular layer is improved, while the others are degraded. This technique is referred to as a Tomographic Filtration Process (TFP). This paper explains the mathematical and physical foundations of the method and the engineering considerations in the design and realization of tomographic filters. Theoretical comparisons between conventional radiography, standard tomography and tomographic filtering are discussed in terms of the thickness of the tomographic layer, the rate of change of the Modulation Transfer Function (MTF) and the signal to noise ratios. Finally, experimental results are shown to demonstrate the effect of tomographic filtering at different depths.

The vague concept of "resolution" of a detector aperture (or of the source distribution in transmission imaging) is quantified in terms of an elementary signal discrimination task. The signal-to-noise ratio that characterizes performance of the discrimination task is obtained from statistical decision theory. Defining efficiency in terms of the square of this ratio, we find the uniformly redundant array to be 15 times as efficient as a simple open aperture or source of the same area for discrimination tasks when the objects are point-like. Otherwise, it is more efficient to use the simple large open aperture or source.

Applications of conical rotating aperture (RA) geometries to digital radiography are described. Two kinds of conical RA imaging systems are the conical scanning beam and the conical scanning grid assemblies. These assemblies comprise coaxial conical surface(s) the axis of which is collinear with the x-ray focal spot. This geometry allows accurate alignment and continuous focusing of the slits or the grid lines. Image receptors which use solid state photodiode arrays are described for each type of conical RA system: multiple linear arrays for the conical scanning beam assembly and multiple area arrays for the conical scanning grid assembly. The digital rotating-aperture systems combine the wide dynamic range characteristics of solid state detectors with the superior scatter-rejection advantages of scanned beam approaches. The high scanning-beam velocities attainable by the use of rotating apertures should make it possible to obtain digital images for those procedures such as chest radiography which require large fields of view and short exposure times.

An attack on background effects which degrade dose efficiency may be made by up to four moving slits installed in a diagnostic image intensifier system. These background effects are scatter, off-focal radiation, and intensifier glare. A version of slit radiography con-sisting of sectors rotating about the central ray of the x-ray tube is studied. Experiments are performed which evaluate the feasibility of this method. System options and trade-offs are discussed.

A computerized imaging system is composed of a detection/conversion stage and an image processing function. Interactive processing/display, being an integral part of computerized radiography, is heuristically approached from a system-design point of view.

A high resolution hybrid system for video densitometry is introduced. The system utilizes 100 mm spot films taken directly from the output phosphor of an image intensifier. These images are magnified using a zoom lens and subsequently digitized by means of a Plumbicon TV camera and a Quantex DS-30 video image processor. The zoomed images are digitized in a 512x512 format with 8 bits/pixel. These images are then transferred into a Data General NOVA 3 computer for video densitometric analysis. The limiting resolution of the system is the resolution at the output phosphor of the image intensifier. Experimental results using phantoms to evaluate the hybrid system's ability to quantitate percentage stenosis are presented.

Blurred-mask subtraction enhancement is useful for improving the contrast and visibility of small objects within radiologic images. The process improves visibility by equalizing large area densities and by enhancing the borders of objects within the image. This form of enhancement is applied to radiologic images in the clinical setting by using a hybrid video/ digital image processing system.

In order to study the mechanical properties of breathing, a dynamic series of digital transmission images is acquired using a conventional gamma camera without collimator as a digital detector and a point source of Tc-99m (5 milli Curie, 140 keV y-rays) about 2 m from the crystal plane of the camera. This radiological set up allows the use of the gamma camera at intrinsic resolution and sensitivity. The transmission images reflect the changes in lung expansion and organ movement during respiration. The image series is processed by means of a temporal Fourier transform to represent amplitude and phase of the harmonic changes in transmission at the fundamental breathing frequency. The application of the image processing techniques allows to differentiate different breathing patterns in normal individuals and to differentiate normal from pathological states. The study therefore illustrates that even crude digital transmission signals from the point of view of spatial resolution may yield dynamic information of physiological and clinical interest.

Intravenous angiography in combination with film subtraction has a limited applicability due to the cumbersome film handling procedure and moreover offers only a limited contrast gain factor. Digital subtraction on the other hand can provide a much larger contrast gain factor so that large-area contrast differences can be visualised much better. Some of the main topics such as contrast gain factor and dynamic range of the two techniques are discussed. Experiments with low contrast phantoms prove that Digital Vascular Imaging (DVI) is superior as regards contrast gain.

Digital subtraction videoangiography has been proven to be suitable for imaging the left ventricle and other cardiac chambers.l,2 The high contrast sensitivity and good cancellation of structural background enable this imaging technique to be used with intra-venous injections, which result in complete mixing of contrast medium with blood in the left ventricle.3 The digital nature of the image processor used in subtraction video-angiography enables convenient data collection and analysis. For these reasons, digital videoangiography is well suited to videodensitometric calculation of physiological quantities such as left ventricular ejection fraction.

One of the most successful applications of digital radiography systems thusfar has been the imaging of vasculature opacified by intravenously administered contrast agents using temporal subtraction. Because the contrast in the arterial structures of interest is small as a consequence of the dilution due to the intravenous injection, it is essential that such imaging systems have high contrast sensitivity. This paper discusses means to assess the contrast sensitivity of imaging systems using iodinated phantoms and temporal subtraction. A difference image is shown demonstrating a 1 mm vessel containing 0.5 mg/cm2 of iodine, i.e. less than 1% radiographic contrast. Initial CDD results are presented. Dependence of CDD analysis on video signal level is demonstrated. The effects of x-ray scatter and image intensifier veiling glare are predicted and measured experimentally. Contrast levels in two digital fluorographic clinical procedures are estimated.

The visualization of blood vessels using noninvasive administration of contrast agents is clearly a desirable clinical goal. It opens up the opportunity for mass screening of vessel disease and the study of asymptomatic people who are at high risk. A number of approaches have been taken to view vessels noninvasively. In this paper we present some initial considerations of a system using measurements at three broad energy spectra.

Intervening tissues in projection radiography often seriously interfere with the observability of other low contrast structures. Dual energy techniques resolve this difficulty by removing a selected substance from the image, enhancing the relative contrast of other materials. Undesired materials will remain; these residues, as well as quantum noise, limit the accuracy of material isolation. Parameterizations of the attenuation coefficient lead to simple expressions for the signal-to-noise ratio (SNR) in the dual energy radiograph. The SNR depends on the absorption path, the imaging energies, and the cancelled material. The magnitude of material residue depends only on the atomic numbers and may be an important consideration in the selection and development of new contrast agents.

One of the potentials of Digital Radiography has been for quantitative measurement, both spatial or contrast. This potential has been realized for contrast in computerized tomography. All projection radiography is somewhat limited in quantitative density analysis by line of sight integration. However, by comparing two successive images of the same subject, differences between them may be assessed. If that difference is due to contrast material which fills an organ, then the image differences can be used to infer geometric organ information. To be successful the imaging system must be of sufficiently low noise, have low external and internal scatter, and have a well known X-ray energy to digital signal transfer function. If these conditions are met, the signal differences may be calibrated absolutely. AS&E's flying spot digital radiographic system has been used to implement two examples of this technique. They are the determination of lung line of sight depth using Xenon as a contrast material and determination of cardiac chamber dimensions by using the cyclic variations in line of sight blood depth.

Developing digital radiography techniques provides greater diagnostic information while utilizing less invasive procedures and/or decreased patient dose. An experimental scanned projection radiography system has been built using a CT detector and data acquisition system to provide increased contrast resolution and flexibility in data manipulation. Modifications to the basic system allow dual energy scanning, and subtraction algorithms relying on the energy dependence of the mass attenuation coefficient have been implemented.

A prototype digital chest unit capable of imaging patients in the conventional upright position is described. The unit employs a scanning slit configuration and discrete detector array. The general features of the detector and data acquisition system are discussed as well as the unit's physical performance parameters and geometry. The resolution along and transverse to the scanning direction is in agreement with theoretical predictions. Measurements of the digital chest unit's signal-to-noise ratio versus detector exposure, detector and data acquisition dynamic range, scatter transmission and exposure linearity are also discussed.

A linear diode array was used to scan a radiation field for pulsed high energy x-ray imaging. With a signal of about 50 pico-coulombs of charge from each diode for each beam pulse, the signal to noise ratio for an image of an open field was 119. An image of a porcine femur in a wax phantom was obtained by scanning the transmitted radiation. The resultant electronic image was superior in quality when compared to the standard radiotherapy port film. The simplicity of the detector system has demonstrated the feasibility of constructing an electronic portal imaging system for clinical use.

Dual energy scanned projection radiography of the abdomen has been performed using an experimental line-scanned radiographic system. Digital images simultaneously obtained at 85 and 135 kVp are combined, using photoelectric/Compton decomposition algorithms to create images from which selected materials are cancelled. Soft tissue cancellation images have proved most useful in various abdominal imaging applications, largely due to the elimination of obscuring high-contrast bowel gas shadows. These techniques have been successfully applied to intravenous pyelography, oral cholecystography, intravenous abdominal arteriog-raphy and the imaging of renal calculi.

Clinical evaluation of CT localization devices for chest imaging have demonstrated the advantages of wide dynamic range and edge enhancement produced by image processing. Major disadvantages in relation to film screen radiography are inferior spatial and temporal resolution. An observer detection study utilizing digitized film screen radiographs of patients with diffuse small pulmonary nodules were performed. Relatively low levels of spatial resolution are required for the detection/discrimination of these abnormal pattern types. More complex observer detection studies are required to determine the spatial resolution requirements for electronic digital radiography of the chest.

A preliminary study was used to test the strengths and weaknesses of our dual energy radiography system as applied to study of chest lesions. Six patients had a pulmonary nodule or pulmonary nodules known to be present on the basis of prior chest films substantiated by full lung tomograms or CT scans of the chest. The results of the dual energy studies as compared to the chest film findings and the other special studies are reported.

A digital subtraction fluoroscopic (DSF) system with two video processors in series has been analyzed and evaluated clinically. In this system the initial processor operates simultaneously in an averaging and differencing mode, resulting in an output that is the difference between the most recent input video frame and an exponentially weighted average of preceeding frames. The content in this processor's memory (the mask image) is continually upgraded by each incoming frame or, alternatively, frozen in time by the operator. The second processor averages the difference image output from the first unit. A continuous, real-time, noise reduced subtraction sequence is viewed on a monitor and stored on an analog disc recorder. A video tape recorder preserves raw data and permits retrospective viewing of an injection sequence using different processor settings.

Digital fluoroscopy is a technique in which it is possible to produce high resolution images at low contrast in a dynamic mode. In order to produce images with the desired resolution of 1 mm at 1% contrast level at T.V. frame rates using conventional broad area image receptors, an input exposure to the image receptor of approximately 1 mR per frame is required.1 Assuming a transmission of the order of 1%, this implies an input exposure to the patient of several hundred R/minute. This is much higher than conventional fluoroscopic exposure rates and is well in excess of normal cinefluorographic rates.

A digital fluoroscopy system has many desirable features for energy selective imaging. Unfortunately it has significant problems which may prevent it from being used for this application. Two techniques for overcoming the limitations of digital fluoroscopy are presented. The first attacks the lack of absolute measurements in data from digital fluoroscopy. The nonlinear processing used in energy selective imaging requires absolute data. Techniques are presented which use relative measurements and still allow selective material imaging. Digital fluoroscopy is normally used for subtraction imaging so that errors in the data tend to cancel. By using a generalization of subtraction imaging called hybrid subtraction, the errors introduced by using fluoroscopic data in energy selective imaging can be significantly reduced.

Digital fluorography (DF) images are degraded from ideal through the action of a variety of random processes associated with the imaging chain. Several studies of the sources of the noises have been reported [1, 2]; these, and earlier studies have concentrated on evaluation of the contributions to total DF system noise output. This paper utilizes previous results along with a signal evaluation to provide suggestions of desired DF system operating parameters. In addition, several additional noise sources which may arise in practical clinical DF systems are evaluated.

When energy subtraction methods are employed in digital radiography it is necessary to combine the initial images in a weighted fashion. This paper discusses the instrumentation developed to automatically generate up to quadratic order combinations of x-ray images at realtime video rates in energy subtraction digital fluorography. Although the energy mode used is three-beam K-edge subtraction, the same instrumentation can also be used for dual beam imaging. The method is implemented by first acquiring logarithmically amplified digital fluorographic images of a calibration phantom with each of three spectrally different x-ray beams. The distribution of iodinated contrast material in the phantom is presumed known and the phantom thicknesses cover the range expected to be encountered clinically. Next, individual pixel values from the images are automatically fed to a PDP 11/10 computer where the weighting coefficients are calculated by least squares. The optimum coefficients are next loaded from the computer into a hardware multiplier network capable of combination of 256 x 256 pixel images at realtime video rates. Once loaded with the fitted coefficients the network is then calibrated and ready to generate energy sub-traction images of arbitrary objects. The complexity of the image combination may range from linear to full quadratic combination of differences of the three initial images. The emphasis of this paper is a discussion of the closed form iodine specific fit used to determine the coefficients and the instrumentation aspects of the system: design of the computer interface and design and operation of the multiplier network. Sample experimental results are presented.

We have recently interfaced an ADAC DPS-4100 "Digital Radiography Processing System" with an existing Siemens X-ray - image intensifier imaging chain and Sierra Scientific DAV-16 video camera with an Amperex 73-XQ plumbicon. This report presents our experiences with the integration process, and preliminary results of the system's evaluation.

It has been shown that perception of the information in angiographic images can be decisively improved through digital image processing1. The application of one such technique, the method of parametric or functional imaging, to studies of the kidney, liver, brain, the heart and other organs has been demonstrated by our group in previous publications2,3,4 First used in nuclear medicines, the fundamental idea of functional imaging is to replace a temporal sequence of images with a single static image which displays a certain aspect of the temporal change. The advantage of the method is two-fold: Often important intensity changes with a striking time structure are hard to perceive directly from the image sequence because the amplitude of the intensity variation is small. Also the human visual system is very well suited to the assessment of static images, but has difficulties in quantifying temporal changes. Parametric image processing thus yields images which enhance certain aspects of temporal change. This paper attempts to summarise the parameters found to be most useful in a variety of clinical applications.

A noninvasive method for diagnosing acute tubular necrosis and rejection would be an important tool for the management of renal transplant patients. From a sequence of digital subtraction angiographic images acquired after an intravenous injection of radiographic contrast material, the parametric images of the maximum contrast, the time when the maximum contrast is reached, and two times the time at which one half of the maximum contrast is reached are computed. The parametric images of the time when the maximum is reached clearly distinguish normal from abnormal renal function. However, it is the parametric image of two times the time when one half of the maximum is reached which provides some assistance in differentiating acute tubular necrosis from rejection.

Using digital electronic techniques it is possible to process the video signal from a conventional image intensified fluoroscopic system so as to allow isolation and amplification of a small iodine signal produced after the intravascular injection of contrast medium. Such a device has been in use in our department for the last two years, primarily for performance of intravenous arteriography. More recently, we have been using the apparatus to perform arteriography after injection of the iodinated contrast medium directly into the arterial system. In this report the technique, advantages and limitations of such an approach will be discussed.

A photoelectronic imaging system has been built and developed at the University of Arizona. Initial studies have been directed toward digital video subtraction angiography (DVSA) using intravenous injections of contrast material. DVSA has been applied clinically to the study of the heart, the aortic arch and great vessels, the kidneys, abdominal and lower extremity vessels, the pulmonary arteries, and the vessels of the head and neck, generally with remarkable success. This paper deals with our clinical experiences with DVSA in evaluation of the vessels of the head.

The clinical experience of DSA of the head and neck in more than 1,000 patients is presented. In a comparative study of 100 conventional carotid angiograms, the accuracy rate in evaluating extracranial carotids was 97% in technically satisfactory studies. Intra cranial vessels were examined in 55 patients with both conventional selective catheterization and intravenous DSA. In 65% of the patients, the DSA examination was diagnostic. DSA is a safe, rapid procedure that can be performed on an outpatient basis and provides diagnostic information comparable to conventional angiography.

Digital radiography, in conjunction with digital subtraction methods can provide high quality images of the vascular system,1-4 Spatial resolution is one important limiting factor of this imaging technique. Since spatial resolution of a digital image is a function of pixel size, it is important to determine the pixel size threshold necessary to provide information comparable to that of conventional angiograms. This study was designed to establish the pixel size necessary to identify accurately stenotic and ulcerative lesions of the carotid artery.

Using perfusion lung scans as a guide, digital video subtraction angiography of the pulmonary arteries was performed in human subjects suspected of having pulmonary embolism. Dogs were employed as a pulmonary embolism model and both routine pulmonary angiography and intravenous pulmonary angiograms were obtained for comparison purposes. We have shown by our preliminary results that the technique is extremely promising as a safe and accurate alternative to routine pulmonary angiography in selected patients.

A computer-aided operator-interactive method has been developed to measure changes in the diameter of the coronary arteries as a result of the contractile motion of the heart and of vasodilators and constrictors. The method has been tested in animal models: mongrel dogs of 15-20 kg weight were subjected to cinefluorograohic procedures taken at a rate of 30 frames/sec. The selected width of segments of coronary arteries were measured with an EyeCom image analyzer interfaced to a VAX 11/780 comnuter at a resolution of -20 microns on 100 frames per injection. Ten scan lines were averaged for each segment: furthermore, in order to reduce scanning noise, each scan line was taken four times and averaged. The blood vessel diameter was determined as the horizontal distance of the base points in the blood vessel image profile. Autocorrelograms, power spectra and an ad-hoc method termed F-technique were used to analyze periodicities in the data. It was found that the amplitudes of periodic oscillations in the blood vessel diameters may be almost as large as the changes caused by some vasodilators and constrictors.

Sequential observations of coronary artery disease depend upon reproducibility of measurements of coronary dimensions. Because the coronary artery segment to be digitized for quantitation occupies only a small portion of the 35 mm cine frame, a cine projector was mounted on a movable X, Y stage so that the area of interest could be focused through a lens turret system onto a video camera. The lens turret provides lx, 3.5x, and 7x magnification. A digital image processing device operating under computer control permits direct application of smoothing and edge finding algorithms to the magnified coronary image. Using separately recorded grids, an external reference marker, and isocentering of the heart within a U-arm radiographic system, it is possible to adjust for distortion and magnification. A split screen technique provides capability for single plane or sequential biplane analyses. A digital lightpen is used to input fiducial points for segment identification and define an initial coronary margin which is superceded by computer defined edges. Final measurements include maximum, minimum and mean diameter of 1.0 cm long coronary segments and the absolute area of the segment profile, along with provision for hard copy image recording. Reproducibility of the minimum and maximum diameter measurements provided by the digitization system and edge finding algorithm are + .06 mm for vessels with an average diameter of 2.3 mm (+ S.D.). When duplicate coronary injections during the same catheterization procedure are made, the coefficient of variation of the minimum diameter is 9.5%, of the maximum diameter 5.0% and of the segment area 3.9%. The capability for area of interest selectivity, variable optical magnification and digital image analysis provides a unique and reproducible system for coronary segment quantitation.

A new approach to blood vessel boundary estimation is presented in this paper. By appropriately modelling the blood vessel as a dynamically evolving state vector, and by taking into account the Poisson statistics of the x-ray imaging noise, we arrive at a state-space system with a non-linear measurement equation which includes non-Gaussian, non-additive noise. MAP smoothing equations are derived for the state vector describing the vessel, and the optimally smoothed state vector is found by a dynamic programming search. This method performs especially well in images with low SNR and low sampling rate. The performance of the proposed method is demonstrated by the boundary estimates obtained by applying the algorithm to a simulated vessel and measurement data as well as to real vessel phantom measurement data at various SNR's.

Computerized digitization and processing of roentgen video images recorded at a rate of 50 per second was tested in intravenous angiocardiography in normal pigs weighing 15 to 20 kg. Roentgen video images were recorded in the 4-chamber view obtained by 30-35 degrees caudocranial angulation of the x-ray tube and 50-60 degrees LAO obliquity in the pig. Significant contrast enhancement was obtained through ECG-gated background subtraction and rescaling after integration of multiple background as well as contrast images. Occasionally, histogram equalization was used to further enhance contrast. To study temporal changes in cardiac motion, time parameter extraction or functional imaging was applied as well. The left and right heart were well visualized after intravenous injection of 1/3-1 cc. 76% Urografin per kg. bodyweight. Special purpose processing like subtraction of the end systolic phase from the end diastolic in the left and right ventricles as well as subtraction of the right ventricular phase from the left ventricular phase was also performed. If the left ventricular end systolic phase was subtracted from the end diastolic, most of the left atrium was also subtracted whereby the left ventricle was seen without continuity or superimposition of the left atrium. Experimentally generated ventricular and atrial septal defects as well as patent ductus arteriosus could be detected using the described technique. The results of the animal experiments became the basis for subsequent applications in children with congenital heart disease.

Digital subtraction video densitometry is utilized in the determination of left ventricular ejection fraction. The mask mode subtracted video images obtained after a peripheral intravenous injection of a bolus of Renografin 76 are used in the derivation of a density time curve. The left ventricular ejection fraction (LVEF) is computed by an algorithm described in this paper. Results for 15 patients are compared to the LVEF's obtained by applying the area-length method to the same patient data. The left ventricular ejection fractions computed by both methods had a good correlation (r=0.96) in fifteen patients.*

Digital subtraction angiography (DSA) has considerable utility in the evaluation of peripheral vascular disease. It is useful in screening selected patients for vascular disease and its relative ease of performance and good patient tolerance make it ideal for serial examinations of post operative patients. When used in conjunction with intra arterial injections, the technique may show "run-off" vessels which were not demonstrated by standard angiography. This paper presents our experience using DSA to image peripheral vascular problems.

Utilizing digital video subtraction angiography, we have successfully imaged peripheral vessels in both pre- and postoperative patients. The intravenous igiograms were matched with routine arteriograms, when available, and the in quality compared favorably in the vast majority of cases. in all cases, the intra-venous angiograms were adequate for diagnosis, sparing patients the risk and discomfort of additional routine arteriograms.

A mathematical filter which improves the contrast response of image intensifiers is derived. The filter is obtained by inverting the point spread function (PSF) of the system. A point spread function which takes into account the spread of light in the image intensi-fier system is proposed using physical arguments. The PSF has two parameters which are the range and the fraction of light which spreads. These two parameters are determined by a least squares fit to the experimentally measured data. The inverse filter is used in the deconvolution of video images. The results indicate that a substantial improvement in the contrast ratio of the system can be achieved by this technique.*

A computerized fluoroscopy method for isolating low fluoroscopic image contrast, which results during the flow of iodinated contrast materials through cardiovascular structures is described. This technique requires applying one of a family of recursive filtering algorithms which have been designed to isolate time varying image contrast. Applications of this technique for intravenous angiography are described. Hardware implementation of these algorithms also is described and preliminary results on human subjects is presented.

To display the wide range of attenuation values in a computed radiographic image, mathematical processing to suppress low frequency information is often used. This increases the apparent visibility of low contrast structures and permits the simultaneous display of these structures in all areas of the image. Low frequency suppression is performed by subtracting from each pixel in the image a fraction of the weighted mean of a number of surrounding pixels. This produces a degree of edge enhancement which results in an image similar in appearance to that of a Xeroradiograph. The fraction of the weighted mean and the number of surrounding pixels to be used in the filtering process can be varied, thus producing varying degrees of edge enhancement and low frequency suppression. We have investigated the effect of different low frequency filtering parameters on the recognition of low density lung nodules. Digital radiographs of an anthropomorphic phantom which contained lesions ranging in size from 3 to 20 mm were obtained using the GE SV 8800 system. Different filters were applied and images were obtained on film. The detectability of the lesions using the different filter parameters was determined.

An imaging system resolution is characterized usually by the Point Spread Function (PSF) and less stringently by a Line Spread Function (LSF). Since no practical system can realize a widthless PSF, such as a Dirac Delta Function, the neighbouring pixels contribute to the grey level of the pixel in question. Then it follows that the contrast in any pixel is modulated by the sum of components of the grey level of the adjacent pixels, the degree of modulation depending on the nature of the PSF and the grey level of the surrounding pixels. Consequently, the system resolution is scene dependent and hence, to some degree, specific to a clinical procedure. Clinical significance of this approach is indicated. Causes of scene dependent artefacts are considered. Method of estimation of this effect in digital imaging systems is suggested. The implications to the choice and application of the electronic imaging system chain, on the basis of the scene dependent component analysis, are discussed. The consequences of this component to digital image subtraction procedures are outlined.

Our institution like many others, is faced with a proliferation of medical imaging techniques. Many of these methods give rise to digital images (e.g. digital radiography, computerized tomography (CT) , nuclear medicine and ultrasound). We feel that a unified, digital system approach to image management (storage, transmission and retrieval), image processing and image display will help in integrating these new modalities into the present diagnostic radiology operations. Future techniques are likely to employ digital images, so such a system could readily be expanded to include other image sources. We presently have the core of such a system. We can both view and process digital nuclear medicine (conventional gamma camera) images, positron emission tomography (PET) and CT images on a single system. Images from our recently installed digital radiographic unit can be added. Our paper describes our present system, explains the rationale for its configuration, and describes the directions in which it will expand.

Pictorial information theory is restated in a form that predicts, with fair precision, the minimum radiation required to detect a given feature size in diagnostic radiology. In contrast to conventional approaches, the probability of detection is made a visible variable in the calculation--an aspect of prime importance for diagnostic interpretation. An accurate knowledge of the absorption coefficients of the tissue concerned and the sizes one wishes to detect is required, but the calculation is relatively easy to carry out with a programmable calculator or microcomputer. The concept of "viewing efficiency" is introduced to compare the performance of a real radiographic system with that of a perfect radiographic system. The theory is applied to mammography, where the viewing efficiencies of film, film/ screen combinations, and xeroradiography are compared. It is shown that a more efficient method for detecting the latent charge image in xeroradiography, coupled with digital data storage and appropriate viewing methods, should result in substantial reductions in exposure compared to the present toning methods.

A microprogrammable ultrasonic image processor and its applications to image enhancement are presented. An image processor has been implemented to input, manipulate and display the ultrasonic image data with a high image quality and also to facilitate the human operation. Although this processor has specially designed circuits for ultrasonic image processing, a microprocessor and its associated arithmetic circuits can be utilized for programmed image manipulations; for example, smoothing, enhancement, contrast stretching, area/arc-length measurement and so on.

A dedicated digital processor is described capable of digitizing a high resolution video signal from a fluoroscopic TV camera into an 810 x 600 matrix in real time. For less demanding applications, a 512 x 512 matrix can be substituted. The sampling clock frequency is 15 Megahertz giving a Nyquist bandwidth limit of 7.5 MHz. A 7 MHz phase equalized eliptical filter at the input prevents aliasing and the production of false artifacts in the picture. Eleven bit digital processing follows an 8 bit analog to digital converter. Noise reduction is accomplished by a one frame recursive filter in which the filter coefficients are adjusted by a patented motion detector on a pixel by pixel basis to reduce motion smear. The lower perceived noise permits X-ray dose reduction of 2 to 8 times while retaining high quality pictures. A noise reduced spot picture can be frozen by a foot controlled switch permitting a further reduction of dosage and eliminating the need for a troublesome disc recorder. This noise reduced picture can also be used as a subtraction mask in an optional version of the equipment. A minimum of front panel operator controls for best human interface is accomplished by the use of a programmed read only memories to control all functions including noise reduction and frame storage.

Subtraction radiography for longitudinal studies requires both reproducible imaging geometry and film contrast characteristics to permit perfect alignment of the radiographs, and achieve good cancellation of diagnostically irrelevant background structures. The standard deviation (SD) of gray levels about the mean in a subtraction image was used as a relative measure of the residual structured noise. In order to estimate the effects of improperly standardized radiographs on SD in the subtraction image, both the imaging angle and film exposure time were systematically varied. The results showed that SD changed linear for small misalignment angles of the central beam, the variance attributable to this error source reaching about the same magnitude as the variance due to anatomical differences for angulation errors within ±20. The SD increase due to large film contrast disparity could be partly reverted for angulation errors within this bound by using a quadratic transformation which matched the first two moments of the gray level distributions in the two parent radiographs. Therefore, in order to use some of the retrospective data obtained under less stringent standardizations for subtraction imagery, it appears possible to adjust for differences in film contrast, and, perhaps, correct for geometric misalignment by a separate algorithm.

This paper discusses the feasibility of an integral communications system to include images for medical applications, for both intra-hospital communications and networking to remote locations and systems. Emphasis is placed on image communications because they will become an important component of telecommunications traffic in the 1980's if the proper facilities are provided. Terminals distributed throughout a hospital and in remote locations could access the databases of images and medical reports. Radiologists could prepare the medical diagnosis reports much faster, shortly after the radiograph was taken, by examining the radiographs on a suitable soft-copy display and dictating the report, ultimately using automated voice recognition systems. To lower the cost of the system the expensive hardware could be shared by means of communications, so that their utilization is maximized. Several configurations are proposed in this paper and their relative cost/performance compared. System evolution is also considered.

Image storage is a necessary consideration in a medical imaging system. Data generated by recently developed digital radiographic techniques may be stored on magnetic tape or disc in either analog or digital form. Video discs have desirable freeze frame and image sequencing properties allowing for flexible image manipulation and display. Whereas digital pre-processing has enabled excellent results to be obtained using analog storage, digital storage is ideal as far as bandwidth and noise properties are concerned. However, serial data rates of digital disc drives are limited to about 10 megabits per second, too slow for recording video information in real-time. Using a standard multi-platter magnetic disc drive we have constructed parallel read/write channels servicing sets of 9 or 10 surfaces simultaneously in order to achieve a data rate of 100 megabits per second. This permits storage of 815 512x512x9 bit images at 30 frames per second. Hardware configurations and applications of the real-time digital disc to subtraction angiography will be discussed.

The intolerance of medical images to error places special demands on data reduction techniques. Quantization in the spatial or transform domains, widely studied in other data compression applications, is irreversible and therefore not applicable to medical image coding. Rather, we must rely on effective prediction and prediction error coding. This paper compares seven differential pulse code modulation (DPCM) predictors on the basis of average prediction error and channel induced error propagation. A Laplacian model of prediction error distribution suggests a simple family of variable length codes and an algorithm for adaptively selecting that code which best suits the anticipated predictor performance.

The emergence of digital display, storage and retrieval equipment has prompted a standard format for transmitting images between different type machines having access to a common channel. The standard assumes that the major cost in image transmission is access to the channel and that computation costs at the sending and receiving machines are small and will continue to get smaller. The standard allows data compression schemes and flexible picture element (pixel) representations. Standard format messages can contain variable numbers of images of different aspect ratios, resolutions, pixel types and compression schemes.