Medical imaging can now take wider advantage of Computer-Aided-Manufacturing through rapid prototyping technologies (RPT) such as stereolithography, laser sintering, and laminated object manufacturing to directly produce solid models of patient anatomy from processed CT and MR images. While conventional surgical planning relies on consultation with the radiologist combined with direct reading and measurement of CT and MR studies, 3-D surface and volumetric display workstations are providing a more easily interpretable view of patient anatomy. RPT can provide the surgeon with a life size model of patient anatomy constructed layer by layer with full internal detail. Although this life-size anatomic model is more easily understandable by the surgeon, its accuracy and true surgical utility remain untested. We have developed a prototype image processing and model fabrication system based on stereolithography, which provides the neurosurgeon with models of the skull base. Parallel comparison of the model with the original thresholded CT data and with a CRT displayed surface rendering showed that both have an accuracy of 99.6 percent. Because of the ease of exact voxel localization on the model, its precision was high with the standard deviation of measurement of 0.71 percent. The measurements on the surface rendered display proved more difficult to exactly locate and yielded a standard deviation of 2.37 percent. This paper presents our accuracy study and discussed ways of assessing the quality of neurosurgical plans when 3-D models a made available as planning tools.

Stereotactic neurosurgery is a technique in which a rigid frame is applied to the patient''s head and pre-operative images acquired. Because the frame and the lesion are visible in the images, the lesion can be located relative to the frame. Devices may then be attached to the frame to direct surgical instruments to the lesion. Conventional stereotactic neurosurgery remains a point by point process, conceptually little changed from the original devices which were designed for use with pneumoencephalograms. The exponential rise in the amount of available imaging information over the past 15 years has not been matched by intrasurgical applications. A new device will be presented which allows the intrasurgical position and trajectory to be displayed on preoperative images. This device has sub-millimetric accuracy and precision and is limited only by the image voxel size. The device can use both CT and MRI image sets concurrently or exclusively. Applications include surgical planning, biopsy, bone flap location and intracranial localization. Both phantom and clinical procedures will be shown.

3D imaging offers the opportunity to gather new kinds of information for diagnosis and therapy. In order to access this information efficiently we need a model for the integration of knowledge which is extracted from the images. In this paper we present such a model for the description of image information as well as results of image analysis and manipulation. The purpose of the model is its use in computer assisted treatment planning. Therefore, we introduce a volume representation based on discrete 3D space to map the model on the computer. We call this representation an extended cell enumeration (XCE) representation. It is an extension of the well-known cell enumeration (CE) representation where graph structures are added to represent high-level knowledge together with the cell information. We show how to transfer information into this representation and how to access the represented information. We give examples of applications in the planning of stereotaxy and plastic surgery.

A Multibeam radiation therapy treatment is a non-invasive technique devoted to treat a lesion within the cerebral medium by focusing photon-beams on the same target from a high number of entrance points. We present here a computer assisted dosimetric planning procedure which includes: (1) an analysis module to define the target volume by using 2D and 3D displays, (2) a planing module to issue a treatment strategy including the dosimetric simulations and (3) a treatment module setting up the parameters to order the robotized treatment system (i.e. chair- framework, radiation unit machine). Another important feature of this system is its connection to the PACS system SIRENE settled in the University hospital of Rennes which makes possible the archiving and the communication of the multimodal images (CT, MRI, Angiography) used by this application. The corporate use of stereotactic methods and the multimodality imagery ensures spatial coherence and makes the target definition and the cognition of the structures environment more accurate. The dosimetric planning suited to the spatial reference (i.e. the stereotactic frame) guarantees an optimal distribution of the dose computed by an original 3D volumetric algorithm. The robotic approach of the treatment stage has consisted to design a computer driven chair-framework cluster to position the target volume at the radiation unit isocenter.

A surgeon typically uses information from a number of tomographic imaging methods (e.g., CT, MR, PET) during the course of a surgical procedure. These imaging techniques represent three-dimensional information as a set of two-dimensional images. To use this information, the surgeon is required to mentally construct a three-dimensional visualization from the set of two- dimensional images. The formation of the mental image becomes more complicated with the inclusion of multiple imaging modalities and multiple imaging planes. We have developed a technique to enhance the mental three-dimensional visualization process through simultaneous graphics and multislice raster image display. The composite display, capable of displaying up to three raster images along with a patient-specific graphics model, is viewed on a 1280 X 1024 monitor. The raster images, displayed in a 512 X 512 format, may be any combination of imaging methods and imaging planes. The graphics model, determined from the imaging data, may be freely rotated as a depth-cued wireframe or shaded-surface model. Regions-of-interest may be incorporated into the graphics model for additional visual cues. Trajectory information may be obtained by moving a three-dimensional cursor in any raster image space or in the graphics model with instantaneous update of the remaining display area. This design allows the surgeon to interactively obtain orientation and visualization information from the images in the operating room. Because the classic imaging planes are used, the surgeon is not required to deal with a new information format or a loss of resolution.

Surface descriptions are difficult to specify. Though image processing techniques are well established to generate nearly any planar or three-dimensionally curved surface, methods to describe such complex shapes are often disorienting. Even the best intentioned interface for surface description can confuse a seasoned user. This paper introduces a surface definition technique that is simple, accurate and intuitive for the needs of routine medical image analysis. We describe a procedure to define a curved surface based on surface intersection points in a series of parallel images. In this medical context, points selected describe a surface that contains pathology of diagnostic interest. Using this technique diagnostic views are generated that conform to natural anatomic shape, physicians are no longer restricted to orthogonal or even single curve surfaces. This user interface provides analytic descriptions to produce surface views that use a Fourier-shift technique for reconstruction. Surfaces through a volume are produced with resolution equal to that of the original data set. Example images are illustrated.

Modern imaging technology places new perceptual and cognitive demands on skilled interpreters of medical images. As a result, medical users sometimes perceive new display systems as clumsy and unfriendly because they are required to interactively manipulate images in ways that are foreign to them. The primary goal of this research is to transfer some of the burden of image manipulation from the observer to the computer. This will be accomplished by using clinical history prompts to set gray-scale imaging parameters that are most likely to enhance relevant diagnostic information called for by the prompt. The image processing is done in a series of invisible, off-line ''steps'' which normalize the gray-scale range, and display an image that has been tuned by an optimum gray-scale Look Up Table (LUT) derived from clinical-history data. The criteria that we have emphasized in designing the workstation are that it (a) is user friendly, (b) incorporates human-perception principles in the display interface, and (c) utilizes gray-scale transfer functions that improve contrast resolution of prompt-targeted diagnostic information. We will report on an experimental display workstation that links clinical-history prompts to an optimized gray-scale LUT. The clinical histories provide the basis for generating rules that call up a gray-scale transfer function designed to optimally display targeted diagnostic information in the medical image. From a perceptual point of view, a workstation that can display optimized images would mean that the initial transformed image not only matches the diagnostic disease category, but also the observer''s perceptual hypothesis so that a global diagnostic interpretation is possible before any fine- tuning of window and level settings. Because the initial perceptual encounter with the image fits observer expectancies, global pattern recognition should be facilitated.

We have developed a system to automatically adjust the display window width and level (WWL) for MR images using a neural network. There were three main points in the development of our system as follows: (1) We defined an index for the clarity of a displayed image, and we call this index ''EW''. EW is a quantitative measure of the clarity of an image displayed in a certain WWL, and can be derived from the difference between gray-level with the WWL adjusted by a human expert and with the WWL adjusted by this automatic system. (2) We extracted a group of six features from a gray-level histogram of displayed images. We designed a neural network which is able to learn the relationship between these features and the desired output (teaching signal), ''EQ'', which is normalized to 0 to 1.0 from EW. Learning was performed using a back-propagation method. As a result, the neural network after learning is able to provide a quantitative measure, ''Q'', of the clarity of images displayed in the designated WWL. (3) Using the ''Hill climbing'' method, we have been able to determine the best possible WWL for displaying images. (a) The maximum Q is searched for and found from roughly sampled WWLs. (b) The WWL sampling intervals are gradually made finer. (c) The WWL with maximum Q searched in (b) is selected as the best possible WWL. We have tested this technique for MR brain images. The results show that this system can adjust WWL comparable to that adjusted by a human expert for the majority of test images. The neural network is effective for the automatic adjustment of the display window for MR images. We are now studying the application of this system to sagittal and coronal images.

Quality Assessment or Quality Assurance (QA) in PACS has its roots in QA procedures which have been developed in the course of many years of radiology practice. The need for QA in all aspects of radiology is being escalated by more complex technology, administrative controls, and economic factors. Growth in PACS is leading to increased demand for QA at the system level, as well as for individual PACS components and modalities.

The advent of picture archiving and communication systems (PACS) promises to confer a number of significant advantages upon the practice of medicine, in general, and upon radiology, in particular. This promise, notwithstanding, numerous important obstacles to a complete, clinically effective, implementation of this technology exist. Not the least of these obstacles pertains to the area of quality assurance (QA). The remainder of this manuscript will review the approach that many conventional departments of radiology have taken to QA problems. This information will be used as a foundation for conjecture about QA issues that are likely to be important in a totally digital environment.

This paper will attempt to identify many of the important issues for quality assurance (QA) of radiological modalities. It is of course to be realized that QA can span many aspects of the diagnostic decision making process. These issues range from physical image performance levels to and through the diagnostic decision of the radiologist. We will use as a model for automated approaches a program we have developed to work with computed tomography (CT) images. In an attempt to unburden the user, and in an effort to facilitate the performance of QA, we have been studying automated approaches. The ultimate utility of the system is its ability to render in a safe and efficacious manner, decisions that are accurate, sensitive, specific and which are possible within the economic constraints of modern health care delivery.

A comprehensive quality assurance (QA) program should be implemented for all teleradiology and picture archival and communications (PACS) systems. In this report we summarize our QA experience with a teleradiology system that includes a laser digitizer for x-ray film. A key component required for the evaluation of laser film scanners is an appropriate test pattern; digitizers should be evaluated with enhanced test patterns specifically designed for this purpose. The phantom pattern should measure high contrast resolution, low contrast discrimination, gray scale linearity, geometric distortion and noise. In addition, a uniformly exposed sheet of film (approx. 0.3 OD) serves as a good phantom for testing screen non- uniformities of viewing-station monitors. It is also suggested that clinical images should be included in a QA program. Finally it is recommended that any discrepancies in the interpretation of teleradiology images should be monitored and investigated.

Various quality control (QC) procedures may be used to evaluate image quality for picture archival and communications (PACS) systems. A standard PACS QC protocol, applied on a regular basis is desirable to insure optimal diagnostic performance. We describe a QC phantom designed especially to test PACS systems performance. We describe a QC phantom designed especially to test PACS systems that acquire images by digitizing x-ray films. The phantom is a sheet of x-ray film upon which a digital test pattern is printed. Multiple parameters of image quality are tested, including resolution, contrast, gray scale, geometric distortion and noise. Individual test patterns are incorporated to detect specific artifacts of laser scanner digitizers. As part of a regular QC protocol, the phantom provides an objective measurement of change in digital image quality over time, as well as an objective means for comparison with other systems.

In 1985 the Society of Motion Picture and Television Engineers (SMPTE) published a Recommended Practice (RP-133) entitled Specifications for Medical Diagnostic Imaging Test Pattern for Television Monitors and Hard-copy Cameras. Since that time the SMPTE monochrome test pattern has been applied to the acceptance testing and quality control of video and image display systems, and hard-copy (film) recorders. The major features of the test pattern will be discussed along with applications and the problems demonstrated with the pattern. Furthermore, the test pattern will be used to demonstrate that color displays may exhibit only one-half of the resolution of a monochrome monitor while the display contrast (due to limited dynamic range) may be only 10 to 20 of that obtained with a monochrome display.

As Picture Archiving and Communication System (PACS) technology has matured, video image capture has become a common way of capturing digital images from many modalities. While digital interfaces, such as those which use the ACR/NEMA standard, will become more common in the future, and are preferred because of the accuracy of image transfer, video image capture will be the dominant method in the short term, and may continue to be used for some time because of the low cost and high speed often associated with such devices. A series of digital phantoms has been developed for display on either a CT9800 or Hilite Advantage scanner. The phantom images have been stored on magnetic tape in the standard tape archive format used by General Electric, so that the images may be loaded onto the scanner at any time. These images are then captured using a commercial video image capture board in a PC/286 computer, where the images are not only to be displayed, but also analyzed with the use of an automated process implemented in a computer program on the same PC. Results of the analyses are saved, together with the data and time of image acquisition, so that the results can be displayed graphically, as trend plots.

The efforts of a manufacturer in the area of component procurement acceptance testing are very visible, while the quality assurance work that goes on during the design and manufacture of medical imaging equipment is frequently not so apparent. Many steps in the design and production of imaging devices are carefully designed to promote not only quality of the devices when initially installed, but also the ability to maintain the device''s quality over a period of time. Customer requirements for aesthetics, performance, safety and regulatory compliance must be considered. Software, which represents over 70 of the design engineering of modern products, presents a special challenge since the quality of software cannot be tested and must be designed in. Specific hardware elements such as monitor phosphor color and uniformity deserve extra consideration. A design that allows quick diagnosis and replacement of failed components is also important.

This paper presents the results of both physical and psychophysical evaluations of the limiting noise of high resolution CRTs at luminance levels from 0.83 to 87.5 ft-L. The study concentrated on the effect of CRT noise on the human observer''s ability to detect small low contrast objects displayed on the CRT. Temporal as well as spatial noise of the CRT was determined physically with the aid of a photomultiplier-based evaluation system. Just- Noticeable-Differences (JNDs) were determined psychophysically using various test patterns and human observers. It was found that the application of the Rose model can be extended from purely photon noise to other noises, particularly those exhibited by CRTs. The results demonstrate a close correlation between physical and psychophysical evaluation: For the luminance levels under consideration, the detection of the human observers is limited by the spatial noise (phosphor granularity); the temporal noise plays a role only at low luminance levels. There is a direct relationship between the JND and the CRT noise. This result implies that for the luminance levels investigated, an optimal display function can be calculated using the results of a fairly easy physical measurement of noise.

Despite the many benefits associated with digital imaging technologies, the adoption of non- film media for primary diagnostic readings has been slowed by two main factors: insufficient screen resolution to permit reliable viewing of many images and insufficient performance to provide an interactive environment. MegaScan has designed a video subsystem which addresses the image quality needs of the radiology community by providing ultra-high spatial resolution. The display monitor has been the subject of several studies and is now in its second generation design. Engineering improvements have addressed the issues of brightness, dynamic range, and brightness uniformity. The MegaScan display controller contains a multiple memory architecture designed to preserve 12-bit data and yet provide fast image loading, image viewing, and image enhancement performance in a networked computer environment.

The quality of the digital image acquired with an image intensifier-TV system depends on a number of parameters. The present work describes procedures to take into account and control the major parameters that influence the image, and how it is perceived by a viewer. This is done both by adjustment of the equipment and by digital image restoration. To further increase the diagnostic yield of chest images, image processing routines for local contrast enhancement were developed and evaluated. Digital images were compared to conventional radiograms by ROC analysis of simulated nodules and line structures on an anthropomorphic phantom. The digital, 1024 matrix, image-intensifier system was found to be well suited for chest imaging, provided optimized equipment and processed images were used.

A study to test the ability of a high-fidelity system to digitize chest radiographs, store the data in a computer, and reprint the film without altering diagnostic observer performance is reported. Two hundred and fifty-two (252) chest films with subtle image features indicative of interstitial disease, pulmonary nodule, or pneumothorax, along with 36 normal chest films were used in the study. Films were selected from a key word search on a computerized report archive and were graded by two experienced radiologists. Each film was digitized with 86 micron pixels and stored in 4000 X 5000 arrays using a research instrument. Replicates were printed using a commercial laser film printer (Eastman Kodak Company) having 80 micron pixels. Originals and replicates were observed separately by two different experienced radiologists. Each indicated a graded response for the three possible pathologies. The agreement of observers between responses for replicates and originals was described by the kappa statistic and compared to the agreement when rereading the original film. The final result of this study supports a hypothesis that the replicate is indistinguishable from the original.

Communication between radiologists and clinicians could be improved if a secondary image (copy of the original image) accompanied the radiologic report. In addition, the number of lost original radiographs could be decreased, since clinicians would have less need to borrow films. The secondary image should be simple and inexpensive to produce, while providing sufficient image quality for verification of the diagnosis. We are investigating the potential usefulness of a video printer for producing copies of radiographs, i.e. images printed on thermal paper. The video printer we examined (Seikosha model VP-3500) can provide 64 shades of gray. It is capable of recording images up to 1,280 pixels by 1,240 lines and can accept any raster-type video signal. The video printer was characterized in terms of its linearity, contrast, latitude, resolution, and noise properties. The quality of video-printer images was also evaluated in an observer study using portable chest radiographs. We found that observers could confirm up to 90 of the reported findings in the thorax using video- printer images, when the original radiographs were of high quality. The number of verified findings was diminished when high spatial resolution was required (e.g. detection of a subtle pneumothorax) or when a low-contrast finding was located in the mediastinal area or below the diaphragm (e.g. nasogastric tubes).

Understanding the mechanisms of ventricular fibrillation and defibrillation requires analysis of epicardial and endocardial potentials throughout the heart. Plunge electrodes permit recording of cardiac potentials at epicardial and endocardial sites, and allow determination of electrical gradients. They also enable us to determine the arrhythmia recurrence sites following failed defibrillation; these sites may be epicardial or endocardial. Therefore, it is necessary to relate the position of the plunge electrodes to the cardiac geometry. We have developed an interactive, computer graphics based system that allows us to locate plunge electrodes on digitized MRI slices of a heart. The system, which can work with any type of image, allows us to identify the epicardial and endocardial points of each plunge electrode on the different MRI slices. Up to 128 different plunge electrodes may be identified to the system. Normalized 3-D coordinates for each epicardial and endocardial electrode point are computed and stored in data files on the computer. Geometry information obtained from this system permits a more thorough understanding of the electrical signals recorded by the plunge electrodes. This information can be used in the study of cardiac excitation and arrhythmias and could help in the development of a more effective lead system for ventricular defibrillation.

A system for display and analysis of neurological digital angiographic images is described. Images are archived along with patient data in a central database. The angiographic images can be displayed simultaneously with images from other modalities for study of anatomy. Digital videodensitometric techniques are used to calculate geometric and dynamic parameters from the angiographic image sequences.

Stroke is the third leading cause of death in the United States. It is caused by ischemic injury to the brain, usually resulting from emboli from atherosclerotic plaques. The carotid bifurcation in humans is prone to atherosclerotic disease and is a site where emboli may originate. Currently, carotid stenoses are evaluated by non-invasive duplex Doppler ultrasound, with preoperative verification by intra-arterial angiography. We have developed a system that uses a color Doppler ultrasound imaging system to acquire in-vivo 3-D color Doppler images of the human carotid artery, with the aim of increasing the diagnostic accuracy of ultrasound and decreasing the use of angiography for verification. A clinical TL Ultramark 9 color Doppler ultrasound system was modified by mounting the hand-held ultrasound scan head on a motor-driven translation stage. The stage allows planar ultrasound images to be acquired over 45 mm along the neck between the clavicle and the mandible. A 3- D image is acquired by digitizing, in synchrony with the cardiac cycle, successive color ultrasound video images as the scan head is stepped along the neck. A complete volume set of 64 frames, comprising some 15 megabytes of data, requires approximately 2 minutes to acquire. The volume image is reformatted and displayed on a Sun 4/360 workstation equipped with a TAAC-1 graphics accelerator. The 3-D image may be manipulated in real time to yield the best view of blood flow in the bifurcation.

The Department of Radiological Sciences, UCLA operates five MR and four CT scanners located in three different buildings and two mobile sites. We have designed and implemented a multi-channel fiber optic broadband video communication system connecting these scanners together. This system consists of baseband fiber optic transmitters and receivers, a multiplexing headend, and broadband fiber optic transmitters and receivers. It can serve up to 5 km. The video signal from each scanner is sent through a baseband fiber optic link to the headend, where it is frequency modulated and distributed over broadband fiber optic links. A receiver, consisting of a demodulator, a TV monitor, and a channel selector, is placed at fourteen strategic locations including the fiber optic hub rooms, chest, neuroradiology, abdomen, bone, gastrointestinal, genitourinary, and pediatric reading rooms as well as scheduling rooms. A radiologist can use any of these fourteen receivers to view a patient''s CT/MR image in real time by selecting the proper channel assigned to the scanner, and use the telephone to communicate with the technologist to monitor the examination. This fiber optic broadband video communication system has been integrated into daily clinical use.

Newly developed algorithms for processing medical ultrasound images use the high frequency sampled transducer signal. This paper describes demands imposed on a sampling system suitable for acquiring such data and gives details about a prototype constructed. It acquires full clinical images at a sampling frequency of 20 MHz with a resolution of 12 bits. The prototype can be used for real time image processing. An example of a clinical in vivo image is shown and various aspects of the data acquisition process are discussed.

The aim of that work is to perform a data compression on X-Ray images of the breast in order to store them within an image data base. A block adaptive Discrete Cosine Transform method is proposed. A homogeneous distribution of the degradation in the decoded image is obtained using a thresholding approach depending on the permitted coding error for each processed block. The proposed method is tested on a representative set of mammograms, containing the various breast pathologies. The decoded images are then visually analyzed by two experts.

We performed volume compression on CT and MR data sets, each consisting of 256 X 256 X 64 or 32 images, using three-dimensional (3D) DCT followed by quantization, adaptive bit-allocation, and Huffman encoding. Cuberille based surface rendering and oblique angle slicing was performed on the reconstructed compression data using a multi-stream vector processor. For CT images 3D-DCT was found to be successful in exploiting the additional degree of voxel correlations between image frames, resulting in compression efficiency greater than 2D-DCT of individual images. During rendering operations, a substantial amount of thresholding, resampling, and filtering operations are performed on the data. At compression ratios in the range 6 - 15:1, 3D compression was not found to have any adverse visual impact on rendered output. Of these two methods, oblique angle slicing, which involves the fewest operations was found to be the most demanding of small compression errors. We conclude that 3D-DCT compression is a viable technique for efficiently reducing the size of data volumes which must be analyzed with various rendering methods.

Irreversible data compression methods have been proposed to reduce the data storage and communication requirements of digital imaging systems. In general, the error produced by compression increases as an algorithm''s compression ratio is increased. We have studied the relationship between compression ratios and the detection of induced error using radiologic observers. The nature of the errors was characterized by calculating the power spectrum of the difference image. In contrast with studies designed to test whether detected errors alter diagnostic decisions, this study was designed to test whether observers could detect the induced error. A paired-film observer study was designed to test whether induced errors were detected. The study was conducted with chest radiographs selected and ranked for subtle evidence of interstitial disease, pulmonary nodules, or pneumothoraces. Images were digitized at 86 microns (4K X 5K) and 2K X 2K regions were extracted. A full-frame discrete cosine transform method was used to compress images at ratios varying between 6:1 and 60:1. The decompressed images were reprinted next to the original images in a randomized order with a laser film printer. The use of a film digitizer and a film printer which can reproduce all of the contrast and detail in the original radiograph makes the results of this study insensitive to instrument performance and primarily dependent on radiographic image quality. The results of this study define conditions for which errors associated with irreversible compression cannot be detected by radiologic observers. The results indicate that an observer can detect the errors introduced by this compression algorithm for compression ratios of 10:1 (1.2 bits/pixel) or higher.

The discrete cosine transform (DCT) type algorithm is a leading technique in irreversible image compression. The full-frame bit-allocation (FFBA) technique based on DCT has been proven to be an outstanding compression method for radiological image. A new version of this technique, which discards the bit-table and uses entropy coding instead of bit-packing, FFEC, has recently been developed. The results showed that the new method improves compression ratio by a factor of two (e.g., 24:1 vs. 13:1) with a controlled mean-square-error. With the same compression ratio, the FFEC produces about 50 less error than does FFBA. The FFEC has been tested and results are demonstrated in this paper along with FFBA and the splitting and remapping method. This new method is characterized as a simple, error controllable, and highly efficient compression technique.

Sunset is a lossless gray scale compression algorithm designed for a simple hardware implementation based on the prediction error context approach for predictive gray-scale compression. Sunset uses adaptive binary arithmetic coding with neighboring prediction error buckets as compact conditioning contexts for directly encoding the prediction error. A prototype card was built to send or receive either compressed or uncompressed images across the IBM PC/AT bus. A special interface was designed to load the memory buffer of a high resolution color display. The result is a component of a workstation prototype for radiologists and the physician who referred the patient. The Sunset approach handles gray scale images where the bits-per-pel precision is simply an input parameter to the algorithm; the compression algorithm itself is relatively insensitive to this parameter. For hardware simplicity, the error bucket (bin) identifier is determined by a leading-one detector (or priority encoder) circuit on the prediction error so the number of prediction error values per bucket is a power of two. The next less significant bits of the prediction error become the ''extra-bits'' which, when encoded, make the algorithm lossless. The number of extra-bits in a final (catch-all) bucket depends on the bits-per-pel parameter of the uncompressed image.

In this paper a reversible image coder is presented for both 2D and 3D images. The coder is based on hierarchical interpolation (HINT) and arithmetic coding. Simulations on medical images show that the decorrelation efficiency of HINT is image dependent; arithmetic coding approximates the entropy of the decorrelated images within a few percent. An investigation is performed as to the gain that can be obtained by applying a HINT-like decorrelator to 3D images. A variance analysis of estimators used in 2D HINT shows that for the class of images with an isotropic negative exponential autocorrelation function, no significant improvement can be obtained by extending a 2D HINT-like decorrelation method to 3D.

Many issues must be addressed and resolved in order to bring a complete imaging workstation into everyday use by radiologists and medical researchers. Important design issues for developing an imaging workstation include image quality, system response time, the user interface and image storage. The Image Computing Systems Laboratory (ICSL) at the University of Washington has been developing a series of inexpensive graphics and image processing workstations with high performance by taking advantage of a sharp decrease in hardware costs, increasingly more powerful VLSI chips, and versatile personal computers and workstations. After gaining experience with two previous image processing systems, UWGSP3 (University of Washington Graphics System Processor #3), a third-generation workstation based on the NeXT Computer and UWGSP3-HI, a host-independent version, that can work with any host computer via an interface card, were developed. UWGSP3, a highly integrated, low-cost workstation, is a complete image display and computing system capable of meeting many of the requirements of a medical imaging workstation provided that a suitable user interface is developed. To demonstrate this capability, RadGSP, a prototype user interface and application software for radiologist use, has been developed. This paper will first describe the UWGSP3-HI system for background information before describing the implementation and evaluation of RadGSP, and current radiology imaging workstation research in progress at ICSL.

Stereoscopic radiography has been used routinely at the Montreal Neurological Institute for many years. Recently, with the advent of stereoscopic acquisition and display techniques for digital angiography, together with the increased use of 3-D display techniques for medical images, we have developed and implemented a stereoscopic display workstation for use in a clinical context. The system is based on standard AT-bus computer hardware and includes a high performance monitor equipped with a liquid-crystal polarizing shutter to display the stereoscopic images. The most significant application of this system has been planning for the stereotactic implantation of EEG recording electrodes. Here the surgeon has the ability not only to view the imaged anatomy in three-dimensions, but he is also able to interact with the images and to plan surgical procedures in a more realistic manner than traditional 2-D approaches. Display modes include vascular anatomy (from stereoscopic digital subtraction or MR angiography), or a combination of DSA images and 3-D volume-rendered MR or CT reconstructions.

Although the computer industry has begun incorporating new features in their newest computers and workstations, it has not been clear how best to utilize these new media to improve the productivity of the user. One problem stems from relative separation of various disciplines. For example, three distinct disciplines have evolved from visual information processing: image processing, computer graphics, and pattern recognition. All of them manipulate image data in some ways. The main difference between them is the domain where each discipline takes the input and produces the output. Recognizing the importance of merging the three distinct disciplines into one so that the image data can be successfully incorporated into the future computer technology, a new discipline, denoted as image computing, has been established to provide for consistency and efficiency in managing image data. In conjunction with other technologies such as video and computer-generated audio, image computing will play a key role in developing an integrated information processing platform that will be used in many areas in the 1990s. Some of the areas where image computing technology can be applied are presented. Requirements specific to each application are also described. Functions required of a typical computing workstation will be listed and each requirement will be investigated in detail. We describe how the continuing advances in technology will benefit image computing, and predict how the software algorithms of the future will be employed in image computing. We also introduce some possible future products that incorporate image computing technology.

Image quality of flat panel displays, especially Liquid Crystal Display (LCD), has been improved for these years. Among those flat panels, an LCD has advantage for space occupancy, electrical consumption and heat dissipation and so on. Furthermore, its brightness is noteworthy for radiological image display. Because of an external light source of an LCD panel, the maximum brightness can be expected to exceed a CRT''s. In this paper, experimental remarks on characteristics of several LCD panels are mentioned. And also possibility of LCD as a display device of PACS workstation is discussed. A portable PACS workstation using a LCD panel based on PC under development is introduced.

The Surgeon''s General of the military services created the Medical Diagnostic Imaging Support (MDIS) project to exploit the results of extensive imaging research efforts over the past ten years. The MDIS project will achieve the objective of implementing filmless medical imaging systems at several military medical treatment facilities over the next 4 years. Filmless medical imaging systems must be presented to decision makers via strategic principles to foster support. MDIS is a superior alternative for health care delivery when compared to film- based image management systems which are inherently limited by film as a hard copy media. Four enabling core technologies make it possible to system integrate an effective filmless system for military medicine. These filmless MDIS systems will be acquired from industry through a contracting approach that (1) functionally describes subsystem and system performance for acceptable clinical operations, (2) validates proposed systems through performance evaluation, and (3) makes a system selection and contract award that based on best value for the government.

A comprehensive picture archiving and communications system (PACS), such as the medical diagnostic imaging support (MDIS) system, consists of several interrelated sub systems. The image acquisition subsystem is the means by which images are introduced into the system and as such it is analogous to the ''eyes'' of the system. Images from digital modalities such as computed tomography (CT) and magnetic resonance imaging (MRI) are readily transferable to a PACS since they are acquired in a digital format. Conventional film based analog images are particularly challenging since at no point in their production or display do they exist in an electronic form suitable for transfer to the MDIS system. In recent years, commercial high resolution film digitizers and computed radiology (CR) devices have become available. These devices now provide us with the means to capture conventional radiographic images in a format suitable for transfer to a PACS. Through the careful selection of acquisition devices we can now design an image acquisition subsystem tailored to meet our clinical needs.

A joint DoD effort is in the final stages of contract acquisition to achieve a ''filmless'' hospital environment in the near future. Success of implementation lays to a large degree on an effective image workstation. This paper will discuss soft copy image display (SCID) of the MDIS system including hardware and software.

Installation of Digital Imaging Networks/Picture Archiving and Communications Systems (DIN/PACS) in the medical environment is complex. Installation planning includes accounting for not only physical requirements of the equipment and facility, but also the regulatory requirements of professional accrediting and fire safety organizations and the physical and knowledge needs of the operators.

The research reported in this paper concerns an evaluation of the impact of compression on the quality of digitized color dermatologic images. 35 mm slides of four morphologic types of skin lesions were captured at 1000 pixels per inch (ppi) in 24 bit RGB color, to give an approximate 1K X 1K image. The discrete cosine transform (DCT) algorithm, was applied to the resulting image files to achieve compression ratios of about 7:1, 28:1, and 70:1. The original scans and the decompressed files were written to a 35 mm film recorder. Together with the original photo slides, the slides resulting from digital images were evaluated in a study of morphology recognition and image quality assessment. A panel of dermatologists was asked to identify the morphology depicted and to rate the image quality of each slide. The images were shown in a progression from highest level of compression to original photo slides. We conclude that the use of DCT file compression yields acceptable performance for skin lesion images since differences in morphology recognition performance do not correlate significantly with the use of original photos versus compressed versions. Additionally, image quality evaluation does not correlate significantly with level of compression.

To make use of the three-dimensional information contained in a series of computed tomography (CT) or magnetic resonance (MR) slices, one must either mentally reconstruct the data or generate a model using a computer display. Computer-aided rendering of a three- dimensional surface allows inspection of the model from different angles. Creation of a three- dimensional model in wireframe or shaded surface format from tomographic slices requires identification of points along the surface of interest in each slice. This task can be accomplished (1) by manually outlining the surface in the images, or (2) by a computer algorithm designed to produce points along the surface of interest. In dealing with image sets of the head, objects outside the head (e.g., support structures, stereotactic frames, and noise) complicate the automatic detection of the perimeter of the head. This paper examines three methods of automatic wireframe generation from CT or MR slices of the head. The three approaches are based on the following techniques: traditional edge detection filters, neural networks, and thresholding combined with region connectivity analysis and region elimination. Results of the three approaches are presented along with a comparison of the relative advantages and disadvantages of each.

There are numerous instances in which it is desirable to compare similar, but subtly different, images in an effort to detect important temporal or spectral changes. In practice there are several approaches to addressing this problem, each with its own advantages and disadvantages. Simple image comparison techniques borrowed from observational astronomy are readily implemented now in very modest microcomputer systems. The advantages of the so-called blink comparison technique are dramatic, particularly when comparing complex or crowded scenes. We have implemented a video image blink system in hardware and software which has been used in several medical applications, including colposcopy, dermatological monitoring, x-ray comparisons and various biotechnology projects. To better understand the potential advantages of this methodology we have performed an ROC analysis of a series of synthetic images presented in a conventional photographic display mode and as a set of video blink pairs. The mechanics of implementing this type of visual image display, and the consequences of its availability, are discussed.

A microcomputer-based image processing system is integrated to do three-dimensional reconstruction of serial sections of brain CT/MRI scans from photographic films. The brain scans are taken with a stereotactic registration frame mounted on the patient''s head as reference, thus enabling quantitative 3D reconstruction to sub-mm accuracy. The software includes digitization, data compression, data archiving, image processing, pattern recognition, features retrieval and 3D reconstruction. With manual driven interactive 3D graphics, the system enables neurosurgeons and medical physicists to do interactive diagnostics, analysis and treatment planning. The system put together in this way satisfies both low cost and high efficiency requirements in the Far East medical environment.

Visualization of the liver in three dimensions (3-D) can improve the accuracy of volumetric estimation and also aid in surgical planning. We have developed a method for 3-D visualization of the liver using x-ray computed tomography (CT) or magnetic resonance (MR) images. This method includes four major components: (1) segmentation algorithms for extracting liver data from tomographic images; (2) interpolation techniques for both shape and intensity; (3) schemes for volume rendering and display, and (4) routines for electronic surgery and image analysis. This method has been applied to cases from a living-donor liver transplant project and appears to be useful for surgical planning.

A PC based gateway which can be used for direct digital image transfer will be described. The gateway PC contains an ACR-NEMA interface, a DR11W interface, an Ethernet interface, and a 9 track tape drive interface. The ACR-NEMA interface, which has been designed to be cost effective and reasonable in performance, will be described. The many usages of the gateway, which has been designed to be used to provide the ACR-NEMA connection to various existing equipment cost-effectively, will be described also.

Compression methods based on Gabor functions are implemented for simulated nuclear medicine liver images with and without simulated lesions. The quality of the compression schemes are assessed objectively by comparing the original images to the compressed images through calculation of the Hotelling trace, an index shown to track with human observers for images from this modality.1 For compression based on thresholding the complex Gabor coefficients, better than 2: 1 compression is obtained without significant degradation in image quality.

A film digitizer is used when X-ray images are entered in the PACS. Dust adhering to the film causes white vertical lines on the readout image displayed on the screen. Once generated, the lines will appear on all subsequent readout film images. These white lines must be removed because they adversely affect CRT diagnosis. This paper shows why vertical lines are generated and describes how these lines are suppressed and effects on aliasing caused when a film radiographed using grid is read by the digitizer.