The MIRIAD project (Medical Image Reconstruction, Interpretation, Analysis, and Display) brings together a multidisciplinary team with the objectives of exploring interactive presentation and model-based interpretation of three-dimensional medical images taken from high and low resolution studies respectively. A digitized atlas of normal anatomy can then be used to provide the personal atlas by which the medical image can be appraised and quantified.

Interactive needs of medical visualization require fast processing of huge amounts of data. There is a need for compact storage and efficient handling of the voxel input from CT and MRI machines. The linear octree data structure is an efficient representation technique which leads to less storage and is amenable to different kinds of geometric operations. This data structure is particularly useful in visualizing thresholded images which are binary images. There are several algorithms to generate a linear octree from binary voxel data with time complexity O(n3) for an input of size n3 voxels. We present an algorithm which first extracts the surface of the object. Based on this surface data, the object is partitioned into a set of parallelepipeds where each parallelepiped is a contiguous run of voxels along one axis. Starting from the lowest level of the octree, the algorithm proceeds iteratively to the highest level, computing maximal overlaps of the parallelepipeds at each level. For any level, the voxels which are not in the overlap are octree nodes and are output at that level. The maximal overlapped parallelepipeds form the input to the next higher level in the algorithm. For a connected object having n3 voxels, the algorithm has a time complexity of O(S) where S is the size of the surface of the object. The algorithm has been implemented and tested for a variety of medical data. We also show how this algorithm can be parallelized.

Intravascular ultrasound imaging is a new method for obtaining high resolution images of sections of the arterial wall. It is particularly interesting for detecting features of vascular pathology that are inaccessible by other conventional techniques. We propose a methodology for acquiring image sequences, that allows us to reconstruct a three-dimensional image of the vessels. The various positions of the catheter within the artery lead to geometric distortions of the ultrasound image. First, we have observed, analyzed, and interpreted the most specific reasons for intravascular image artefacts, using calibrated phantoms. Second, sequences of in- vitro pathological segments are acquired. Some pre-filtering methods are tested, in order to ease the segmentation step. Finally, the corresponding 3-D image is reconstructed and visualized, using various volume rendering techniques.

We present a new paradigm, called shell rendering, for volume visualization of surfaces. It significantly overcomes the two major impediments of current volume rendering methods -- high computational cost and storage requirements -- and makes interactive volume rendering feasible in a workstation environment via portable software. It imparts the notion of a structure -- a shell -- to what is usually treated as an amorphous semi-transparent volume and makes morphometrics of surfaces feasible in the volume rendering paradigm.

The development of specialized dyes that essentially prevent light from crossing the film base in duplitized screen-film systems has made it possible to design screen-film combinations with unusual performance characteristics. Specifically, by combining front and back screens with differing light emission and resolution properties with asymmetric films with differing front and back sensitometric characteristics, it is now possible to design screen-film systems that have some or all of the following features: (1) density-dependent image blur, (2) previously impractical sensitometric curve shapes, and (3) screen-dependent system contrast. Performance characteristics of two specific systems are summarized, including sensitometric data, contrast transfer functions, noise equivalent quanta, and detective quantum efficiency. Initial clinical applications of this technology are also described, with an emphasis on thoracic radiography.

Equalization is advantageous in some radiographic applications in order to reduce the dynamic range of the x-ray signal incident on the detector. A technique is discussed which employs several filter wheels, which are radio-transparent wheels mounted near the x-ray tube which can be rotated under computer control. The wheels are designed to contain complex patterns of attenuator material on the annular region of each wheel which intersects the x-ray beam. Rotation of the wheels changes the attenuator pattern presented to the x-ray beam, and therefore this system is capable of regional exposure compensation. The use of multiple filter wheels provides a large selection of compensation patterns, for example, an 8 wheel system with 30 patterns per wheel would allow 1011 patterns. Two different design strategies are discussed, one aimed at digital subtraction angiography, and another at chest equalization. Clinical data bases of 191 DSA images and 250 chest radiographs were employed with computer simulation to evaluate the potential of the filter wheel equalization technique.

A prototype high resolution charge coupled photodiode device (CCD) array coupled to an analog to digital converter (ADC) personal computer system has been developed to assist in the measurement of line spread functions generated in diagnostic radiology for the purpose of objectively evaluating focal spots and film-screen systems. Presently, this task is accomplished with the use of a microdensitometer and associated hardware in order to achieve small enough sampling to make the measurements accurate. An ADC card in a personal computer system samples the output of the 2048 element CCD array, upon which a film containing the LSF is projected. Conversion of the transmitted light through the film by the CCD array permits digitization of the output video signal for subsequent analysis. The CCD computer system is relatively inexpensive and can be made portable, with the potential of providing in the field assessment of LSFs of similar accuracy to the microdensitometer. In addition, the direct computer interface allows for on-line data manipulation and corrections for the characteristic curve, background biases, and nonuniformities, as well as direct calculation of the corresponding MTFs through Fourier transformation.

Magnetic resonance imaging (MRI) has progressed rapidly within the last decade. In addition to better soft tissue discrimination than CT, MRI is capable of imaging a volume in multiple planes. If these planes can be accurately registered, it is possible to create a 3-D image database. This 3-D database can then be used to generate considerably more accurate target coordinates for use in applications such as stereotaxic neurosurgery or radiation oncology where precise measurements are required. Despite these advantages, use of MRI in these applications has been limited because of fundamental questions about the geometric accuracy of the images. Distortion in MR images is inevitable because of nonlinearities and inhomogeneities in the magnetic gradient fields that are used for spatial encoding of the signal. We have developed a model for this distortion which has allowed us to minimize the localization errors using MR images and to improve the geometric correspondence between images acquired in different planes. A special phantom is imaged and analyzed to determine machine dependent parameters for the distortion model. These parameters may then be applied to the geometric correction of patient images, which are acquired separately. The model has been incorporated into an integrated stereotaxic planning system designed for interactive use in the operating room. In this paper, we present details of the model and experimental results.

In a picture archiving and communications system (PACS), images are acquired from several modalities including computed radiography (CR). This modality has unique image characteristics and presents several problems that need to be resolved before the image is available for viewing at a display workstation. A set of preprocessing functions have been applied to all CR images in a PACS environment to enhance the display of images. The first function reformats CR images that are acquired with different plate sizes to a standard size for display. Another function removes the distracting white background caused by the collimation used at the time of exposure. A third function determines the orientation of each image and rotates those images that are in nonstandard positions into a standard viewing position. Another function creates a default look-up table based on the gray levels actually used by the image (instead of allocated gray levels). Finally, there is a function which creates (for chest images only) the piece-wise linear look-up tables that can be applied to enhance different tissue densities. These functions have all been implemented in a PACS environment. Each of these functions have been very successful in improving the viewing conditions of CR images and contribute to the clinical acceptance of PACS by reducing the effort required to display CR images.

The traditional quality assurance programs in a radiology department will undergo a significant transformation as we introduce new technologies with computers such as PACS. A well designed quality assurance program will help assure that these new technologies are used effectively to improve our ability to manage our departments and provide quality care for our patients.

While acceptance of standards for digital image transfer may make video image capture obsolete, this technique of getting an image from a device such as a CT scanner will be in use for many years. Because the devices are inherently analog, these circuits are susceptible to errors in image capture, which can lead to degradation in image quality. We have designed a series of phantom images and used them to periodically measure the quality of captured images. The CT images are displayed at specific window width and window level settings, so that the value of each pixel is known, and is analyzed automatically by a computer program. The procedure involves capturing each of the four quality assurance (QA) images and storing them on the image capture computer. The QA software may be run immediately or it may be run at a later date, when it will analyze images collected over a period of time. The results of the analysis are stored in the computer in a database. This allows displays of the captured image quality, including tables, graphs, charts, and trend plots. A video frame grabber was connected to a CT advantage computed tomography independent console. The images were captured once per week over a period of three months to determine the range of variation which could be expected in the first part of the device's useful life.

Historically, image quality issues involved with matching the `look' of an image on a modality monitor, such as CT, MRI, or Ultrasound, with the `look' at either a remote viewing station or on printed film could be dealt with by adjusting the individual components on an ad hoc basis. As medical imaging moves away from point to point connections and into a networked environment, a standard is needed that insures image integrity with respect to tonescale and guarantees that the look of the image at various imaging systems and printers can match the original modality monitor look. A standard way of describing image presentation using look- up tables is proposed. This provides a `standard' network that each manufacturer can connect to and provides consistent image quality with respect to tonescale. This method expands upon the current ACR-NEMA implementation, which already provides some information that allows for consistent image presentation.

A PACS in clinical use must provide a high degree of reliability in terms of uptime and data integrity. For a number of reasons, this is difficult to achieve. There may exist, on a PACS network, a multitude of imaging and display sub-systems each with its own interface peculiarities and requirements. Because a PACS is usually a multi-vendor system, the interface problems are complex. Furthermore, once a PACS is in clinical use, the nature of the process level interactions are complicated and random. To improve overall system reliability requires not only rigorous component testing but also a complete system evaluation. A testing methodology for the evaluation of a multi-vendor PACS is being developed at our institution. Key features of this methodology include the design of component and system stress tests, and system boundary tests that stimulate the depletion of critical resources. Preliminary results from running such tests at the University of Pennsylvania Medical Center are presented.

This paper discusses the optimization of the display function of an image display with respect to properties of the human eye/brain system. The optimization minimizes the contrast information loss through the display/human observer system and is based on the threshold contrast curve or the curve of just noticeable differences, JND, which one can acquire by psychophysical experiments or physical measurements. It was found that, given the luminance dynamic range of an image display system, the optimum display function is the inverse of the scaled visual response function which is approximately independent of the absolute values of the threshold contrast. In particular, in good approximation, the optimum display function is independent of the spatial frequencies of the displayed image. With an analytical model of the threshold contrast curve, it is shown how the perceived dynamic range depends on factors such as the display device noise, internal scatter and the maximum luminance. The optimum display function based on the threshold contrast curve was employed with a CRT monitor. Preliminary results indicate that an improvement in the overall contrast resolution of clinical images can be achieved compared to operating the CRT with its original display function. The optimum display function guarantees maximum utilization of the contrast information transfer capabilities of a display device uniformly over the available display range. Its suitability as standard for image displays is reiterated. Often, however, for given display tasks, the best display function considering intrinsic image noise, perception requirements, and the desired certainty of perceiving a given contrast detail may be different from the optimum display function. A look-up table should then adapt the display function from the standard optimum state to a task-specific function.

A book cassette containing both a conventional film-screen radiographic system (FR) and a phosphor storage radiographic plate (SR) was used to obtain simultaneous bedside chest images in 22 patients in the Post Operative Cardiac and Surgical Intensive Care Units (ICUs). Twenty-five potential findings of normal structures, lung and pleural disease, and life support devices were recorded for each image in a five point rating format. The FR images are all considered of good diagnostic quality. The original FR films, the laser digitized FR images (DF) displayed on a workstation (WS), and the SR images displayed on a WS were compared. The WS viewing was on a 1 K X 1.2 K, 8 bit monitor. Free adjustment of window level, window width, and black-white inversion was allowed. Magnification allowed access to the 2 K data set. ROC analysis supports the null hypothesis that there is no difference in the diagnostic yield of good quality bedside obtained FR, DF made from good quality FR viewed on a workstation, and SR viewed on a workstation. Analysis of the subset of interstitial and airspace edema indicated that readers gave higher scores for interstitial disease on the WS for both false positive and true positive findings.

The display of a large number of projection radiographs (e.g., AP chest images) for comparison purposes poses potential problems for any electronic environment. In an attempt to assess the concept of rapid sequential viewing, 10 series of AP chest images were each reviewed on a high-resolution workstation under two conditions: (1) simultaneous display of each series in a mosaic configuration; and (2) separate image display in which each image was viewed individually in a rapid sequential mode. In our study, the sequential display was believed subjectively to be of comparable or higher quality by four of six readers. Diagnostic performance (patient improved; no change; patient condition worsened) was comparable for both display modes. Readers were somewhat more comfortable with the simultaneous (mosaic) configuration. Our preliminary results indicate that after minimal training, rapid sequential viewing of AP-chest images may be a reasonable alternative for the display of a series of AP chest images in the ICU.

We are investigating how radiologists's readings of standard intensity windowed (IW) chest computed tomography (CT) films compare with readings of the same images processed with contrast limited adaptive histogram equalization (CLAHE). Previously reported studies where CLAHE has been tested have involved detection of computer generated targets in medical images. Our study is designed to evaluate CLAHE when applied to clinical material and to compare the diagnostic information perceived by the radiologists from CLAHE processed images to that from the conventional IW images. Our initial experiment with two radiologists did not yield conclusive results, due in part, to inadequate observer training prior to the experiment. The initial experimental protocol was redesigned to include more in-depth training. Three new radiologist observers were recruited for the follow-up study. Results from the initial study are reviewed and the follow-up study is presented. In the new study we find that while CLAHE and IW are not statistically significantly different overall, there are specific clinical findings where the radiologists were less comfortable reading CLAHE presentations. Advantages and disadvantages of using CLAHE as a replacement or as an adjunct to IW are discussed.

Optical, non-contact three-dimensional range surface digitizers are employed in the 360-degree examination of object surfaces, especially the heads and faces of individuals. The resultant 3- D surface data is suitable for computer graphics display and manipulation, for numerically controlled object replications, or for further processing such as surface measurement extraction. We employed a scanner with a basic active sensor element consisting of a synchronized pattern projector employing flashtubes that illuminate a surface, with a CID camera to detect, digitize, and transmit the sequence of 24 images (per camera) to a digital image processor for surface triangulation, calibration, and fusion into a single surface description of the headform. A major feature of this unit is its use of multiple (typically 6) stationary active sensor elements, with efficient calibration algorithms that achieve nearly seamless superposition of overlapping surface segments seen by individual cameras. The result is accurate and complete coverage of complex contoured surfaces. Application of this system to digitization of the human head in the planning and evaluation of facial plastic surgery is presented.

A data-, machine-, and application-independent software system called 3DVIEWNIX has been developed and is maintained on an ongoing basis by the Medical Image Processing Group. It is aimed at serving the needs of biomedical visualization researchers as well as end users. Unlike existing visualization packages, 3DVIEWNIX is not designed around a fixed methodology or set of methods providing a fixed set of tools. It incorporates the basic imaging transforms common to most visualization and analysis methods and provides a facility to combine them in meaningful ways. The result is a powerful exploratory environment that not only provides the commonly used standard tools but also an immense variety of others. In addition to visualization, it incorporates a variety of unique tools for multidimensional image analysis. Its design is mostly image data dimensionality independent to make it just as convenient to analyze 2-D and 3-D data as it is to analyze 4-D and higher-dimensional data. It is based on UNIX, C, X Window and our own multidimensional generalization of the 2-D ACR-NEMA standards for image data representation.

This paper presents a case study of the effects of compression error on computerized tissue characterization of normal and fatty ultrasound liver images. Two compression techniques were studied, pruned-tree structured vectored quantization (PTSVQ) and PTSVQ with splitting. Vector quantization is a technique for representing a block of image values, or vector, by the vector in a codebook that is closest to the original vector. Splitting is a technique for decomposing image pixel values into the high and low values. The high values are compressed reversibly while the low values are compressed via PTSVQ. Tissue characterization was accomplished by extracting features from a region of interest (ROI). These features included measuring fractal dimension and statistics concerning run length and co-occurrence probabilities of pixels separated by a given direction and distance. The results were: (1) PTSVQ with splitting produced less image distortion at moderate bit rates than PTSVQ as measured by mean square error; (2) PTSVQ with splitting produced more degradation of the tissue characterizer; and (3) Rotation of the ROIs greatly reduced the degradation of the tissue characterizer for both types of compression. This type of rotation uses interpolation to derive pixel values for rotated lattice points that fall between original lattice points. A possible explanation for these results is that PTSVQ caused irregular distortions at edges depending upon the amount of region information included in the design of the codebook. The interpolation during rotation reduces these irregularities.

The aim of this paper is to propose a new scheme for image compression. The method is very efficient for images which have directional edges such as the tree-like structure of the coronary vessels in digital angiograms. This method involves two steps. First, the original image is decomposed at different resolution levels using a pyramidal subband decomposition scheme. For decomposition/reconstruction of the image, free of aliasing and boundary errors, we use an ideal band-pass filter bank implemented in the Discrete Cosine Transform domain (DCT). Second, the high-frequency subbands are vector quantized using a multiresolution codebook with vertical and horizontal codewords which take into account the edge orientation of each subband. The proposed method reduces the blocking effect encountered at low bit rates in conventional vector quantization.

Adaptive histogram equalization is a contrast enhancement technique in which each pixel is remapped to an intensity proportional to its rank among surrounding pixels in a selected neighborhood. We present work in which adaptive histogram equalization is performed on the codebook of a tree-structured vector quantizer so that encoding with the resulting codebook performs both compression and contrast enhancement. The algorithm was tested on magnetic resonance brain scans from different subjects and the resulting images were significantly contrast enhanced.

This study is a continuation of the work presented previously. We develop the storage aspect of x-ray images of the breast after processing them with an adaptive discrete cosine transform (DCT) coding method. The decoded images are analyzed by experts. The analysis consists to compare the original and decoded mammograms using keywords, i.e., stellate opacity with short or long spicules and with low or high contrast. A real time coding allows experts to choose the optimal coding rate before storage.

Lossy data compression generates distortion or error on the reconstructed image and the distortion becomes visible as the compression ratio increases. Even at the same compression ratio, the distortion appears differently depending on the compression method used. Because of the nonlinearity of the human visual system and lossy data compression methods, we have evaluated subjectively the quality of medical images compressed with two different methods, an intraframe and interframe coding algorithms. The evaluated raw data were analyzed statistically to measure interrater reliability and reliability of an individual reader. Also, the analysis of variance was used to identify which compression method is better statistically, and from what compression ratio the quality of a compressed image is evaluated as poorer than that of the original. Nine x-ray CT head images from three patients were used as test cases. Six radiologists participated in reading the 99 images (some were duplicates) compressed at four different compression ratios, original, 5:1, 10:1, and 15:1. The six readers agree more than by chance alone and their agreement was statistically significant, but there were large variations among readers as well as within a reader. The displacement estimated interframe coding algorithm is significantly better in quality than that of the 2-D block DCT at significance level 0.05. Also, 10:1 compressed images with the interframe coding algorithm do not show any significant differences from the original at level 0.05.

Data compression can improve imaging system efficiency by reducing the required storage space and the image transmission time. Transform compression methods have been applied to digital radiographs with good results. Block transform compression is usually based on 8 X 8 or 16 X 16 transform blocks for the sake of simplicity and speed. Compression with these small sizes tends to require accurate coefficient representations to prevent blocking artifacts. Weighted quantization of block transform coefficients can reduce the blocking effects and improve compression performance. Full frame compression has the advantage of eliminating blocking effects but the disadvantage of heavy demand for computing resources. Small block compression can retain local variation better and has a simpler and faster implementation. We have evaluated the performance tradeoffs for different block sizes and their effects on the image quality of chest radiographs. The results showed that there is no significant difference in root-mean-square error nor in power spectra between different block sizes for visually lossless compression (at about 10:1 compression ratio).

Acquisitions of 3-D (x,y,t) and 4-D (x,y,z,t) medical image sequences become usual, especially in the case of dynamic studies performed with modalities such as MRI, fast x-ray CT, or PET. It results in a significant increase of data volume and, consequently, the necessity, which is emphasized by the development of PACS, to reduce storage space and transmission bandwidth. A multidimensional image sequence coding method based on a multiple decorrelation scheme is presented. In a first step, an orthogonal transform, such as factor analysis (similar to discrete Karhunen-Loeve transform), is applied to decorrelate initial data according to temporal dimension. In a second step, the resulting orthogonal image sequences are decorrelated using a 3-D discrete cosine transform (DCT). The resulting dataset is quantized using an optimal Max quantizer. This irreversible coding is applied to MRI, fast x-ray CT, and PET image sequences. Its evaluation is performed using quantitative and observer independent methods taking into account the specificity of medical dynamic data.

A digital video-based system was designed and implemented to assess imaging workstation human-user interfaces through time and motion studies in diagnostic radiology and radiotherapy treatment planning. This system provides a means for recording and analyzing the activities which take place at imaging workstations during initial training and active clinical use in radiology. On time-synchronized and event-stamped video tapes, the system simultaneously records the soft copy display images and workstation environment.

A medical image workstation has been conceived and prototyped which is designed to act as a cooperative dialogue partner in a number of routinely performed tasks in diagnostic radiology. The system can automatically select relevant information (e.g., current and previous examinations, images, image sequences) and generate a meaningful and appropriate image arrangement on the display screen. This is shown to be an effective feature to simplify and speed up radiological image access and presentation. For many cases in diagnostic image reading, the users' interaction may be as simple as switching from one patient to the next. Furthermore, the installed mechanisms offer a solution for the automatic pre-fetching of images to avoid transmission delays in the course of diagnostic work sessions. The cooperative system response is based on explicit (formalized and computer-accessible) models of diagnostic information requirements. These models are context-dependent and take into account that diagnostic information needs vary with radiological work procedures, workstation users, and patient cases. Initial information requirement models have been acquired from expert radiologists in two European hospitals and were integrated in a cooperative workstation prototype. For the representation of models, rule-based and object-oriented techniques were applied. The rule base was designed with a distinct modular structure, separating between rule sets for general, task- and user-dependent information requirements. The paper reviews the objectives for the design of cooperative workstation user interfaces and describes the acquisition, structuring, formalization, and representation of context-dependent information requirement models. The rule-base is explained by examples. A layered workstation architecture consisting of model-, object-, and realtime-layer is presented. Difficulties in the implementation of cooperative workstations are discussed which point to future research topics and standardization activities.

Workstations are becoming more commonly used in medical environments, and are being used increasingly for viewing medical images. In most clinical environments, counter and wall space is not readily available, and there is a strong motivation to make the equipment small, while making the displayed images as large as possible to preserve image detail. This precludes the use of a separate text monitor for user interaction, and any menus or displays on the image monitor use valuable space -- pixels backed up with 256-shade grayscale capability. We have developed a method for user interaction which requires essentially no screen area for permanent menus, but uses much of the image screen for `invisible' menus -- menus which are in windows which are always open (active) but only obscure the underlying image for the small portion of time that they are actually in use. These invisible menus respond to movements of the mouse, and become visible when the mouse is moved into the window which holds the menu. The menu becomes invisible again after a period of mouse inactivity. Because these windows are always active, a given item may be selected multiple times by simply pressing the mouse button repeatedly. This `type-ahead' capability is not normally available on systems which do not include a keyboard, and may be easily used for common repetitive functions, analogous to pressing the NEXT IMAGE key multiple times. This invisible window concept can also be used to display analysis results, so that the results do not cover any of the active image area, but are immediately available for on-screen viewing.

Many functional, technical, and perceptual considerations must be met before a workstation will be adequate for primary diagnosis. The focus of this paper is the functional aspect of the workstation. Specifically, we are concerned with determining how images and data must be presented to the radiologist, for the purpose of primary diagnosis, under the constraints imposed by the digital workstation. We have developed an interface that is being used to acquire detailed information about the current diagnostic process. The purpose of this interface is twofold. First, this interface enables us to monitor the image and information access patterns of the radiologists in the process of interpreting films. This information is used to automate the presentation of images and information to the radiologist in future cases. Second, this interface provides a continuously evolving tool to capture the physical attributes, or navigational cues, necessary for the radiologist to develop a mental model of the operation of the diagnostic workstation. This report describes the current operation and future goals of this interface.

As hospitals geographically spread and radiologic services are required in remote locations, the radiologist increasingly must conduct a remote practice. Rapid image transmission from the remote site to the radiologist is important but only half the problem. First, the radiologist may need to view and discuss the images with the technologist to verify image quality or to specify the location of follow-up images. Second, the radiologist may need to discuss the case with another radiologist for a second opinion or for the advice of a sub-specialist. Third, and most importantly, the radiologist may need to discuss the case with the referring physician to better understand the text data, clinical history, and referring physician's clinical questions and concerns, and to better convey the location and extent of the clinical findings. In this paper we detail the requirements for a remote consultation workstation, present previous work on remote computer interaction, and describe the FilmPlane remote consultation workstation in detail. We then discuss the MICA medical communications project in which FilmPlane will be used for a remote consultation study between the UNC family medicine clinic and the main hospital 1/2 mile away.

Picture archival and communication (PACS) and teleradiology systems require workstations for image display, however not all areas demand the same functionality and performance. A comparison was made between the Vortech Personal Display System (PDS), the Dupont Clinical Review System (CRS), and the dual 2 K X 2.5 K Megascan Diagnostic Workstation (MDW) under development within the department.

The tasks of the radiologist include interpretation of images, communication of findings to referring physicians, and assistance in treatment planning. CT, MRI, SPECT, and PET are usually analyzed from film recorded images. On-screen analysis allows images to be quickly modified, measured, reprocessed by 2-D and 3-D methods, etc. These images are also filmed. An ideal clinical diagnostic radiology workstation must be flexible in design and provide a user interface which is consistent, intuitive and simple for every use of the system. Both efficient image analysis and filming should be addressed. This poster describes the implementation of such a workstation (CEMAX VIP) using a standard RISC Unix platform (SUN SPARCstation) and a modular software design.

Research into all aspects of PACS increased in the 1980s as advances in technology enabled key components of PACS to be implemented. The University of Washington has worked on many issues of PACS, such as workstation development, teleradiology, and PACS-RIS (Radiology Information System) interconnection. In the past, special-purpose workstations have had to be developed to support the high demands of medical imaging. However, recently developed general-purpose workstations are now being considered as potential components of a PACS. Our latest research, a joint UW/IBM study, has involved the development of a system based on the IBM RISC System 6000 (RS6000). Evaluations of radiology workstation user interfaces led to the conclusion that a Graphical User Interface is the most acceptable. Since a general-purpose workstation is employed, it was felt that the software should be as portable as possible. These two requirements led to the selection of the X Window System for software development. What is discussed in this paper is the development of image display and processing functions on the RS6000 in the X Window System environment, and how this development effort satisfies the requirements of a PACS workstation.

Integrated diagnostic support system (IDSS) is a system which manages not only text data but also medical images on the central database and provides doctors most medical care information on a single workstation. IDSS consists of host database management, a gateway, and workstation. Medical images are sent to the host from medical imaging modalities through a gateway program and are saved in the database. When images are required at workstations, those images are sent from the host to the workstation and saved in the workstation database. The workstation has functions to display images which are in local disks. Doctors can select single or multiple images to display by mouse operation. There are three database layers such as a host cash, a long term storage, and a local workstation. Medical images which are referred to frequently are held in the host cash but moved to the long term storage at a certain point. In order to see medical images on the workstation, the images have to be loaded into the local database by the host applications or a doctor's operations.

X-ray fluoroscopy is a significant source of x-ray dose to patients and hospital staff. One technique proposed for reducing dose is pulsed fluoroscopy at reduced frame rates, typically 15 frames/sec for cardiac angiography. Because the human visual system acts as a temporal low-pass filter, simply reducing the frame rate may not allow a dose reduction. In fact, one can argue that for equivalent visualization, the dose per frame should be doubled when the frame rate is halved. We address the question of proper dose by comparing detectability in simulated pulsed and continuous fluoroscopy displays on a unique device called the video tachistoscope. The visual task for the subjects is to detect stationary, computer-generated, low- contrast disks on a flat background in the presence of noise. The disks form a contrast-detail phantom and are displayed in continuous and pulsed mode on either side of a single video monitor. We find similar detectability and visibility in noise when the dose for pulsed is reduced by 25% of that for continuous.

Typical user interaction in image processing is with command line entries, pull-down menus, or text menu selections from a list, and as such is not generally graphical in nature. Although applying these interactive methods to construct more sophisticated algorithms from a series of simple image processing steps may be clear to engineers and programmers, it may not be clear to clinicians. A solution to this problem is to implement a visual programming language using visual representations to express image processing algorithms. Visual representations promote a more natural and rapid understanding of image processing algorithms by providing more visual insight into what the algorithms do than the interactive methods mentioned above can provide. Individuals accustomed to dealing with images will be more likely to understand an algorithm that is represented visually. This is especially true of referring physicians, such as surgeons in an intensive care unit. With the increasing acceptance of picture archiving and communications system (PACS) workstations and the trend toward increasing clinical use of image processing, referring physicians will need to learn more sophisticated concepts than simply image access and display. If the procedures that they perform commonly, such as window width and window level adjustment and image enhancement using unsharp masking, are depicted visually in an interactive environment, it will be easier for them to learn and apply these concepts. The software described in this paper is a visual programming language for imaging processing which has been implemented on the NeXT computer using NeXTstep user interface development tools and other tools in an object-oriented environment. The concept is based upon the description of a visual language titled `Visualization of Vision Algorithms' (VIVA). Iconic representations of simple image processing steps are placed into a workbench screen and connected together into a dataflow path by the user. As the user creates and edits a dataflow path, more complex algorithms can be built on the screen. Once the algorithm is built, it can be executed, its results can be reviewed, and operator parameters can be interactively adjusted until an optimized output is produced. The optimized algorithm can then be saved and added to the system as a new operator. This system has been evaluated as a graphical teaching tool for window width and window level adjustment, image enhancement using unsharp masking, and other techniques.

Multimodal PACS or high performance teleradiology systems require high throughput in order to be productive. Acquisition and availability of data and on-line storage capacity are significant factors for evaluating the acceptability of a PACS or teleradiology system in a clinical environment. Digital medical images contain an enormous amount of data. For example, a computed radiography image (2048 X 2048 X 10-bit) is composed of 8 megabytes (8-bit boundaries) of digital data. A CT examination consisting of 64 one-half megabyte images (512 X 512 X 12-bit) is composed of 32 megabytes (8-bit boundaries). The overall throughput of these systems is impacted by the protocol overhead, data collisions network media, and the I/O speed of the computer's disk system. This paper addresses I/O bandwidth, capacity, and data integrity of the disk storage system and specifically how parallel disk array implementations can be utilized in a PACS or teleradiology environment to improve throughput.

The recognized incidence of cutaneous malignant melanoma in the United States is now rising faster than any other cancer, increasing by 83% from 1980 to 1987. Recent revelations that depletion of the earth's ozone layer is accelerating at a more rapid rate than previously believed can only exacerbate current projections for the increased incidence of this deadly disease. Because there is no good treatment for metastatic melanoma even small cancers often prove fatal if not detected early. Melanoma allowed to invade the subcutaneous tissue is associated with a five-year survival rate of only 44%. Ironically, few cancers provide a greater opportunity for early discovery and cure. Cutaneous melanoma is not only located where it is readily observed, but typically undergoes a `radial growth' phase prior to metastasis. During this phase the net growth is superficial and circumferential, gradually increasing the area of the lesion and changing its coloration. Screening measures for the early detection of melanoma must concentrate on two primary tasks: (1) detection of lesion changes indicative of the radial growth stage of malignancy and (2) alerting the patient and physician to the existence of a new or changed lesion on the skin. To accomplish these goals we have experimented with the applicability of a microcomputer based video imaging system which stores an image archive of historical reference images for each patient. With the acquisition of new images of the patient, easily registered with the archival images through a technique we have developed we are able to perform a blink comparison of the image pairs. This technique appears to be far more effective than currently used techniques for detecting changed lesions on a comprehensive basis.

Modern dedicated image processing workstations and even general purpose computers offer enhanced user interface capabilities. Hardware management of the user interface allows a fast, easy, and powerful dialogue between man and machine. The application design must take into account these new possibilities in order to make optimal use of the hardware. Physicians are special users in that they need to customize their working environment to carry out specific tasks. Specific medical applications in the area of 3-D display and multimodality imaging need to accommodate a sequential organization of the physician's tasks, access to the various tools (image processing features, 3-D display, environment configuration, etc. ...) and the powerful dedicated workstations the physician may require. This paper sets out a number of general rules applicable to user interface design and defines the specific features of medical imaging brought into play in the definition of the environment we have developed for medical imaging user interface design. Examples in 2-D and 3-D display mode are presented.

UWGSP4 is configured with a parallel architecture for image processing and a pipelined architecture for computer graphics. The system's peak performance is 1,280 MFLOPS for image processing and over 200,000 Gouraud shaded 3-D polygons per second for graphics. The simulated sustained performance is about 50% of the peak performance in general image processing. Most of the 2-D image processing functions are efficiently vectorized and parallelized in UWGSP4. A performance of 770 MFLOPS in convolution and 440 MFLOPS in FFT is achieved. The real-time cine display, up to 32 frames of 1280 X 1024 pixels per second, is supported. In 3-D imaging, the update rate for the surface rendering is 10 frames of 20,000 polygons per second; the update rate for the volume rendering is 6 frames of 128 X 128 X 128 voxels per second. The system provides 1280 X 1024 X 32-bit double frame buffers and one 1280 X 1024 X 8-bit overlay buffer for supporting realistic animation, 24-bit true color, and text annotation. A 1280 X 1024- pixel, 66-Hz noninterlaced display screen with 1:1 aspect ratio can be windowed into the frame buffer for the display of any portion of the processed image or graphics.

In this study, we have explored two compression techniques, i.e., 3-D displacement estimated interframe coding algorithm (DEICA) and 2-D DCT algorithm, applied them in ultrasound image compression and assessed their feasibility. A sequence of 96 128 X 128 X 8- bit parallel slices and a set of 25 360 X 264 X 8-bit time-sequence images have been used in our experiments, and compression ratios in the range of 6 - 20 to 1 and 4 - 5 to 1 have been obtained in DEICA, while the mean square error was kept to less than 4.0. While DEICA has achieved higher compression ratios in x-ray CT images than the 2-D DCT algorithm at the same distortion, DEICA has not shown any improvement in compressing ultrasound images. This is mainly due to the fact that statistics within the difference image are sometimes degraded in ultrasound images. After fine-tuning the DEICA by adjusting key parameters based on the statistics of ultrasound images and reducing the speckle noise, higher compression ratios are expected. In the future, 3-D or time-sequence images will have more frame-by-frame correlation and less speckle noise patterns with the technical advances in ultrasound probes and data acquisition & image computing technologies. Furthermore, 3-D ultrasound imaging systems will become more popular and the amount of data generated by these devices will increase even further. These new developments might make DEICA more attractive.

This paper discusses conventional 3-D reconstruction for bone visualization and presents a case study to demonstrate the dangers of performing 3-D reconstructions without careful selection of the bone threshold. The visualization of midface bone lesions directly from axial CT images is difficult because of the complex anatomic relationships. Three-dimensional reconstructions made from the CT to provide graphic images showing lesions in relation to adjacent facial bones. Most commercially available 3-D image reconstruction requires that the radiologist or technologist identify a threshold image intensity value that can be used to distinguish bone from other tissues. Much has been made of the many disadvantages of this technique, but it continues as the predominant method in producing 3-D pictures for clinical use. This paper is intended to provide a clear demonstration for the physician of the caveats that should accompany 3-D reconstructions. We present a case of recurrent odontogenic keratocyst in the anterior maxilla where the 3-D reconstructions, made with different bone thresholds (windows), are compared to the resected specimen. A DMI 3200 computer was used to convert the scan data from a GE 9800 CT into a 3-D shaded surface image. Threshold values were assigned to (1) generate the most clinically pleasing image, (2) produce maximum theoretical fidelity (using the midpoint image intensity between average cortical bone and average soft tissue), and (3) cover stepped threshold intensities between these two methods. We compared the computer lesions with the resected specimen and noted measurement errors of up to 44 percent introduced by inappropriate bone threshold levels. We suggest clinically applicable standardization techniques in the 3-D reconstruction as well as cautionary language that should accompany the 3-D images.

In digital radiographs with significant collimation, there is a noticeable amount of unexposed background remaining in the image. This unexposed portion appears as bright white on the video display. We have developed a method that automatically removes this unexposed background and converts it to black background. This is less distracting to the viewer when the image is displayed on a video monitor. The algorithm first searches for the raw edges of an image. The next stage is to reconcile these edges to find the corner points. From these corner points, we interpolate to obtain straight line boundaries. The estimated boundaries are then checked with an error-correction procedure to verify that these points are outside the image. The error-corrected contour is then drawn using smooth line techniques and setting points outside the contour to black. This method has been implemented in a PACS environment and works well for images with significant collimation (i.e., pediatric cases and hand images). The amount of distracting unexposed background has been noticeably reduced. In a survey of CR images needing background removal, 89.5% of all images had all or part of the background removed and 10.5% had no background removed.

A cost-effective display system has been developed and incorporated into a low-cost magnetic resonance imager. A single graphics board (Peritek VCD) with an on-board 68030 CPU, ACRTC graphics controller and 1024 X 1024 12 bit video RAM is interfaced to a DEC Micro-VAX II host computer. All memory and devices on the graphics board are designed to be tri-ported so that they may be accessed by all three processing engines: 68030, VAX, and ACRTC with minimum arbitration delay. The software is designed to time share between the processing engines to provide maximum throughput without overloading the host CPU. Using the on-board hardware zoom of 2, the available video RAM can be partitioned to allow 16 frame movie loops to be produced. The ACRTC graphics controller allows other video modes such as PAL and NTSC to be selected by software control. This allows direct video taping without the expense and resolution degradation of scan convertors or secondary filming off the monitor screen. A direct pathway off the graphics board to a digital laser camera is also provided. This system is also applicable to workstation displays where it would not be the primary display but could be used for the tasks which workstations can not do such as interfacing to analog multiformat cameras, videotape recorders, and digital laser cameras.

The radiology department of Jackson Memorial Hospital processes 255,000 clinical examinations each year -- 65,000 of which are portable x rays. Film transportation and loss are major obstacles to the smooth operation of this department. To assist in the solution of these problems we have designed and begun the piecemeal installation of a clinical PACS. This system is based on a platform of IBM RISC/6000 computers and software developed by Genesys Corporation. The initial installation involved the digitization of the portable x rays from three ICUs. The images (in the form of a matrix of 2048 X 1648 pixels) are then entered into the network and can be viewed simultaneously in the radiology department and in the ICU. The second phase of installation, involving the images from two CT scanners and two MRI scanners is currently underway. We have evaluated the system from several standpoints. The first is user acceptance. The users are the radiologists who must make the diagnosis at the workstation and the referring physicians who need the diagnosis quickly but also require the image. The radiologists must be comfortable with their diagnosis based on the images presented at the two viewer workstation. This is compared to the use of a multiviewer which presents many radiographs simultaneously. The second parameter for evaluation involves the impact on patient care in terms of the time elapsed between the taking of the radiograph and the presentation to the physician of the image and the diagnosis.

We present a method for recursively rendering a realistic image of a volumetric dataset consisting of a mixture of sampled and synthetic objects. We describe several volume visualization tools that are based on the use of recursive ray tracing. These include shadows, mirrors, specularity, and constructive solid geometry. We discuss several ways to enhance our method both in terms of image quality and rendering speed by introducing the principles of adaptive traversal and probabilistic rendering.

We have analyzed the imaging characteristics of a cathode-ray tube multiformat printer. These include the dependence of gray scale response on the printer's settings, the modulation transfer function of the printer, the acutance (or sharpness) and the noise power spectrum of the printed images. A comparison of theoretical results and experimental data is presented.

Image compression is essential to handle a large volume of digital images including CT, MR, CR, and digitized films in a digital radiology operation. The full-frame bit allocation using the cosine transform technique developed during the last few years has been proven to be an excellent irreversible image compression method. This paper describes the effect of using the hardware compression module on diagnostic accuracy in hand radiographs with subperiosteal resorption and chest radiographs with interstitial disease. Receiver operating characteristic analysis using 71 hand radiographs and 52 chest radiographs with five observers each demonstrates that there is no statistical significant difference in diagnostic accuracy between the original films and the compressed images with a compression ratio as high as 20:1.

Digital techniques are used more often than ever in a variety of fields. Medical information management is one of the largest digital technology applications. It is desirable to have both a large data storage resource and extremely fast data transmission channels for communication. On the other hand, it is also essential to compress these data into an efficient form for storage and transmission. A variety of data compression techniques have been developed to tackle a diversity of situations. A digital value decomposition method using a splitting and remapping method has recently been proposed for image data compression. This method attempts to employ an error-free compression for one part of the digital value containing highly significant value and uses another method for the second part of the digital value. We have reported that the effect of this method is substantial for the vector quantization and other spatial encoding techniques. In conjunction with DCT type coding, however, the splitting method only showed a limited improvement when compared to the nonsplitting method. With the latter approach, we used a nonoptimized method for the images possessing only the top three-most-significant- bit value (3MSBV) and produced a compression ratio of approximately 10:1. Since the 3MSB images are highly correlated and the same values tend to aggregate together, the use of area or contour coding was investigated. In our experiment, we obtained an average error-free compression ratio of 30:1 and 12:1 for 3MSB and 4MSB images, respectively, with the alternate value contour coding. With this technique, we clearly verified that the splitting method is superior to the nonsplitting method for finely digitized radiographs.

A software module has been created in the biomedical image analysis and display package ANALYZE' which enables experimentation with 2-D and 3-D image compression based on wavelet transforms. In particular, this module allows the user to interactively determine the desired amount of compression (by viewing the results) and to interactively define specific subregions of the image that are to be preserved with full fidelity during the compression (again, viewing the results). Examples of the tool's operation and results on a variety of medical images will be presented.