We describe superfaces, a new method for simplifying closed polyhedra. The superfaces algorithm performs the simplification based on a bounded approximation criterion that produces a simplified polyhedron that approximates the original one to within a prespecified tolerance. The vertices in the simplified polyhedron are a proper subset of the original vertices, so the algorithm is well-suited for creating hierarchical representations of polyhedra.

In this paper, we present a new approach for the interpolation of grey data of arbitrary dimension that generalizes the shape-based method from binary data to grey data. The basic idea here is to express the given n-dimensional image as a (closed) surface in the (n + 1)- dimensional space, to interpolate the surface based on its shape, and then to collapse the new surface back to the image form. This method is more general and flexible in many respects than other methods. In addition to being able to handle data of arbitrary dimension, it allows intermixing operations on structures and images. The traditional shape-based interpolation method becomes a particular case of this new methodology. Our preliminary observation is that in regions of smooth as well as sharp intensity changes, the new method (in its simplest form) performs better than linear grey-level interpolation. In regions of scattered, fractal-like structures its performance seems to be inferior to the linear method.

In order to enable the interaction with and manipulation of 3-D data sets in the realm of medical diagnosis and therapy planning we developed a modified Z-merging algorithm that includes transparency and texture mapping features. For this an extended shape based interpolation model creates isotropic grayscale data volume in case of spatial image sequences. Interesting anatomical regions such as soft tissue, organs, and bones are detected by automatic and interactive segmentation procedures. Following that, a fully automatic surface construction algorithm detects the 3-D object boundaries by fitting geometric primitives to the binary data. The surface representations support the user with a fast overview about the structure of the 3D scene. Texture mapping is implemented as the projection of the gray values of the isotropic voxels onto a polygonal surface. Adaptive refinement, Phong's normal interpolation, and transparency are the most important features of this raytracer. The described technique enables the simultaneous display of multimodal 3D image data.

Volume rendering uses classification based on fuzzy segmentation, and the key to making meaningful images lies in a proper choice of the opacity and color transfer functions. The smooth and continuous transfer functions introduce less artifacts than any binary operations such as thresholding which disrupts the continuity of the data. Despite this advantage, density based fuzzy segmentation still has several limitations. To find suitable transfer functions for real clinical data may be a laborious task, and methods facilitating the automated generation of transfer functions are very useful. Furthermore, the standard transfer functions are based on the density of a resampled point. This results in several shortcomings. To overcome those limitations, we propose a suitable extension of the standard transfer functions called generalized transfer functions. These functions use both density based as well as non-density based information about classified structures. We show the usefulness of the generalized transfer functions in 3D neuroimaging (neuropathology) from MRI data. Three approaches are discussed: contour-enhanced volume rendering, ROIs-enhanced volume rendering, and slice density corrected transfer functions.

A computer software package named BDREC was designed and implemented in order to store and retrieve registration data information and to have some registration tools available. The aim is to facilitate multimodal application development by managing all geometrical issues.

Three-dimensional-VIEWNIX is a data-, machine-, and application-independent software system, developed and maintained on an ongoing basis by the Medical Imaging Processing Group. It is aimed at serving the needs of biomedical visualization researchers as well as biomedical end users. Three-dimensional-VIEWNIX is not designed around a fixed methodology or a set of methods packaged in a fixed fashion for a fixed application. Instead, we have identified and incorporated in 3DVIEWNIX a set of basic imaging transforms that are required in most visualization, manipulation, and analysis methods. In addition to visualization, it incorporates a variety of multidimensional structure manipulation and analysis methods. We have tried to make its design as much as possible image-dimensionality- independent to make it just as convenient to process 2D and 3D data as it is to process 4D data. It is distributed with source code in an open fashion. A single source code version is installed on a variety of computing platforms. It is currently in use worldwide.

Arterio-venous malformations (AVMs) are a congenital disorder that affects a small percentage of the population. They are treated by blocking or reducing the blood supply followed by surgery. This paper looks in a preliminary way at visualizing the cerebral vasculature and ultimately the AVMs. These visualizations provide support for the surgeons and radiologists. Our concern is to substantiate the point that there are deficiencies in the data correctable with reference to digital subtraction angiograms and we conjecture that knowledge based processing of this data may lead to improved results. The paper explores the basis of the difficulty and it compares the performance of several algorithms. Simple geometric objects are studied and the dependence of error on several parameters is shown. A comparison is drawn between the richness of the data available from x-ray angiograms (XRAs) and magnetic resonance angiograms (MRAs). Inferences are drawn on approaches that may be appropriate for the evolution of a description of the vasculature. Comment is also made on the way in which different representations may be compared.

The aim of radiotherapy in the treatment of cancer is to destroy the tumor or inhibit its growth by means of radiation. Multiple beams are often used to spare healthy regions, concentrating the radiation to the extent of the tumor. Critical structures cannot receive large amounts of radiation, and thus they must be easily recognized within the data to allow for a more efficient planning. The design of radiation bemas must then be visualized so that the plan can be evaluated. It is also very important to visualize radiation doses within the data when evaluating a treatment plan. This paper gives an overview of techniques developed to address some of these issues.

The utility of 3D medical images has been limited by difficulties in analyzing and making measurements on the images. To tackle these problems, interactive workstation-based systems have been devised for visualizing and quantitating structures in 3D images. Unfortunately, these systems generally demand time-consuming, subjective, error-prone human interaction. To reduce the disadvantages of purely interactive techniques, some recent efforts have combined automatic image analysis with human interaction. These efforts demonstrate that for complex 3D medical applications, a judicious combination of human interaction and automatic computer-based processing is essential. We describe a system called INTERSEG (INTERactive SEGmentation) that combines human interaction and automatic processing for 3D radiological image analysis. INTERSEG is a graphical user interface system that can be used to: (1) construct interactively defined region cues; (2) invoke automatic image analysis; and (3) peruse/visualize analyzed images. The cues, which are easy to construct, convey problem-specific information and define the image-analysis task. The paper focuses on INTERSEG's capabilities for human interaction and visualization.

The 3D reconstruction of an opacified vasculature from a set of subtracted 2D x-ray projections is an ill-posed problem for which prior information must be taken into account in order to stabilize the solution. The high contrast and sparseness characteristics of opacified blood vessels may be used through the introduction of a region of support (ROS) of the object to be reconstructed. This paper compares different techniques to build and use an ROS in 3D vascular reconstruction. The ROS is obtained either by segmenting the 2D projections or by segmenting in 3D a coarse estimate of the reconstructed object. The conclusion is that the use of an ROS does not improve the quality of the reconstruction, although it does dramatically reduce its computational requirements. Another conclusion is that the 3D segmentation approach seems to be more robust than the 2D segmentation one.

Stereotactic radiosurgery is the precise application of high dose radiation to cranial lesions. A combination of isocenters or `shots' is utilized by the physician to map the contours of the target lesion. This pattern of shots results in a high dose of radiation to the lesion while delivering a lesser, non-lethal dose to healthy tissue elsewhere in the brain. In an effort to improve the efficacy and efficiency of stereotactic treatment planning, a method for the automatic correlation of the intrinsic data sets has been developed. Individual coordinate transforms for each of the data sets are calculated using a coordinate system provided by a stereotactic frame fixed to the patient's skull. Each data set is then correlated within this common coordinate system. Through the use of multiple, cascaded coordinate transforms the accuracy of the correlated display is maximized. Each data set is displayed on its own graphical layer allowing the physician to visualize any combination of the individual data sets.

We describe a 3-D user interface for preoperative neurosurgical planning based on the physical manipulation of familiar real-world objects in free space. Using these passive interface props, neurosurgeons can apply their existing skills to specify spatial relationships in a natural and direct manner. The interface currently employs a head viewing prop, a cutting- plane selection prop, and a trajectory selection prop. Each prop is a simple real-world tool, the position and orientation of which is tracked by the computer. The behaviors associated with each prop serve as `interaction primitives' which can be composited to describe complex spatial relationships, resulting in a powerful, expressive, and conceptually simpler user interface. From the surgeon's perspective, the interface is analogous to holding a miniature skull which can be `sliced' and `pointed to' using the cutting-plane and trajectory props. Our informal evaluation sessions have shown that with a cursory introduction, neurosurgeons who have never seen our interface can understand and use it without training.

Interactive, image-guided (IIG) surgery requires three processes or devices: a 3-D spatial localizer, a technique for mapping physical space into image space, and the display of present surgical position onto medical images. A number of types of 3-D localizers have been tried including articulated arms, magnetic field sensors, and ultrasonic time-of-flight devices. Optical localization techniques have been suggested in the past but rejected due to problems with angular sensitivity, system size, and data rates. In this paper we present an optical localization device for use in neurosurgery that addresses those problems and provides an accurate, easy to use localizer that, when coupled to the IIG software and image display, allows for enhanced surgical guidance.

This paper describes an interactive 3-D display system for supporting image-guided surgery. Different from conventional CRT-based medical display systems, this one can provide true 3- D images of the patient's anatomical structures in a physical 3-D space. Furthermore, various tools for view control, target definition, and simple treatment simulation, have been developed and can be used for directly manipulating these images. This feature is very useful for a surgeon to intuitively recognize the precise position of a lesion and other structures and to plan a more accurate treatment. The hardware system is composed of a volume scanning 3-D display for 3-D real image presentation, a 3-D wireless mouse for direct manipulation in a 3-D space, and a workstation for the data control of these devices. The software is for analyzing X-CT, MRI, or SPECT images and for organizing the tools for treatment planning. The system is currently aimed at being used for stereotactic neurosurgical operations.

The stereotactic procedures allow, with the aid of a reference frame supporting an instrument holder, us to define a trajectory to access to a target in a coordinate system recognized by the imaging systems and corresponding to the frame coordinate system. In the field of stereotactic research, a number of major problems in Medical Imaging are encountered: 3-D imaging, multimodal data fusion, 3-D segmentation, PACS, etc. These problems resolved, the design of a computerized application allowing the neurosurgical stereotactic act simulation has its own problems: development of man/computer interface, 2-D and 3-D display tools, validation, integration of computerized application into the clinical environment, etc.

We demonstrate the use of multi-modality and 3-D stereoscopic imaging in the context of image-guided neurosurgery. We consider here the integration of anatomical data (MRI), vascular data (DSA), and functional data (PET) derived from the same patient. Our workstation, which has a stereoscopic 3-D image display, is interfaced to a hand-held probe whose position coordinates in real space are constantly relayed to the computer during the procedure. This enables the probe to be visualized in images relating to individual or combined modalities during the surgical procedure. The integration of multi-modality data in this manner provides the surgeon with a comprehensive overview of brain structures on which he is performing surgery, or through which he is passing probes or cannulas, enabling critical vessels and/or structures of functional significance to be avoided.

Computed radiography (CR) is a relatively new technique for projection radiography. Few hospitals have CR devices in routine service and only a handful have more than one CR unit. As such, the clinical knowledge base does not yet exist to establish quality control (QC) procedures for CR devices. Without assurance that CR systems are operating within nominal limits, efforts to optimize CR performance are limited in value. A complete CR system includes detector plates that vary in response, cassettes, an electro-optical system for developing the image, computer algorithms for processing the raw image, and a hard copy output device. All of these subsystems are subject to variations in performance that can degrade image quality. Using CR manufacturer documentation, we have defined acceptance protocols for two different Fuji CR devices, the FCR 7000 and the AC1+, and have applied these tests to ten individual machines. We have begun to establish baseline performance measures and to determine measurement frequencies. CR QC is only one component of the overall quality control for totally digital radiology departments.

By the middle of 1994, most of the radiology diagnoses at Madigan Army Medical Center will be performed on diagnostic workstations rather than from hardcopy films. Consistent, high- quality images must be produced by the different monitors to help support accurate diagnoses and consultation. We have applied several quality control tests to various types of monitors attached to a clinically-functioning medical diagnostic imaging support (MDIS or PACS) system at Madigan to ensure the high quality of diagnostic image display. Many characteristics of a monitor can degrade over time, including luminance, spatial uniformity, and spatial resolution. Tests to measure these characteristics are explained and results are shown. A photometer and the SMPTE test pattern are used in many of the tests. The results of different monitors on the same workstation are compared as well as the results from the same monitor over time. Technicians from the PACS vendor adjust the maximum brightness of the monitors periodically at Madigan so that they meet the specification. Our data shows that this adjustment alone is not sufficient to ensure that the other characteristics of a monitor are within acceptable values.

Through the Medical Diagnostic Imaging Support (MDIS) Program, the U.S. military has installed picture archiving and communication systems and teleradiology systems in three medical centers. The primary image acquisition modality for these systems is computed radiography. This presentation reviews the distribution of downtime for the computed radiography devices at the three PACS based hospitals and at one non-PACS based medical center. Data on both the Fuji AC1+ and the Fuji Digiscan 7000 is included. Reported downtime is divided into four categories; user induced, hardware, software, and preventive maintenance. Analysis of the distributions of these four categories in the context of the equipment's operational environment points out the impact of workload, interface complexity, preventive maintenance, and human factors upon both the number of equipment failures and the reason for the failures.

Distributing a radiographic image with the verbal report would help to avoid misunderstanding and would aid the referring physician in understanding better the severity of the disease described. With magnetic resonance imaging it is customary in our practice to distribute a copy of the examination to the referring physician. Distributing a film copy of the image is expensive. Based on our initial experience with the Scitex paper printer, we believe that this system provides a paper image of sufficient quality that it would be acceptable as a referrer's copy. Paper copies are cheaper to produce and can be more easily filed with the remainder of the patient's paper based medical record. This presentation discusses the printing method employed by the Scitex printer, demonstrates comparisons between Scitex and laser print images, and discusses the current problems of interfacing the printer to image acquisition devices. As we develop our image management and communication system (IMAC), we anticipate that a need for hard copy images will remain. We discuss the role that we believe this paper printer serves in an IMAC film independent system.

A Du Pont Linx computed radiography (CR) system and associated radiology image viewing station (RIVS) were evaluated with respect to CR image data transfer and image quality performance. Image transfers were compared with conventional film digitizers and CR film processing. Image quality was determined using limiting spatial resolution and low contrast detectability. Processing times differed for CR film output and softcopy display on the RIVS station, ranging between 2 and 10 minutes depending on whether single or batch processing was being used. In general, images from the CR were available on the RIVS display within 100 to 160 seconds compared with a minimum of 270 seconds for producing CR film images. Low contrast detectability performance for CR films was strongly influenced by choice of CR image processing algorithm. In general, there was very little difference between CR film and the corresponding images displayed on a RIVS. Digitization of film and subsequent review on a RIVS, however, generally showed a marked deterioration in low contrast detectability performance.

The perceptual linearization of video display monitors plays a significant role in medical image presentation. First, it allows the maximum transfer of information to the human observer. Second, for an image to be perceived as similarly as possible when seen on different displays, the two displays must be standardized. Third, perceptual linearization allows us to calculate the perceived dynamic range of the display device, which allows comparison of the maximum inherent contrast resolution of different devices. This paper provides insight into the process of perceptual linearization by decomposing it into the digital driving level to monitor luminance relationship, the monitor luminance to human brightness perception relationship, and the construction of a linearization function derived from these two relationships. We compare and contrast the results of previous work with recent experiments in our laboratory and related work in vision and computer science. Based on these analyses we give recommendations for using existing methods when appropriate, and propose new methods or suggest additional work where the current methods fall sort. Finally, we summarize the significant issues from all three component areas.

The use of standards in hardware and software development has long been the subject for debate within medical-imaging companies. Adhering to standards in hardware often yields lower cost and more flexible and easily upgraded systems. Software standards reduce engineering cost and speed the introduction of new products, while providing more consistent and well-understood user interface environments. Unfortunately, within medical imaging, the advantages of standards have been much less clear. The special requirements of medical imaging have often overwhelmed both hardware and software standards. Custom system development was frequently the only way to produce adequately functional workstations. Now, as the industry moves towards broadly available imaging workstations, it is important to review the advantages of standards. If available standards can be made to fit the needs of the medical-imaging community, it is appropriate and important for medical-imaging companies to begin to provide systems based on industry-wide standards. We argue that the best approach is a balanced, semi-custom system that provides high performance with as much compliance with hardware and software standards as possible without compromising system efficacy.

This paper discusses the use of a novel model of neural networks, the generalized neural network model, to build the primitives for an adaptive compression system. This model adds to the today's connectionist model paradigms to include the behave-act, evolve-learn, and behave-control functions of neural networks, which allow the definition of connectionist systems that overcome the drawbacks of previous feedforward neural network-based compression systems. The approach yields a compression system that surpasses known compression algorithms in three main aspects: very high compression rate with a low introduced distortion, ability to tackle a broad set of data, and feasibility for on-line real-time compression.

The potential of overlapped transform coding for data compression of medical x-ray image series has been investigated and a comparison with conventional block-based transform coding was made. We found that overlapped transform coding clearly produces less blocking artifacts than conventional block-based transform coding at compression ratios in the range 8 - 16.

Since most medical images are composed of different image characteristics, it has been demonstrated that applying a combined method to various image components can preserve a higher image quality. The major image components are: (a) smooth areas, (b) sharp edges, (c) texture, and (d) noise. In practice, sharp edges and general textures are two main components to be concerned for radiological images compression. A unified perspective of transform coding is reviewed to find out how a high compression ratio can be achieved with a lossy compression technique. Theoretically, the image quality in resolution power is associated with a composed module transfer function (MTF) when an image obtained from x-ray device coupling with a digitization module and processed by a lossy compression. It is very difficult to use a global MTF (or a band of MTF) to represent such a system. In this paper, we concentrate on clinical considerations for various applications in radiological image compression. Three different applications and associated compression strategies are discussed. Based on these compression strategies, we believe that many compression methods are suitable for clinical implementation with some clinical guidance and technical modifications.

This work concerns the compression of x-ray images of the breast using an original coding method based on a 2D wavelet transform. A tree-structured analysis/reconstruction system is used with Daubechies wavelet filters in order to decompose the original image into subbands; then we code only the low resolution subimage with the JPEG algorithm. This hybrid method has been compared with the JPEG approach, and results are proposed for different compression ratio.

This paper reports on a lossy compression method applicable to cardiac angiography. Full frame DCT coding has been investigated, using an optimized bit allocation and quantization scheme. We compared it to the standard JPEG method in the environment of a cardiac angiography system with dedicated visualization devices and post-processing. At a compression ratio 12:1, the image quality appeared to be better than the JPEG base-line compression. Owing to the principle of our method, no blocking effect is induced, whereas this is a critical drawback of the JPEG algorithm. Furthermore, the sharpness of fine details is better preserved.

We have developed a compression algorithm that achieves high compression ratio and excellent reconstruction quality for video rate or sub-video rate angiograms. Commercially available technology such as JPEG and MPEG do not satisfy medical requirements due to their severe block artifacts. Our new method takes advantage of the high temporal correlation of the angiographic frames using MPEG motion estimation software, but performs the error frame encoding using wavelet transform which is much more sensitive to localized block artifact features than the conventional discrete cosine transform. Our results show that at ratios up to 20:1, excellent image quality is obtained in reconstructed angiograms. Because of the excellent energy gathering and spatial localization properties of wavelet transform, high image fidelity and compression ratio can be simultaneously achieved. The algorithm is parallelizable and therefore can be implemented relatively easily in real time hardware.

A Test Director's Workstation has been developed to collect radiological exams in a PACS for studies. While the efficacy of PACS systems is being accepted in the radiological community, there remain many questions about the performance of a PACS to be answered. Some of the answers can be supplied by studies comparing the performance of radiologists using the PACS under different circumstances. A workstation that is useful in collecting images from the PACS at Madigan Army Medical Center for studies has been developed. The workstation retrieves the exam from the PACS, writes the exam to a local optical disk for permanent retention, and collects information about the exam in a Test Director's master exam file. In the process of collecting the exam the patient identification is stripped from the images to preserve the patient's privacy. The identification of the patient is retained in the Test Director's master file in order to follow the progress of the patient and determine the `truth' of any diagnosis. The Test Director's Workstation completes its task by helping the test director organize selected exams for reader's disks and by setting up a reader's worklist of exams to be read.

The medical diagnostic imaging support workstation has been in clinical use at selected military medical centers since March 1992. The workstation is a critical component in picture archiving and communications systems representing the interface between the system and the end user. The workstation has undergone several software changes over the last year based on feedback from end users. The present performance of the workstation in terms of image manipulation and navigation, response time, database, and reliability is emphasized. Discussion includes clinical acceptance, lessons learned, and future enhancements.

A very common task for a radiologist is to compare a previous examination with a current one. With film, the typical method varies somewhat, but usually involves putting up comparable views on two adjacent lightboxes. The reader then looks back and forth between the two films to assess changes. For images that can be placed on adjacent lightboxes, this method works reasonably well, but if images are placed several lightboxes apart, the task becomes difficult. A method that may help improve the comparison operation is possible with a workstation, but difficult (at best) with film. That is, to present the images to be compared as if they were on top of each other. The reader then uses a cursor to move a reveal bar or region of interest that shows the images `underneath.' This application has some potential for simplifying the comparison task for the radiologist. It avoids the extremes of head movement, and may keep the area of interest nearly coincident.

The UWGSP5, also known as MediaStation 5000, is the latest entry in a series of high- performance medical imaging workstations developed at the University of Washington Image Computing Systems Laboratory. In addition to its image processing and graphics performance, the UWGSP5 has multimedia capabilities that set it apart from traditional medical workstations. It supports continuous streams of video and audio data for real-time acquisition and playback, and implements real-time compression and decompression algorithms including the MPEG, JPEG, and P X 64 standards. Also, other compression algorithms optimized for medical images can be easily implemented on the UWGSP5. The fast decompression capability also allows for rapid reconstruction and display of stored compressed images, making the UWGSP5 suitable for a medical consultation workstation as well as a specialty image processing and graphics workstation. In this paper, we present the hardware architecture and software organization of the UWGSP5 and its performance on various image display, processing, and compression algorithms.

Accessibility of multimedia information related to radiology in a timely manner is a key to success in practicing radiology in the future. In this paper we describe the concept of multimedia in the radiology environment and its implementation in our department at UCSF. This paper emphasizes the various types of databases related to radiology including HIS, RIS, PACS image database, digital voice dictation system, electronic mail and library information system. A method to interconnect these databases is through a comprehensive network architecture that also is described. As an application, we introduce the concept of a departmental image file server, for any of the 150 Macintosh users in the department to access this multimedia information.

To make images and data available to all the radiologists, fellows, and residents at the University of Florida an electronic teaching file was required that was accessible to a wide range of people in a number of locations and could display not only data about cases, but the relevant images to go with the case. The multimedia electronic teaching system (METS) uses Oracle as its database management system with addresses of images stored in the tables. To avoid tedious data entry, a connection to the medical center's radiology information system was made. Findings, diagnosis and ACR codes are entered at the data entry screen along with the image numbers of the relevant images for CT, MRI, and US. These images are retrieved from the PACS system for storage on the METS disks. The user interface to the retrieval and display operation allows a search of any or all of the fields in the database through a query by example facility. Simple image manipulation is possible. A Sun workstation was used to implement the system. The user interface was initially written using the MOTIF window manager that could be run over the network on any other Sun workstation. An upgrade uses a MOSAIC with forms to implement data entry and query facilities.

Three dimensional medical scanners are widely available in today's hospitals to acquire a dataset of the human body without the need for surgery. The usefulness of this diagnostic information is limited by the lack of techniques to visualize the datasets. With the increasing computer power of today's workstations it is possible to make a transparent view of the 3D dataset. An interactive mode is necessary, however, to fully explore the 3D dataset. If both a high resolution and a high interactive speed is required, the necessary computational power is enormous. Therefore it is necessary to map the algorithms for volume visualization in a rather specific way onto (dedicated) chips to overcome the performance gap. This paper discusses a high-performance special-purpose low-power system, the Real-Time Volume Rendering Engine (RT-VRE), capable of rendering a 3D dataset of 256³ voxels onto a display of 750² pixels with an interaction rate of 25 images per second. The RT-VRE allows biomedical engineers to interactively visualize and investigate their data.

Most texture mapping involves a backward projection which maps from screen space back into the texture data. Bilinear/trilinear interpolation is usually used to sample data in the 2D/3D textures. To accelerate texture mapping, traditional graphics hardware requires a complete copy of the texture image at each parallel computation node to allow all nodes to operate in parallel. This paper introduces a new backward projection approach for rendering volume data, treating the volume as a 3D texture. It employs an innovative parallelization scheme in which a computation node operates on a subvolume, thereby avoiding the need to store the whole volume at each node. This approach has been implemented on the Kubota Kenai Denali workstation, a high performance system for imaging and 3D graphics. It currently supports functions for maximum intensity projection, multiplanar reformatting, volume resampling, ray sum, and isosurface rendering. Progressive refinement schemes can be utilized to reduce sampling rate and to realize interactive rendering speeds.

Neuroradiologists require rapid access to large multislice, multisequence image datasets, as well as tools for planning interventional procedures and predicting the probable outcome of an intervention. To address these challenges, we are integrating virtual reality (VR) technology with a large-scale picture archiving and communication system (PACS). The VR-based applications that we are developing include a `virtual view box' for rapid browsing and pairwise image comparison of large datasets using a head-mounted display (HMD), and `virtual intervention' tools for planning, simulating, and executing interventional radiology procedures. VR in radiology offers novel, `hands-free' techniques for image navigation, simulation, and teleintervention.

The full frame discrete cosine transform (FFDCT) compression method has the capability to eliminate block artifacts that exist in JPEG and MPEG compressed images. Previous FFDCT algorithm used an adaptive bit allocation (BA) method to pack quantized DCT coefficients. In this work, we replaced BA with arithmetic coding (AC) to improve compression performance markedly. We also show that by partitioning the quantized cosine domain much higher arithmetic coding efficiency can be achieved, especially in highly quantized images.

A comparative engineering performance evaluation was performed on video colonoscope systems from all three of the current U.S. suppliers: Fujinon, Olympus, and Pentax. Video system test methods, results, and conclusions based on their clinical significance are the focus of this paper.

Acquiring axial, coronal, sagittal data, T1-weighted, T2-weighted, PD-weighted, and T1 contrast enhanced images, several studies of the same patient can be collected. Usually all these studies are analyzed and processed as individual datasets. However, their spatial combination can improve the quality of volume rendered images and the accuracy of mensuration. This paper addresses the combination of contours (used to delineate a tumor) and the combination of volumetric MRI data. We have observed a substantial variability of the tumor volume calculated either from axial, coronal, or sagittal contours, despite the tumor delineation in all three directions being done by the same human expert. The volume error depends highly on the number of slices, slice thickness, and gap. The spatial combination of contours allows the following: finding the most probable tumorous region, determining the maximum extent of the tumor, and detecting inconsistent contour data. Combination of volumetric MRI data increases the quality of volume rendered images. For a given direction, data is interpolated using the gray-scale interpolation along with the shape information from the orthogonal datasets.

A hierarchical representation processing scheme has been proposed in this paper, in order to provide a framework for the anatomical/pathological object definition in medical imaging workstations. Preliminary experimental results show that the representation processing hierarchy can be used as a basic medical imaging representation tool to improve the efficiency of image processing in the workstation environment.

Image processing and visualization are very important in the medical field. This is clearly demonstrated by the wide variety of imaging schemes used such as magnetic resonance imaging, ultrasound imaging, x-ray imaging (computed tomography, CT), single photon emission tomography, and positron emission tomography. All of these modalities produce 2-D images. It is evident that image processing and visualization routines could expedite research in the field. In fact, if routines that already exist could be reused, it would further assist the progress. Environments that provide all the functionality required are not available (aside from the fact that all the functionality might not be understood, yet). A visual programming environment (VPE) tailored to the integration of all needed functions, through the combination of the best features of different packages into a single unified interface, was the main goal to be accomplished by the design of VMIPVS. Important issues addressed by VMIPVS are interoperability, data conversion, display formats, and extendibility.

We have developed a compression algorithm based on discrete wavelet transform (DWT) and arithmetic coding (AC) that satisfies the requirements of a radiological image archive. This new method is far superior to the previously developed full frame DCT (FFDCT) method as well as the industrial standard JPEG. Since DWT is localized in both spatial and scale domains, the error due to quantization of coefficients does not propagate throughout the reconstructed picture as in FFDCT. Since it is a global transformation, it does not suffer the limitation of block transform methods like JPEG. The severity of error as measured by NMSE and maximum difference increases very slowly with compression ratio compared to FFDCT. Normalized nearest neighbor difference (NNND), which is a measure of blockiness, stays approximately constant, while JPEG's NNND increases rapidly with compression ratio. Furthermore, DWT has an efficient FIR implementation which can be put in parallel hardware. DWT also offers total flexibility in the image format; the size of the image does not have to be a power of two as in the case of FFDCT.

Due to the increasing number of medical imaging modalities, the integration capacity of a PACS is necessary. Indeed, the PACS will lead the way in the introduction of advanced technology in the hospital. Among the elements that compose a PACS the elaboration of a software implemented in the workstation is the most essential part because this software realizes the interface between the user and the system. The introduction of a workstation in the hospital imposes a fundamental change in the work of radiologists and clinicians. This paper deals with the design and the development of a clinical workstation for pulmonary disease diagnostic. The usual working method of the clinician is taken as a reference for the design of the workstation. A software with image manipulating and processing facilities running in an X Window environment is described. The evaluation of the workstation is then undertaken in the service of pneumology of the pneumo-cardiologic hospital in Lyon, France. The conviviality and flexibility of the software is optimized in the function of the suggestions and criticism of the users.

This paper proposes a new service element (SE) of the OSI presentation layer, that contains functions for multimode medical images manipulation. The order of the image data compression and processing is determined within each mode on the basis of the reduction of total response time of the conferencing system, as well as the amount of transferred data through the underlying network. The introduction of the SE functions within the communication function stack of a conferencing medical system enhances the total performance of the conferencing entities (e.g., workstations, databases, modalities).

The design of hospital and medical diagnostic centers for the 21st century will include medical image capture, formatting and display systems designed to support multiple modalities from a single computing environment. These systems will be based on an architectural approach allowing upgrades and modifications, as required by new technologies, applied to the current modalities or requirements for yet unbuilt modalities. Changes will be needed to incorporate new hardware and software. The need to maintain a cost effective operation dictates flexible, easily up-gradeable architectures. This paper presents an architecture for an integrated system that can be built today to serve the needs for the hospitals and medical centers of tomorrow. These systems will be totally independent from any one manufacturer, a cost effective solution to processing images from multiple modalities, and based on the maximum use of commercial- off-the-shelf hardware and software.

A simple image review station has been designed for the surgical intensive care unit (SICU) of the Hospital of the University of Minnesota. The review station provides clinicians with digitized copies of all conventional films taken while a patient is in the SICU. The image review station consists of a single screen SPARCstation IPX. Images are acquired through a Kodak film digitizer and managed in cooperation with a bedside electronic medical record system on this unit. The review station software maintains a detailed log of all interactions that is routinely downloaded to a relational database for analysis of usage patterns.

With the tremendous growth in imaging applications and the development of filmless radiology, the need for compression techniques that can achieve high compression ratios with user specified distortion rates becomes necessary. Boundaries and edges in the tissue structures are vital for detection of lesions and tumors, which in turn requires the preservation of edges in the image. The proposed edge preserving image compressor (EPIC) combines lossless compression of edges with neural network compression techniques based on dynamic associative neural networks (DANN), to provide high compression ratios with user specified distortion rates in an adaptive compression system well-suited to parallel implementations. Improvements to DANN-based training through the use of a variance classifier for controlling a bank of neural networks speed convergence and allow the use of higher compression ratios for `simple' patterns. The adaptation and generalization capabilities inherent in EPIC also facilitate progressive transmission of images through varying the number of quantization levels used to represent compressed patterns. Average compression ratios of 7.51:1 with an averaged average mean squared error of 0.0147 were achieved.

Large volumes of high resolution radiological images acquired by a picture archiving and communication system require large archival storage and high network bandwidth. We use compression to increase storage capability and improve network throughput. Decompressing image data at the image display workstation, however, may drastically interfere with the interactive operation on the station. We have developed a method which optimizes the centralized long-term image data storage with image data compression, improves local availability of image data on display workstations, and minimizes the delay at the receiving image display workstation for an interactive session.

An important aspect of interventional neuroradiological procedures is the ability to access and interact with digital angiographic images to select and clinically evaluate intravascular therapeutic treatment. These issues can be adequately addressed and successfully accomplished using PACS. PACS provides the key technologies needed to access, display, and analyze sequential angiographic images. A dedicated neuroradiology PACS network consisting of workstations for the review of diagnostic data, quantitative analysis of arterial blood flow, and therapeutic assessment has been established at UCLA. The PACS provides not only a reliable and efficient infrastructure for on-line image retrieval and delivery of all digitally acquired angiographic images, but could also serve as a network supercomputing resource for computationally intensive calculations of hemodynamic parameters in the cerebral vasculature. In summary we have developed an innovative application of PACS in neuroangiography. It offers: (1) automatic optimized image display upon acquisition; (2) automatic retrieval of archived cases for a current patient; and, (3) a rapid, streamlined user interface for the quantitation of hemodynamic flow phenomena in normal and diseased cerebral vessels.

Although commercially available image processing technology exists today that could facilitate a real-time multi-modality workstation, there are multiple challenges in building such a system. The first is defining a common image format. In multi-modality imaging, the basic concern is combining two source images such that the resultant composite image retains the best qualities of each individual modality. One critical requirement is that the images be spatially matched before combination. The second challenge is to determine the system requirements for a multi-modality workstation. This would typically consist of what algorithms are to be used and the bandwidth requirements of the system. Assuming these issues are resolved, determining the appropriate hardware for each system requirement remains a significant hurdle. Array processing (systolic) and pipelined processing are two high performance hardware architectures available to meet these requirements. While each performs well with some operations, they falter with others. The key is the effective combination of these two architectures for maximum benefit. This paper addresses these challenges from a systems point of view.

Digital mammography is likely to replace conventional mammography within a few years. In anticipation of this, our group has been exploring the implications of image processing in digital mammography. Some of our findings are reported here.

Stereotactic radiosurgery is the precise delivery of high dose radiation to cranial lesions. A combination of isocenters or `shots' is utilized by the physician to map the contours of the target lesion. The accurate localization and definition of the lesion is a crucial element of this procedure. A system has been developed that allows the physician to define 3-D wire frame objects from a series of displayed images. Each image of the series is displayed as the background of a graphics screen. In order to facilitate more accurate definition of the lesion, variable contrast mapping and stretching routines allow the physician to adjust the histogram of the image for maximum visualization of the lesion and neighboring structures. A segmentation routine automatically detects the location of reference coordinates provided by a stereotactic frame fixed to the patient's skull. The physician is then able to create precise, 3-D wire frame representations of cranial lesions and structures, quickly and easily, buy selecting points on the displayed images. These wire frame representations can then be used to display the lesion stereoscopically in 3-D as opposed to 2-D cross-sections.

This paper describes two VLSI implementations that provide an effective solution to compressing medical image data in real time. The implementations employ a lossless data compression algorithm, known as the Rice algorithm. The first chip set was fabricated in 1991. The encoder can compress at 20 Msamples/sec and the decoder decompresses at the rate of 10 Msamples/sec. The chip set is available commercially. The second VLSI chip development is a recently fabricated encoder that provides improvements for coding low entropy data and incorporates features that simplify system integration. A new decoder is scheduled to be designed and fabricated in 1994. The performance of the compression chips on a suite of medical images has been simulated. The image suite includes CT, MR, angiographic images, and nuclear images. In general, the single-pass Rice algorithm compression performance exceeds that of two-pass, lossless, Huffman-based JPEG. The overall compression performance of the Rice algorithm implementations exceeds that of all algorithms tested including arithmetic coding, UNIX compress, UNIX pack, and gzip.