Medical application of volume-rendering techniques requires algorithms with a very high spatial accuracy. In this paper, a new volume-rendering method called 'Iso-Surface Volume Rendering' will be introduced. This method is able to visualize an iso-surface in a volumetric dataset with the highest possible accuracy. Furthermore, this new method is very fast, allowing interactive visualization on low-cost workstations without compromising quality. Given the point- spread function and the sample distance (voxel distance) of the acquisition device it is possible to calculate the spatial error that is made when the surface of an object is estimated by an iso-surface. It will be shown that this error depends on the sample distance, the sample location, and the interpolation function used. When the sample distance is very small, the error approaches zero.

A graphics-based simulation has been proved as a powerful tool in surgical and therapeutic applications. In a computer aided surgical simulation system, it must provide real-time rendering performance as well as flexible 3D object manipulation. In this paper, we propose a new volume rendering scheme, namely SSB (Scan-line Based Semi-Boundary). The time complexity of SSB is O(MN) plus a 2-Dimensional image rotation, and the SSB is theoretically faster than the original Semi-Boundary algorithm that will take O(LMN) time, where L,M,N are the Semi-Boundary nodes in X,Y,Z axes, respectively. We experimentally evaluate SSB rendering performance using large pelvis volume data. Our preliminary results show that SSB can achieve an interactive rendering performance for large volume pelvis data set with size of 512 by 512 by 245.

Image interpolation is an important operation that is widely used in medical imaging, image processing, and computer graphics. A variety of interpolation methods are available in the literature. However, their systematic evaluation is lacking. At a previous meeting, we presented a framework for the task independent comparison of interpolation methods based on a variety of medical image data pertaining to different parts of the human body taken from different modalities. In this new work, we present an objective, task-specific framework for evaluating interpolation techniques. The task considered is how the interpolation methods influence the accuracy of quantification of the total volume of lesions in the brain of Multiple Sclerosis (MS) patients. Sixty lesion detection experiments coming from ten patient studies, two subsampling techniques and the original data, and 3 interpolation methods is presented along with a statistical analysis of the results. This work comprises a systematic framework for the task-specific comparison of interpolation methods. Specifically, the influence of three interpolation methods in MS lesion quantification is compared.

A new approach is presented for determining left ventricular (LV) shape in a sequence of gated blood pool single photon emission computed tomography data. The approach consists of the automatic extraction and fitting of an analytical ellipsoidal surface that well-approximates the shape of the LV. Data from all slices and geometric characteristics of the analytical surface are exploited to enable robust and (over- constrained) recovery of the LV shape parameters. Examination of 3D renderings of the surface coupled with computation of LV volume and motion parameters over the systolic cycle can allow qualitative and quantitative determination of LV function.

Three dimensional (3D) plain film radiographic cephalometric analysis of boney skull landmarks may be used for patient diagnosis, treatment planning, prosthetic design, intra- operatively, and outcome assessment. To test the accuracy and reliability of 50 cephalometric landmarks, three dry human skulls, with and without metallic markers affixed to the landmarks, were digitized in our 3dCEPH software by 4 operators. The average inter-operator variability about mean landmark position, across all operators, for all 3 skull image pairs, was 3.33 mm. Ten landmarks exhibiting least variability were 1.15 mm average distance from the mean, including: B point 0.69 mm, Lower Incisal Edge 0.85 mm, and Anterior Nasal Spine 0.90 mm. The average rms error from the metallic fiducials for these 4 operators across all 50 landmarks, and 3 skulls was 5.03 mm. The 10 landmarks with the least variability exhibited 2.01 mm average distance from the fiducial, including: B point 1.69 mm, upper incisal edge 1.71 mm, lower incisal edge 1.78 mm. Additional studies are needed to test the robusticity of the hypothesis of homologous anatomy. Homology of landmarks is important to cephalometric comparisons between image pairs representing patient and 'normative,' pre- and post-surgical alteration, and different ages of the same patient.

A new method for calibrating a free-hand 3D ultrasound system based on a DC magnetic tracking position sensor is presented. The method uses a linear affine transformation for registering the individual 2D US images in the position sensor's reference coordinate system. The affine transformation allows for automatic scaling of the digitized ultrasound image in physical dimensions. We have also introduced a spherical tissue-mimicking phantom to facilitate the calibration procedure. The system's calibration was validated in a clinical setting by determining its precision in localizing a single point and its accuracy in measuring distances and volumes. Precision in localizing a point was consistently below 1.9 mm (rms error). The bias in distance measurements was 0.06 mm and the standard deviation was 0.75 mm. The accuracy of measuring volumes for a one-pass can was below 1% whereas for multiple pass scans it was below 4%. Furthermore, the performance of the DC magnetic tracking system was tested separately from the ultrasound scanner. We conclude that the ultrasound scanner rather than the position sensor is the limiting factor in the geometric accuracy of the 3D US system.

Intra-vascular ultrasound (IVUS) has significant potential for providing new information about the structure and condition of the coronary arteries. However, these image datasets are inherently noisy and characterized by frequent drop-outs and shadowing, considerably constraining their use for quantitative evaluation of coronary artery integrity and stent augmentation. A feasibility study was conducted to test the effectiveness of a software toolkit for processing, segmenting, and visualizing pre- and post-stent IVUS datasets, as a precursor to detailed quantitative evaluation of IVUS with regard to characterization of coronary luminal wall properties. Frame averaging, histogram processing, and anisotropic diffusion were used to reduce speckle noise and compensate for image dropout and shadowing. A region of interest (ROI) analysis was conducted to determine the effectiveness of the image processing. Preliminary results suggest that the image processing steps are effective at increasing the contrast to speckle ratio. In addition, contrast between the arterial wall and lumen was improved, producing an edge enhancement effect. Image registration was used to align images within a volume (i.e. 2-D registration) and between volumes (i.e. 3-D registration). A voxel matching method was used for the 2-D registration and a surface matching method was used for 3-D registration. A morphological connect method for segmentation was used to extract large structures in the data. A new edge-based algorithm was developed to segment the data in regions where the other method failed. Volume rendering methods were used to visualize the data. These preliminary visualizations of processed, segmented and registered IVUS datasets illustrated encouraging potential for further quantitative analysis of coronary artery morphology and pathology.

With the number of men seeking medical care for prostate diseases rising steadily, the need of a fast and accurate prostate boundary detection and volume estimation tool is being increasingly experienced by the clinicians. Currently, these measurements are made manually, which results in a large examination time. A possible solution is to improve the efficiency by automating the boundary detection and volume estimation process with minimal involvement from the human experts. In this paper, we present an algorithm based on SNAKES to detect the boundaries. Our approach is to selectively enhance the contrast along the edges using an algorithm called sticks and integrate it with a SNAKES model. This integrated algorithm requires an initial curve for each ultrasound image to initiate the boundary detection process. We have used different schemes to generate the curves with a varying degree of automation and evaluated its effects on the algorithm performance. After the boundaries are identified, the prostate volume is calculated using planimetric volumetry. We have tested our algorithm on 6 different prostate volumes and compared the performance against the volumes manually measured by 3 experts. With the increase in the user inputs, the algorithm performance improved as expected. The results demonstrate that given an initial contour reasonably close to the prostate boundaries, the algorithm successfully delineates the prostate boundaries in an image, and the resulting volume measurements are in close agreement with those made by the human experts.

We are building a 3D eye model from multimodal data and this paper describes our progress in using images from the Visible Human project and Scheimpflug images. The former is being targeted at muscles and the eyeball and the latter is being used to model the cornea. In this regard, the orbital anatomy of the Visible Human is being segmented and the relevant segmented data are then used to create geometric models. The lessons and methods we have learnt here will then be applied to MRI data when these become available. For Scheimpflug data, we are mutually registering individual images to reconstruct the cornea first, after which volumetric models can then be obtained. Our registration method is based on the fact that each image has a common point which can be used as a landmark. Future work includes registration of cornea and eyeball models, the incorporation of CT and biometric norms, as well as the identification of anatomical landmarks for deformation and registration.

This paper presents ongoing research in the field of volume visualization, interactive volumetric analysis and teleradiology. To cover the complete scenario from image acquisition to computer-based diagnosis and therapy support, interactive tools for volume visualization and volumetric analysis have been integrated into a state of the art teleradiology system through a general plug-in mechanism. Visualization is demonstrated as a hybrid approach integrating precise volume visualization based on the Heidelberg Raytracing Model with fast surface rendering. With respect to volumetric analysis, tools extend the 3D-viewing. They allow to measure distances, angles, areas and volumes through 3D- pointing and user controlled segmentation processes. Various input devices can be connected to control the navigation of the viewer or objects in the scene. Moreover, a haptic interface has been investigated. The realistic perception through force feedback while navigating within the 3D- reconstruction was positively judged by the medical users as well as software developers. The work is closely related to research in the field of heart, liver and cranio-facial surgery planning. In cooperation with our medical partners the development of tools as presented proceed the integration of image analysis into the clinical routine.

The following article introduces a maxillofacial surgical planning tool based on a general two- and three-dimensional image processing and visualization concept. By applying various rendering methods to computer tomography volume data (CT), it is possible to produce stereoscopic views of the patient. These reconstructed images enable the surgeon to carry out a computer-aided three-dimensional maxillofacial profile analysis and operation plan. The 'Medical Visualization Tools' (MeVisTo) concept developed by our workgroup serves as the basis for this. Besides an easy-to- handle user interface for 3D tasks, speed and quality of the volume rendering process are of great importance for the visualization. With 'Virtual Sphere Rendering' (VSR) a method is presented that, by means of pre-calculating views, enables a visualization of data in real-time with more than 25 images per second.

Five patients with cerebral palsy, hip dysplasia, pelvic obliquity, and scoliosis were evaluated retrospectively using three dimensional computed tomography (3DCT) scans of the proximal femur, pelvis, and lumbar spine to qualitatively evaluate their individual deformities by measuring a number of anatomical landmarks. Three dimensional reconstructions of the data were visualized, analyzed, and then manipulated interactively to perform simulated osteotomies of the proximal femur and pelvis to achieve surgical correction of the hip dysplasia. Severe deformity can occur in spastic cerebral palsy, with serious consequences for the quality of life of the affected individuals and their families. Controversy exists regarding the type, timing and efficacy of surgical intervention for correction of hip dysplasia in this population. Other authors have suggested 3DCT studies are required to accurately analyze acetabular deficiency, and that this data allows for more accurate planning of reconstructive surgery. It is suggested here that interactive manipulation of the data to simulate the proposed surgery is a clinically useful extension of the analysis process and should also be considered as an essential part of the pre-operative planning to assure that the appropriate procedure is chosen. The surgical simulation may reduce operative time and improve surgical correction of the deformity.

Three dimensional microscopy visualization has the potential of playing a significant role in the study of 3D cellular structures in biomedical research. Such potential, however, has not been fully realized due to the difficulties of current visualization methods in coping with the unique nature of microscopy image volumes, such as low image contrast, noise and unknown transfer functions. In this paper, we present a new 3D microscopy imaging approach that integrates volume visualization and 3D image processing techniques for interactive 3D data exploration and analysis. By embedding 3D image enhancement procedures into the volume visualization pipeline, we are able to automatically generate image- dependent transfer functions to reveal subtle features that are otherwise difficult to visualize. It allows the users to interactively manipulate a small number of parameters to achieve desired visualization effects. Other 3D image processing techniques, such as quantification and segmentation, may also be integrated within the data exploration process for interactive image analysis.

The achievement of volume geometry data from middle ear structures and surrounding components performs a necessary supposition for the finite element simulation of the vibrational and transfer characteristics of the ossicular chain. So far those models base on generalized figures and size data from anatomy textbooks or particular manual and one- or two-dimensional distance measurements of single ossicles, mostly obtained by light microscopy, respectively. Therefore the goal of this study is to create a procedure for complete three-dimensional imaging of real middle ear structures (tympanic membrane, ossicles, ligaments) in vitro or even in vivo. The main problems are their microscopic size with relevant structures from 10 micrometer to 5 mm, representing various tissue properties (bone, soft tissue). Additionally, these structures are surrounded by the temporal bone, the most solid bone of the human body. Generally there exist several established diagnostic tools for medical imaging that could be used for geometry data acquisition, e.g., X-ray computed tomography and magnetic resonance imaging. Basically they image different tissue parameters, either bony structures (ossicles), or soft tissue (tympanic membrane, ligaments). But considering this application those standard techniques allow low spatial resolution only, usually in the 0.5 - 1mm range, at least in one spatial direction. Thus particular structures of the middle ear region could even be missed completely because of their spatial location. In vitro there is a way out by collecting three complete data sets, each distinguished by 90 degree rotation of a cube-shaped temporal bone specimen. That allows high-resolution imaging in three orthogonal planes, which essentially supports the three-dimensional interpolation of the unknown elements, starting from the regularly set elements of the cubic grid with an edge extension given by the original two-dimensional matrix. A different approach represents the application of a micro- tomographic imaging device. Therefore an X-ray beam focused down to few microns passes the object in a tomographic arrangement. Subsequently the slices become reconstructed. Generally spatial resolution down to 10 micrometer may be obtained by using this procedure. But there exist few devices only, it is not available as standard equipment. The best results concerning spatial resolution should be achieved by applying conventional histologic sectioning techniques. Of course the target will become destroyed during the procedure. It is cut into sections (e.g., 10 micrometer thick), every layer is stained, and the image acquired and stored by a digital still-camera with appropriate resolution (e.g., 2024 X 3036). Three-dimensional reconstruction is done with the computer. The staining allows visual selection of bones and soft tissues, resolutions down to 10 micrometer are possible without target segmentation. But there arise some practical problems. Mainly the geometric context of the layers is affected by the cutting procedure, especially if cutting bone. Another problem performs the adjustment of the -- possibly distorted -- slices to each other. Artificial markers are necessary, which could allow automatic adjustment too. But the introduction and imaging of the markers is difficult inside the temporal bone specimen, that is interspersed by several cavities. Of course the internal target structures must not be destroyed by the marker introduction. Furthermore the embedding compound could disturb the image acquisition, e.g., by optical scattering of paraffin. A related alternative is given by layered ablation/grinding and imaging of the top layer. This saves the geometric consistency, but requires very tricky and time-consuming embedding procedures. Both approaches require considerable expenditures. The possible approaches are evaluated in detail and first results are compared. So far none of the above-mentioned procedures has been established as a standard tool for three-dimensional geometry data acquisition of the middle ear. Otherwise the establishment of a high-resolution imaging technique for those structures, even in vivo, would be of high interest in diagnostics, anatomy and middle ear modeling and research at all.

Three dimensional (3-D) images have recently received wide attention in applications involving medical treatment. Most current 3-D imaging methods focus on the internal organs of the body. However, several medical image applications such as plastic surgery, body deformities, rehabilitation, dental surgery and orthodontics, make use of the surface contours of the body. Several techniques are currently available for producing 3-D images of the body surface and most of the systems which implement these techniques are expensive, requiring complex equipment with highly trained operators. The research involves the development of a simple, low cost and non-invasive contour capturing method for facial surfaces. This is achieved using the structured light technique, employing a standard commercial slide projector, CCD camera and a frame-grabber card linked to a PC. Structured light has already been used for many applications, but only to a limited extent in the clinical environment. All current implementations involve extensive manual intervention by highly skilled operators and this has proven to be a serious hindrance to clinical acceptance of 3-D imaging. A primary objective of this work is to minimize the amount of manual intervention required, so that the system can be used by clinicians who do not have specialist training in the use of this equipment. The eventual aim is to provide a software assisted surgical procedure, which by merging the facial data, allows the manipulation of soft tissue and gives the facility to predict and monitor post-surgical appearance. The research focuses on how the images are obtained using the structured light optic system and the subsequent image processing of data to give a realistic 3-D image.

Computed radiography is used for a wide range of projection radiography examinations. To produce useful diagnostic images it is necessary to apply an appropriate tone scale to the raw CR data. This paper presents a new automated tone scaling method for computed radiography and presents the results of a clinical study of the algorithm encompassing a wide range of clinical examinations.

Softcopy reading and CAD in mammography are both dependent upon the characteristics of the source of the digital data, either direct digital mammography or digitized screen-film mammography. In this work, a standardization is proposed based on geometric and intensity transformation that are discovered using a set of calibration images. A genetic algorithm is used to search for the best transformations. Results indicate that standardization for display can be achieved, but that CAD performance varies even after standardization: both sensitivity and false-positive rates appear to be reduced.

Multiple light reflections in the emissive structure and electron backscattering processes, in the vacuum of cathode- ray tubes (CRT) are sources of luminance spread and loss of image quality. To perform glare measurements in CRTs, a detector that can record low intensity signals from small dark fields in the presence of a bright surrounding field is required. We have designed and constructed a luminance probe consisting of a cone-shaped nose and an internally baffled barrel that measures light generated in a spot of abut 4 mm diameter. The probe minimizes light scattering from the probe back into the faceplate, and minimizes contributions from bright surrounding regions. In this paper, we report results on the performance of the probe and demonstrate its application for characterizing CRT devices. The probe is shown to be capable of measuring contrast ratios of at least 800 for dark fields with a radius of 2 cm. Measurements on a medical imaging CRTs illustrate how significant veiling glare in the device can result is contrast ratios of about 100 and how the reflection of ambient light can reduce the contrast ratio to below 25.

Many of the critical basal ganglia structures are not distinguishable on anatomical magnetic resonance imaging (MRI) scans, even though they differ in functionality. In order to provide the neurosurgeon with this missing information, a deformable volumetric atlas of the basal ganglia has been created from the Shaltenbrand and Wahren atlas of cryogenic slices. The volumetric atlas can be non-linearly deformed to an individual patient's MRI. To facilitate the clinical use of the atlas, a visualization platform has been developed for pre- and intra-operative use which permits manipulation of the merged atlas and MRI data sets in two- and three-dimensional views. The platform includes graphical tools which allow the visualization of projections of the leukotome and other surgical tools with respect to the atlas data, as well as pre- registered images from any other imaging modality. In addition, a graphical interface has been designed to create custom virtual lesions using computer models of neurosurgical tools for intra-operative planning. To date 17 clinical cases have been successfully performed using the described system.

In minimally invasive surgery (MIS), endoscopes are used in real-time to enhance visualization and minimize invasion of healthy tissue. Unfortunately, the field of view provided by the scope is limited. In interactive image guided surgery (IIGS), the display of present surgical position on preoperative tomographic images enhances the surgeons field of view and provides knowledge of surgical anatomy. However, changes in the anatomy during surgery are not realized by the current IIGS techniques. This manuscript details the initial experiments conducted to merge the strengths of MIS with IIGS. This incudes: (1) developing a technique for accurately tracking an endoscope in physical space and (2) determining a transformation to map endoscopic image space into physical (patient) space.

An important clinical application of biomedical imaging and visualization techniques is provision of image guided neurosurgical planning and navigation techniques using interactive computer display systems in the operating room. Current systems provide interactive display of orthogonal images and 3D surface or volume renderings integrated with and guided by the location of a surgical probe. However, structures in the 'line-of-sight' path which lead to the surgical target cannot be directly visualized, presenting difficulty in obtaining full understanding of the 3D volumetric anatomic relationships necessary for effective neurosurgical navigation below the cortical surface. Complex vascular relationships and histologic boundaries like those found in artereovenous malformations (AVM's) also contribute to the difficulty in determining optimal approaches prior to actual surgical intervention. These difficulties demonstrate the need for interactive oblique imaging methods to provide 'line-of-sight' visualization. Capabilities for 'line-of- sight' interactive oblique sectioning are present in several current neurosurgical navigation systems. However, our implementation is novel, in that it utilizes a completely independent software toolkit, AVW (A Visualization Workshop) developed at the Mayo Biomedical Imaging Resource, integrated with a current neurosurgical navigation system, the COMPASS stereotactic system at Mayo Foundation. The toolkit is a comprehensive, C-callable imaging toolkit containing over 500 optimized imaging functions and structures. The powerful functionality and versatility of the AVW imaging toolkit provided facile integration and implementation of desired interactive oblique sectioning using a finite set of functions. The implementation of the AVW-based code resulted in higher-level functions for complete 'line-of-sight' visualization.

A method is described for computing the registration of computed tomography (CT) images of the head with physical space using the outer surface of the skull for registration. A head phantom is constructed by placing a latex swim cap over a plastic skull and pouring a gelatin mixture into the space between skull and cap. A Panametrics 10.0 MHz immersion type ultrasound transducer has been rigidly attached to an optically tracked probe. This device is swept across the phantom to localize the surface of the skull. A standard deviation sliding filter is applied to the ultrasound signals to determine the location of the skull echoes. A CT derived model of the surface of the skull is generated using contours obtained from a deformable model-based segmentation algorithm. The CT and ultrasound surfaces are registered using a modified iterative closest point (ICP) algorithm. Results of a phantom experiment, validated with the Vanderbilt fiducial marker system, show target registration errors less than 4 mm. This result shows the feasibility of performing physical space to CT space registration using the outer surface of the skull.

Multimodality medical image fusion is becoming increasingly important in clinical applications, which involves information processing, registration and visualization of interventional and/or diagnostic images obtained from different modalities. This work is to develop a multimodality medical image fusion technique through probabilistic quantification, segmentation, and registration, based on statistical data mapping, multiple feature correlation, and probabilistic mean ergodic theorems. The goal of image fusion is to geometrically align two or more image areas/volumes so that pixels/voxels representing the same underlying anatomical structure can be superimposed meaningfully. Three steps are involved. To accurately extract the regions of interest, we developed the model supported Bayesian relaxation labeling, and edge detection and region growing integrated algorithms to segment the images into objects. After identifying the shift-invariant features (i.e., edge and region information), we provided an accurate and robust registration technique which is based on matching multiple binary feature images through a site model based image re-projection. The image was initially segmented into specified number of regions. A rough contour can be obtained by delineating and merging some of the segmented regions. We applied region growing and morphological filtering to extract the contour and get rid of some disconnected residual pixels after segmentation. The matching algorithm is implemented as follows: (1) the centroids of PET/CT and MR images are computed and then translated to the center of both images. (2) preliminary registration is performed first to determine an initial range of scaling factors and rotations, and the MR image is then resampled according to the specified parameters. (3) the total binary difference of the corresponding binary maps in both images is calculated for the selected registration parameters, and the final registration is achieved when the minimum number of mismatch pixels gives the optimal scaling factor and rotation angle. Cross-modality quantification is then performed by incorporating the probabilistic pixel memberships from one modality (e.g., MRI) and the functional activities from another modality (e.g., PET or CT) within any given region of interests. The consistent nature of the comparative clinical studies show that our matching technique result in a robust and accurate image fusion of MRI and PET/CT scans.

We have developed a new ultrasound scan conversion algorithm that can be executed efficiently on modern mediaprocessors. Our algorithm is designed to handle the address calculations, and input and output (I/O) data loading concurrently with the interpolation. The processing unit's computing power can be dedicated to performing pixel interpolations while the other operations are handled by an independent direct memory access (DMA) controller. By making intelligent use of the I/O transfer capabilities of the DMA controller, the algorithm avoids spending the processing unit's valuable computing cycles in address calculations and nonactive pixel blanking. Furthermore, the new approach speeds up the computation by utilizing the ability of very long instruction word (VLIW) processors to perform multiple operations in parallel. Our scan conversion algorithm was implemented on a multimedia and imaging system based on the Texas Instruments TMS320C80 Multimedia Video Processor (MVP). The algorithm was also simulated on a future mediaprocessor called the High- performance Multimedia Processor using VLIW (HMPV). Computing cycles are spent only on predeterminable nonzero output pixels. For example, for a 512X512 16-bit image with 101,829 nonzero output pixels, an execution time of 13.3 ms was achieved on the TMS320C80 while it takes 4.4 ms on the HMPV. This algorithm demonstrates a substantial improvement over previous scan conversion algorithms, and its optimized implementation enables modern commercially-available programmable processors to support scan conversion at video rates.

For a reliable diagnosis through visual telecommunication systems, it is very important to reproduce patient images with correct color. Natural color television system previously proposed is capable of reproducing the original color of the objects using multispectral images, even when the illumination environment for observing images is different from the illumination used for recording. In this paper, to render this system practical for telemedicine system, we reduce the number of color bands of multispectral camera by utilizing the principal components of spectral reflectances in human skin class. In addition, we investigate a simple method, using color charts, to obtain the recording information of both illumination and camera. As color chart selection criterion, it becomes clear that principal components for human skin class need to be described by the linear combination of spectral reflectances of color charts. Principal components for human skin class are obtained through measurements and analysis of 484 human skin spectral reflectances. In the experiment, natural color reproduction for human skin is realized from limited number of color bands using color charts for acquiring the recording information instead of spectral measurement devices.

Media processors, with high-level programming languages, Application Programmer Interfaces (APIs) and rich media libraries, are capable of providing an effective solution for medical imaging products. Video, audio, 3D graphics, printing and communications functions become cost-effective by sharing one media processor. This paper includes an overview of media processors, their application including medical imaging uses, and projections for future media processors.

The continued economic pressure that is being placed upon the healthcare industry creates both challenge and opportunity to develop cost effective healthcare tools. Tools that provide improvements in the quality of medical care at the same time improve the distribution of efficient care will create product demand. Video Conferencing systems are one of the latest product technologies that are evolving their way into healthcare applications. The systems that provide quality Bi- directional video and imaging at the lowest system and communication cost are creating many possible options for the healthcare industry. A method to use only 128k bits/sec. of ISDN bandwidth while providing quality video images in selected regions will be applied to echocardiograms using a low cost video conferencing system operating within a basic rate ISDN line bandwidth. Within a given display area (frame) it has been observed that only selected informational areas of the frame of are of value when viewing for detail and precision within an image. Much in the same manner that a photograph is cropped. If a method to accomplish Region Of Interest (ROI) was applied to video conferencing using H.320 with H.263 (compression) and H.281 (camera control) international standards, medical image quality could be achieved in a cost-effective manner. For example, the cardiologist could be provided with a selectable three to eight end-point viewable ROI polygon that defines the ROI in the image. This is achieved by the video system calculating the selected regional end-points and creating an alpha mask to signify the importance of the ROI to the compression processor. This region is then applied to the compression algorithm in a manner that the majority of the video conferencing processor cycles are focused on the ROI of the image. An occasional update of the non-ROI area is processed to maintain total image coherence. The user could control the non-ROI area updates. Providing encoder side ROI specification is of value. However, the power of this capability is improved if remote access and selection of the ROI is also provided. Using the H.281 camera standard and proposing an additional option to the standard to allow for remote ROI selection would make this possible. When ROI is applied the ability to reach the equivalent of 384K bits/sec ISDN rates may be achieved or exceeded depending upon the size of the selected ROI using 128K bits/sec. This opens additional opportunity to establish international calling and reduced call rates by up to sixty- six percent making reoccurring communication costs attractive. Rates of twenty to thirty quality ROI updates could be achieved. It is however important to understand that this technique is still under development.

Digital vascular computer systems are used for radiology and fluoroscopy (R/F), angiography, and cardiac applications. In the United States alone, about 26 million procedures of these types are performed annually: about 81% R/F, 11% cardiac, and 8% angiography. Digital vascular systems have a very wide range of performance requirements, especially in terms of data rates. In addition, new features are added over time as they are shown to be clinically efficacious. Application-specific processing modes such as roadmapping, peak opacification, and bolus chasing are particular to some vascular systems. New algorithms continue to be developed and proven, such as Cox and deJager's precise registration methods for masks and live images in digital subtraction angiography. A computer architecture must have high scalability and reconfigurability to meet the needs of this modality. Ideally, the architecture could also serve as the basis for a nonvascular R/F system.

Radiology images are acquired electronically using phosphor plates that are read in Computed Radiology (CR) readers. An automated radiology image orientation processor (RIOP) for determining the orientation for chest images and for abdomen images has been devised. In addition, the chest images are differentiated as front (AP or PA) or side (Lateral). Using the processing scheme outlined, hospitals will improve the efficiency of quality assurance (QA) technicians who orient images and prepare the images for presentation to the radiologists.

Trends in medical imaging indicate that the storage requirements for digital medical datasets require a more efficient, scalable storage architecture for large-scale RIS/PACS to support high-speed retrieval for multiple concurrent clients. As storage and networking technologies mature, the cost of applying such technologies in medical imaging has become more economically viable. We propose to take advantage of such economies of scale in technology to provide an effective network workstation storage solution for achieving (1) faster display and navigation response time, (2) higher server throughput and (3) better data storage management. Full-field direct digital mammography presents a challenging problem in the design of digital workstation systems for screening and diagnosis. Due to the spatial and contrast resolution required for mammography, the digital images are large (exceeding 5K X 6K X 14 bits approximately equals 60MB per image) and therefore difficult to display using commercially available technology. We are developing clinically useful methods of storing, displaying and manipulating large digital images in a medical media server using commercial technology. In this paper we propose an Intelligent Grid-based Data Layout Mechanism to optimize the total response time of a reading by minimizing the speed of image access (data I/O time) and the number of data access requests to the server (queueing effects) during the image navigation. A Navigation Threads Model is developed to characterize the performance of many navigation threads involved in the course of performing a reading session. In our grid-based data layout approach, a large 2D direct-digital mammogram image is divided spatially into many small 2D grids and is stored into an array of magnetic disks to provide parallel grid-based readout services to clients. Such a grid- based approach not only provides fine-granularity control, but also provides a means of collecting statistical information about the distribution of Region of Interests (ROI) for a given image in the storage systems. Hence, it provides statistical rules to guide image navigation and guidelines for reorganizing the data layout within the storage server (replication of ROI blocks) dynamically; hence, better load balancing can be achieved and the overall image navigation throughput for the system can be maximized. Given the same buffer capacity and data access mode, this technique can statistically guarantee the maximum image retrieval time, and can scale-up easily without significant performance degradation. Throughout this paper, we assume that a high- speed network is used in our client/server model and network latency (data communication cost) is minimal compared to data I/O cost. In addition, the cost of reporting diagnostic results associated with the total response time is assumed to be negligible.

Computed tomographic colography (CTC or virtual colonoscopy) is a new technique for imaging the colon for the detection of colorectal neoplasms. Early clinical assessment of this procedure has shown that the performance of this test is acceptable for colorectal screening examinations. The current version of CTC utilizes an interactive combination of axial, reformatted 2D and 3D images (from an endoluminal perspective) that are generated in real time. Retained fluid in the lumen of the colon is a commonly encountered problem that can obscure lesions. Prone imaging in addition to standard supine views are often required to visualize obscured colonic segments. Although the colorectum is often seen optimally with combined supine and prone views, twice as much interpretation time is required with both acquisitions. The purpose of this study is to describe a novel system of synchronous display of supine and prone images of the colon. Simultaneous display of synchronized (anatomically registered) views of the colon eliminates the need for two separate readings of the colon and shortens interpretation time. This tool has all of the features of the original CTC interpretation system (presented in 1995) and includes recent innovations such as virtual pathology which is present in another paper within this Proceeding. The anatomic levels are indexed to match each other and advanced synchronously as the radiologist interprets the data set. Axial, reformatted 2D and 3D images are displayed and simultaneously updated for both prone and supine images on the same computer screen. The colon only needs to be reviewed once with the diagnostic benefit of both scans. In many cases, the two scans can be interpreted nearly as quickly as one. Conclusion: Synchronous display of prone and supine images of the colon is a new enhancement for CTC that combines with advantages of prone and supine views without the added interpretations time of reviewing two separate scans.

Diagnostic workstations have generally lacked acceptance due to awkward interfaces, poor usability and lack of clinical data integration. We developed a new methodology for the design and implementation of diagnostic workstations and applied the methodology in diagnostic neuroradiology. The methodology facilitated the objective design and evaluation of optimal diagnostic features, including the integration of images and reports, and the implementation of intelligent and adaptive graphical user interfaces. As a test of this new methodology, we developed and evaluated a neuroradiological diagnostic workstation. The general goals of diagnostic neuroradiologists were modeled and directly used in the design of the UCLA Digital ViewBox, an object-oriented toolkit for medical imaging workstations. For case-specific goals, an object-oriented protocol toolkit was developed for rapid development and integration of new protocols, modes, and tools. Each protocol defines a way to arrange and process data in order to accomplish diagnostic goals that are specific to anatomy (e.g., a spine protocol), or to a suspected pathology (e.g., a tumor protocol). Each protocol was divided into modes that represent diagnostic reading tasks. Each mode was further broken down into functions supporting that task. Via a data mediator engine, the workstation communicated with clinical data repositories, including the UCLA HIS, Clinical RIS/PACS and individual DICOM compatible scanners. The data mediator served to transparently integrate, retrieve, and cache image and report data. Task-oriented Reading protocols automatically present the appropriate diagnostic information and diagnostic tools to the radiologist. We describe a protocol toolkit that enables the rapid design and implementation of customized reading protocols. We also present an intelligent layer that enables the automatic presentation of the appropriate information. This new methodology for diagnostic workstation design led to an improved neuroradiology workstation. Future research will focus on developing protocols for other diagnostic specialties.

Picture Archiving and Communication Systems (PACS) will evolve into more network-centric, distributed systems in the near future. The reasons for this are better manageability, higher scalability and reduced cost of ownership. Medical image display software should be prepared for this development by making optimized use of networked resources, for instant large medical image files. Most software products currently used, however, is not designed for networked operation and imposes architectural limitations on the possible performance. We will propose an alternative software architecture for medical image display systems that can overcome this limitation without affecting the performance of conventional disk based I/O.

This paper presents a study on quantifying and understanding the impact of multiple lossy JPEG compression cycles on medical images. In medical imaging applications, an image is compressed using a specific technique and parameters and then transmitted to a user (i.e., physician) where it is decompressed for viewing. The user may then wish to retransmit the decompressed image (or a portion of the image) to another user for further review. This second transmission might use the same compression parameters as the first transmission, or they might change because of system constraints. This cycle of compression/decompression might be repeated several times. When performing multiple JPEG compression cycles, several scenarios are likely to occur between cycles: (1) changes in the compression ratio; (2) changes in the quantization table; and (3) cropping of the image. In this study, we considered combinations of high (40:1), medium (20:1), and low (10:1) compression ratios and visually-based q-tables designed for viewing distances of 2, 3, and 4-picture heights. The compression ratio and/or q-table were varied between compression cycles. We also consider cropping of the image by one-pixel increments between compression cycles. Root-mean- squared error (RMSE) between the original image and the decompressed image was used to quantify the impact of these parameters. This study examined these issues for the specific case of computed radiography images.

To make the archival and transmission of medical images in PACS (Picture Archiving and Communication Systems) and teleradiology systems user-friendly and economically profitable, the adoption of an efficient compression technique is an important feature in the design of such systems. An important category of lossy compression techniques uses the wavelet transformation for decorrelation of the pixel values, prior to quantization and entropy coding. For the coding of sets of images, the images are mostly independently compressed with a two-dimensional compression scheme. In this way, one discards the similarity between adjacent slices. The aim of this paper is to compare the performance of some two- dimensional and three-dimensional implementations of wavelet compression techniques and investigate some design issues as decomposition depth, the choice of wavelet filters and entropy coding.

Wavelet-based image compression is a very effective technique for medical images, giving significantly better results than the JPEG algorithm. Recent advanced wavelet techniques, such as embedded zerotree wavelet coding or set partitioning in hierarchical trees (SPIHT), have further improved compression efficiency by exploiting the natural relationship between corresponding coefficients at different scales and by progressively refining coefficient values. It is also well known that full 3-D wavelet compression of 3-D data sets is significantly more efficient than 2-D compression of individual slices, but the memory requirements are very high. We explore the extension of the SPIHT algorithm to three dimensions, and also of intermediate approaches such as the encoding of a 3-D image in 'slabs' of 16 slices at a time or simple subtraction of neighboring slices, which yield different tradeoffs between speed and memory requirements on the one hand and compression efficiency on the other for a variety of possible approaches on typical 3-D CT and MRI data sets.

Experiments using NASA's Advanced Communications Technology Satellite were conducted to provide an estimate of the compressed video quality required for preservation of clinically relevant features for the detection of trauma. Bandwidth rates of 128, 256 and 384 kbps were used. A five point Likert scale (1 equals no useful information and 5 equals good diagnostic quality) was used for a subjective preference questionnaire to evaluate the quality of the compressed ultrasound imagery at the three compression rates for several anatomical regions of interest. At 384 kbps the Likert scores (mean plus or minus SD) were abdomen (4.45 plus or minus 0.71), carotid artery (4.70 plus or minus 0.36), kidney (5.0 plus or minus 0.0), liver (4.67 plus or minus 0.58) and thyroid (4.03 plus or minus 0.74). Due to the volatile nature of the H.320 compressed digital video stream, no statistically significant results can be derived through this methodology. As the MPEG standard has at its roots many of the same intraframe and motion vector compression algorithms as the H.261 (such as that used in the previous ACTS/AMT experiments), we are using the MPEG compressed video sequences to best gauge what minimum bandwidths are necessary for preservation of clinically relevant features for the detection of trauma. We have been using an MPEG codec board to collect losslessly compressed video clips from high quality S- VHS tapes and through direct digitization of S-video. Due to the large number of videoclips and questions to be presented to the radiologists and for ease of application, we have developed a web browser interface for this video visual perception study. Due to the large numbers of observations required to reach statistical significance in most ROC studies, Kappa statistical analysis is used to analyze the degree of agreement between observers and between viewing assessment. If the degree of agreement amongst readers is high, then there is a possibility that the ratings (i.e., average Likert score at each bandwidth) do in fact reflect the dimension they are purported to reflect (video quality versus bandwidth). It is then possible to make intelligent choice of bandwidth for streaming compressed video and compressed videoclips.

In this paper, we describe an algorithm for compression of pre-scan-converted ultrasound tissue data. Between compression ratios of 8:1 and 120:1, we have found that there is a significant reduction in distortion using compression prior to scan conversion rather than after scan conversion. We have also found an additional improvement using wavelets, zerotree quantization and arithmetic coding rather than standard block DCT compression.

This study presents an approach to build a 3D vascular system of coronary for the development of a virtual cardiology simulator. The 3D model of the coronary arterial tree is reconstructed from the geometric information segmented from the Visible Human data set for physical analysis of catheterization. The process of segmentation is guided by a 3D topologic hierarchy structure of coronary vessels which is obtained from a mechanical model by using Coordinate Measuring Machine (CMM) probing. This mechanical professional model includes all major coronary arterials ranging from right coronary artery to atrioventricular branch and from left main trunk to left anterior descending branch. All those branches are considered as the main operating sites for cardiology catheterization. Along with the primary arterial vasculature and accompanying secondary and tertiary networks obtained from a previous work, a more complete vascular structure can then be built for the simulation of catheterization. A novel method has been developed for real time Finite Element Analysis of catheter navigation based on this featured vasculature of vessels.

To improve the benefit of surgical simulators for education and research a visual convincing modeling of the operation scenario and the involved tissues is not sufficient. It is also necessary to simulate the deformation and resulting inner forces of tissue under influence of external forces caused by, for example, medical instruments or gravity. In this paper, we present a hybrid neuro-fuzzy system, which was designed for the description and simulation of tissues. The neuro-fuzzy system can be used to simulate the physical behavior like stiffness, viscosity and inertia of deformable or elastic tissues in surgical simulation. The parameters of a physical model or prior expert knowledge in the form of linguistic terms can be used to initialize the network parameters. Using a neural network structure, local changes to the system like cuts or ruptures can be performed during simulation. As an application example, some simulation results in the area of gynecological laparoscopy are given.

Complex anatomical information can be obtained from a high- resolution, 3D, radiologic image by interactively navigating through it in a manner similar to an endoscopic examination. Real-time computation of these 'virtual' endoscopic views, however, is needed to permit truly interactive navigation. We present such a method, based on the temporal-coherence concept. Included results show the method's efficiency. The method constitutes parts of a complete virtual-endoscopic system we have devised. This system is illustrated for 3D pulmonary analysis.

Purpose: To improve the diagnosis of pathologic modified airways, a visualization system has been developed and tested based on the techniques of digital image analysis, synthesis of spiral CT and the visualization by methods of virtual reality. Materials and Methods: 20 patients with pathologic modifications of the airways (tumors, obstructions) were examined with Spiral-CT. The three-dimensional shape of the airways and the lung tissue is defined by a semiautomatic volume growing method and a following geometric surface reconstruction. This is the basis of a multidimensional display system which visualizes volumes, surfaces and computation results simultaneously. To enable the intuitive and immersive inspection of the airways a virtual reality system, consisting of two graphic engines, a head mounted display system, data gloves and specialized software was integrated. Results: In 20 cases the extension of the pathologic modification of the airways could be visualized with the virtual bronchoscopy. The user interacts with and manipulates the 3D model of the airways in an intuitive and immersive way. In contrast to previously proposed virtual bronchoscopy systems the described method permits truly interactive navigation and detailed exploration of anatomic structures. The system enables a user oriented and fast inspection of the volumetric image data. Conclusion: To support radiological diagnosis with additional information in an easy to use and fast way a virtual bronchoscopy system was developed. It enables the immersive and intuitive interaction with 3D Spiral CTs by truly 3D navigation within the airway system. The complex anatomy of the central tracheobronchial system could be clearly visualized. Peripheral bronchi are displayed up to 5th degree.

Computed tomographic colonography (CTC or virtual colonoscopy) is a new technique for imaging the colon for the detection of colorectal neoplasm. Early clinical assessment of this procedure has shown that the performance of this test is acceptable for colorectal screening examinations. The current version of CTC utilizes an interactive combination of axial, reformatted 2D and 3D images (from an endoluminal perspective) that are generated in real time. Navigating the colorectum within the virtual endoscopy (VE) metaphor is tedious. To alleviate this problem and free the radiologist to concentrate on diagnosis, a two pass approach has been adopted. In the first pass, the colon is semiautomatically navigated and its midline is recorded. Then in a second pass, the radiologist moves the view point (virtual endoscope tip) along this midline. This second pass alone takes from 15 to 40 minutes per scan. Additional volume rendered displays have been developed which show longer segments of the colon in formats analogous to views which may be seen at autopsy. These include a technique called planar virtual pathology' (PVP) which uses image planes of the two longitudinal transcolonic sections as cut planes within isometric volume rendered images. Scenes are rendered with rays passing orthogonally through these planes from both sides. Each of the four resulting images depicts approximately one half of the inner surface of a 12 cm bowel segment. Interpretation is performed by viewing these colon segments at approximately 8 cm intervals. In this way, the entire colon can be rapidly examined with a minimum of navigational input from the radiologist. Other virtual pathology views have also been developed. These include cylindrical virtual pathology (CVP) and mercator virtual pathology (MVP). CVP views are formed by casting rays perpendicular to a straight line which approximates a segment of the colon midline. These views are analogous to the result of splitting a segment of excised colon and opening it to expose its inner surface. Interpretation can be performed interactively or by viewing a series of precomputed CVP views as described for PVP above. With MVP, views are generated by casting rays from a single viewpoint in all directions. The resulting view is a panoramic image. Interpretation of CVP is performed interactively in a manner similar to that of VE. Conclusion: A new display mode, referred to as virtual pathology (VP) has been developed which enables the radiologist to interpret long segments of the colon at one time. This new display metaphor has great promise in reducing interpretation time for CT Colonography.

There are many situations in 3D imaging applications wherein object and non-object regions with identical image properties are situated in close proximity. Although the commonly used opacity functions are adequate in capturing object gradation in these situations, the non-object regions obscure the objects of interest in volume renditions based on these opacity assignments. These obscurations cannot be removed using simple techniques except by manually masking them out. We present a method based on fuzzy connectedness principles which effectively removes the obscuration automatically. The methods are illustrated using craniomaxillofacial CT patient data for separating thin bones from skin and muscle, and muscle from skin and bone boundary. The resulting remarkable clarity of renditions are illustrated.

In this paper, we present a new volume rendering algorithm for visualizing multiple objects from separate volume datasets. Octree volumes are defined to represent volumetric objects encapsulated in regular volume datasets. An adaptive space subdivision is used to extract target blocks from the final volumetric scene to optimize the reconstruction computation and minimize the amount of empty space in the target blocks. A template-based block projection process then projects all the target blocks onto the projection plane to generate the image. Both raycasting and hardware assisted 3D texture mapping schemes are implemented. The algorithm is efficient in terms of both speed and memory requirements.

Purpose: The purpose of this study is to compare the accuracy of facial linear measurements obtained from volumetric spiral CT using 2D versus 3D reconstruction, and test the repeatability of these measurements. Material and Methods: The population consisted of 5 cadaver heads that were scanned to a Spiral CT scanner (120 Kvp and 200 mA, Toshiba Xpress S/X Toshiba-America, Medical System Inc., Tustin, CA) with high- resolution contiguous slices. Heads were scanned with 3 mm thick axial slices and a 2 mm/sec table feed. The CT data were archived on optical disks, and then transferred to a networked computer workstation (Sun Microsystems with Cemax version 1.4 software, Fremont, CA), to generate 2D and 3D images for manipulation and analyses. Repeated measurements were done on 2D and 3D images reconstructed from spiral CT scans on the workstation. Linear measurements were done by 2 observers with 2 sessions each, using several unique and conventional craniometric anatomic landmarks. The soft tissues were then partially removed and physical measurements of the same landmarks were repeated by an electromagnetic (3 space) digitizer (Polhemus Navigation Sciences Division, Mc Donnell Douglas Electronic Company, Colchester, VE). Analyses of variance were done to compare 2D versus 3D methods, and the accuracy of measurements between both imaging techniques. Results: The results showed statistically significant differences between 2D and 3D images for the majority of measurements. The 3D image measurements were not statistically different from the physical measurements. However, some of the 2D image landmarks differed from physical measurements. The repeatability of measurements was high by spiral CT-based craniofacial imaging. Conclusion: New computer graphics technology combined with 3D volumetric imaging by spiral CT can distinguish the craniofacial anatomy with greater accuracy than previously reported measurements and with greater accuracy than measurements from 2DCT images. These 3D measurements are essential to diagnostic and treatment planning of craniofacial injuries, anomalies and for craniofacial identification.

The purpose of this work is to characterize the 3D morphology of the bones of the rear foot using MR data. This work has two subaims: (1) to study the variability of the various computed architectural measures as a result of the subjectivity and variations in the various processing operations; (2) to study the morphology of the bones included in the peritalar complex. Each image data set utilized in this study consists of 60 longitudinal slices of the foot acquired on a 1.5 T commercial GE MR system. Our description of the rear foot morphology is based mainly on the principal axes, which represent the inertia axes of the bones, as well as on the bone surfaces. We use the live-wire method for segmenting and forming the surfaces of the bones. In the first part of this work, we focus on the dependence of the principal axes system on segmentation and on scan orientation. In the second part, we describe the normal morphology of the rear foot considering the four bones (calcaneus, cuboid, navicular, talus) and compare them to a population from the Upper Pleistocene. We conclude that this non-invasive method can be used in live patients to characterize the bone morphology or as a comparative method to classify population of bones. in spite of the variations involved in the various processing operations.

Prostate needle biopsy is used for the detection of prostate cancer. The protocol of needle biopsy that is currently routinely used in the clinical environment is the systematic sextant technique, which defines six symmetric locations on the prostate surface for needle insertion. However, this protocol has been developed based on the long-term observation and experience of urologists. Little quantitative or scientific evidence supports the use of this biopsy technique. In this research, we aim at developing a statistically optimized new prostate needle biopsy protocol to improve the quality of diagnosis of prostate cancer. This new protocol will be developed by using a three-dimensional (3-D) computer- based probability map of prostate cancer. For this purpose, we have developed a computer-based 3-D visualization and simulation system with prostate models constructed from the digitized prostate specimens, in which the process of prostate needle biopsy can be simulated automatically by the computer. In this paper, we first develop an interactive biopsy simulation mode in the system, and evaluate the performance of the automatic biopsy simulation with the sextant biopsy protocol by comparing the results by the urologist using the interactive simulation mode with respect to 53 prostate models. This is required to confirm that the automatic simulation is accurate and reliable enough for the simulation with respect to a large number of prostate models. Then we compare the performance of the existing protocols using the automatic biopsy simulation system with respect to 107 prostate models, which will statistically identify if one protocol is better than another. Since the estimation of tumor volume is extremely important in determining the significance of a tumor and in deciding appropriate treatment methods, we further investigate correlation between the tumor volume and the positive core volume with 89 prostate models. This is done in order to develop a method to estimate the tumor volume from the corresponding positive core volumes. Finally, we propose an algorithm for developing a statistically optimized prostate needle biopsy protocol. Preliminary experimental results are also presented.

Virtual colonoscopy, a recent three-dimensional (3D) visualization technique, has provided radiologists with a unique diagnostic tool. Using this technique, a radiologist can examine the internal morphology of a patient's colon by navigating through a surface-rendered model that is constructed from helical computed tomography image data. Virtual colonoscopy can be used to detect early forms of colon cancer in a way that is less invasive and expensive compared to conventional endoscopy. However, the common approach of 'flying' through the colon lumen to visually search for polyps is tedious and time-consuming, especially when a radiologist loses his or her orientation within the colon. Furthermore, a radiologist's field of view is often limited by the 3D camera position located inside the colon lumen. We have developed a new technique, called multi-planar geometry clipping, that addresses these problems. Our algorithm divides a complex colon anatomy into several smaller segments, and then splits each of these segments in half for display on a static medium. Multi-planar geometry clipping eliminates virtual colonoscopy's dependence upon expensive, real-time graphics workstations by enabling radiologists to globally inspect the entire internal surface of the colon from a single viewpoint.

Stereotactic ultrasound can be incorporated into an interactive, image-guide neurosurgical system by using an optical position sensor to define the location of an intraoperative scanner in physical space. A C-program has been developed that communicates with the Optotrak^TM system developed by Northern Digital Inc. to optically track the three-dimensional position and orientation of a fan-shaped area defined with respect to a hand-held probe. (i.e., a virtual B-mode ultrasound fan beam) Volumes of CT and MR head scans from the same patient are registered to a location in physical space using a point-based technique. The coordinates of the virtual fan beam in physical space are continuously calculated and updated on-the-fly. During each program loop, the CT and MR data volumes are reformatted along the same plane and displayed as two fan-shaped images that correspond to the current physical-space location of the virtual fan beam. When the reformatted preoperative tomographic images are eventually paired with a real-time intraoperative ultrasound image, a neurosurgeon will be able to use the unique information of each imaging modality (e.g., the high resolution and tissue contrast of CT and MR and the real-time functionality of ultrasound) in a complementary manner to identify structures in the brain more easily and to guide surgical procedures more effectively.

The properties of a novel dry medical recording system are presented in this paper. This system, developed through a cooperative effort between Sterling Diagnostic Imaging and Tektronix Corporation, is based on offset phase change solid inkjet technology and renders very stable, diagnostic quality medical images. A new technique of combining microdroplets of ink to construct grayscale is used and in most cases for a given gray level several possible combinations arise. Although these redundancies provide equivalent optical density, the resulting noise can vary considerably and criteria used to minimize the noise are reviewed.

We propose a method for improving the visualization of potential target regions in radiographic projection images. This is accomplished by modifying the intensity lookup tables (LUT) to compensate for, (1) cathode ray tube (CRT) dynamic range limitations and, (2) for contrast sensitivity characteristics of the human eye. We match the resulting LUT to selected regions of the image to optimize the use of the available intensity range.

In this paper we address the correction of the convergence error of a monochrome multi-beam Cathode Ray Tube (CRT) by means of digital video-signal processing. Correction of the convergence (horizontal misalignment) of the three electron beams with respect to each other in the CRT, will improve the resolution and brightness of the CRT. We apply the theory of fractional delay filtering to design a digital Finite Impulse Response (FIR) filter that is capable of interpolating the digital video signal. Emphasis is on small four-tap filters, to reduce the necessary amount of processing power. The variable filter has been implemented on a TMS320C80 signal processor to assess the performance of standard DSP hardware on this type of filtering. The analysis of four-tap fractional delay filters has led to useful designs in our application. The implementation on the TMS320C80 processor shows that the variable filter can be implemented with about 10 (parallel) instructions, yielding a maximum throughput of 16 M pixels/s on the TMS320C80 DSP at 40 MHz. We demonstrate (results of) a real-time DSP (TMS320C80) implementation of a variable delay video processing for horizontal convergence correction. The image quality, MTF and brightness are quite satisfactory and well in the diagnostic application area.

A unified lossless and lossy compression algorithm is proposed using integer-based reversible wavelet transforms. The proposed algorithm is based on improved pre-quantization, integer-based wavelet transform, and bit plane coding of wavelet bands with QM coder. From simulation, reconstructed image by the proposed algorithm shows higher peak signal-to- noise ratio (PSNR) compared to those obtained by float-point- based wavelet transform (Daubechies) and the JPEG. Peak errors in the reconstructed images by the integer-based wavelet transform are also much lower than those obtained by the JPEG and the float-point-based wavelet coding. In near lossless compression, lossless encoding of the wavelet coefficients of the pre-quantized image shows higher PSNR than those with the bit rate control by the bit plane-band ordering of the wavelet coefficients of the original image without pre-quantization. Using pre-quantization, peak error can also be uniquely controlled, which may be an important feature in medical image compression.

To achieve both high compression ratio and information preserving, it is an efficient way to combine segmentation and lossy compression scheme. Microcalcification in mammogram is one of the most significant sign of early stage of breast cancer. Therefore in coding, detection and segmentation of microcalcification enable us to preserve it well by allocating more bits to it than to other regions. Segmentation of microcalcification is performed both in spatial domain and in wavelet transform domain. Peak error controllable quantization step, which is off-line designed, is suitable for medical image compression. For region-adaptive quantization, block- based wavelet transform coding is adopted and different peak- error-constrained quantizers are applied to blocks according to the segmentation result. In view of preservation of microcalcification, the proposed coding scheme shows better performance than JPEG.

This paper presents a new region-based multi-scalar video coding algorithm, which employs an efficient spatial and frequency decomposition scheme within each frame. Spatial decomposition is implemented through a simple region growing block segmentation method, yielding arbitrary-shaped (relatively) homogeneous regions. Frequency decomposition is implemented by finding the optimum wavelet packet for each of the spatial regions by pruning a 3-level full wavelet tree. In order to adapt the coding to the motion within the sequence, that is, to exploit the temporal redundancy between frames, a differential pulse code modulation (DPCM) loop is used. The optimum coding is based on a rate-distortion criterion. The bit allocation to all subbands of the wavelet packet for each of the regions of a frame is done such that the overall distortion D is minimized under a bit rate R constraint. The constrained optimization problem is transformed into an unconstrained minimization of the Lagrangian cost function J equals D plus (lambda) R. It then follows that the optimum bit rate allocation is achieved when all subbands of all regions are operating at points with identical slopes, on their respective R-D curves.

Noise greatly degrades the quality of the image and the performance of any image compression algorithm. This paper presents an approach to representation and compression of noisy images. A new concept RBP region-based prediction model is first introduced, and then the RBP model is utilized to work on noisy images. In the conventional predictive coding techniques, the context for prediction is always composed of the individual pixels surrounding the pixel to be processed. The RBP model allows for the regions instead of the individual pixels surrounding the pixel to be predicted. A practical algorithm for implementation of RBP is developed. In our experiments, the practical algorithm is applied to noisy synthetic images. By encoding we achieve the bit rate 1.10 bits/pixel of the noisy synthetic image. The decompressed image achieves the peak SNR 42.59dB. Compared with the peak SNR 41.01dB of the noisy synthetic image, the decompressed synthetic image is better in the MSE sense. The image compression standard JPEG provides peak SNR 33.17dB for the noisy synthetic image at the same bit rate, and the conventional median filter with 3 by 3 window provides the peak SNR 25.89dB. The RBP model is also applied to a medical MRI image. The subjective evaluation of the decompressed MRI image shows that our method is superior to the conventional methods.

In this paper, we present an image compression scheme based on the automatic segmentation of regions of interest (RoI) and a lossy wavelet compression algorithm adapted to this segmentation. Quasi-lossless compression is applied to the RoI while lossy compression is allowed outside the RoI, preserving at best the visual quality of the decoded image within a defined RoI. In fact, for diagnostic accuracy purposes, quasi- lossless compression is often mandatory, while high compression ratio can only be achieved by lossy compression methods. The proposed technique is applied to heart MR images where the RoI is the entire heart. First, an unsupervised segmentation of the heart is performed in the original MR images, and the RoI is modeled by an ellipse fitted by means of a genetic algorithm. This model is defined by only 5 parameters, providing an efficient representation of the surrounding shape of the RoI. Compression is then applied using a wavelet-based multiresolution scheme. The quantization factor applied to the wavelet coefficients is adapted to the region and the subband, leading to the quasi-lossless compression in the RoI and lossy compression outside this RoI. The quantization difference between inside and outside RoI is also optimized for the desired compression ratio. Finally, the compressed bitstream is transmitted to the decoder together with the parameters of the RoI, allowing the reconstruction of the RoI surrounding shape in the decoder. Results of this RoI- based compression scheme are presented and further compared with the JPEG standard.

This paper presents a region adaptive encoding scheme with JPEG decoder compatibility. This scheme adequately changes the Lagrange multiplier in R-D cost function for each block according to the segmentation information so that the important region for diagnosis can be coded with less distortion. The results show that the proposed scheme not only outperforms original JPEG but also outperforms R-D optimal thresholding JPEG in important regions with respect to image fidelity.

One remaining barrier to the clinical acceptance of electronic imaging and information systems is the difficulty in providing intuitive access to the information needed for a specific clinical task (such as reaching a diagnosis or tracking clinical progress). The purpose of this research was to create a development environment that enables the design and implementation of advanced digital imaging workstations. We used formal data and process modeling to identify the diagnostic and quantitative data that radiologists use and the tasks that they typically perform to make clinical decisions. We studied a diverse range of radiology applications, including diagnostic neuroradiology in an academic medical center, pediatric radiology in a children's hospital, screening mammography in a breast cancer center, and thoracic radiology consultation for an oncology clinic. We used object- oriented analysis to develop software toolkits that enable a programmer to rapidly implement applications that closely match clinical tasks. The toolkits support browsing patient information, integrating patient images and reports, manipulating images, and making quantitative measurements on images. Collectively, we refer to these toolkits as the UCLA Digital ViewBox toolkit (ViewBox/Tk). We used the ViewBox/Tk to rapidly prototype and develop a number of diverse medical imaging applications. Our task-based toolkit approach enabled rapid and iterative prototyping of workstations that matched clinical tasks. The toolkit functionality and performance provided a 'hands-on' feeling for manipulating images, and for accessing textual information and reports. The toolkits directly support a new concept for protocol based-reading of diagnostic studies. The design supports the implementation of network-based application services (e.g., prefetching, workflow management, and post-processing) that will facilitate the development of future clinical applications.

Bone age assessment by a radiological examination of a hand and wrist image is a procedure frequently performed in pediatric patients to evaluate growth disorders, determine growth potential in children and monitor therapy effects. The assessment method currently used in radiological diagnosis is based on atlas matching of the diagnosed hand image with the reference set of atlas patterns, which was developed in 1950s and is not fully applicable for children of today. We intent to implement a diagnostic workstation for creating a new reference set of clinically normal images which will serve as a digital atlas and can be used for a computer-assisted bone age assessment. In this paper, we present the initial data- collection and system setup phase of this five-year research program. We describe the system design, user interface implementation and software tool development for collection, visualization, management and processing of clinically normal hand and wrist images.

X-ray mammography is an important diagnostic imaging modality for early detection of breast cancer. The early detection of the breast cancer can reduce the mortality of middle-aged women, especially in the developed country. Computer aided diagnosis (CAD) technologies have been developed to assist radiologists to detect breast cancer in early stage. This paper presents a KCAD (KAIST Computer-Aided Diagnosis) system for detection of breast cancer, which consists of personal computer, high resolution X-ray film scanner, high-resolution display and application softwares. There are three algorithms implemented in the application softwares. The first algorithm is the enhancement of the digitized X-ray mammograms based on the gradient operation. The second algorithm is to detect the clustered microcalcifications based on the statistical texture analysis, which is called surrounding region dependence method (SRDM). The SRDM matrix is computed for each ROI, which has 128 by 128 pixels. The SRDM matrix characterizes the small and high-density regions in mammograms, which can be recognized as microcalcifications. Four textural features are computed from the SRDM matrix. Using these features, the neural network classifies the regions as normal or microcalcification region. The third algorithm is the classification of the clustered microcalcifications as malignant or benign based on the shape analysis. The microcalcifications are segmented using SRDM. Four shape features are extracted from each microcalcification and five representatives are computed for each shape feature. Twenty-one shape-based values containing the number of microcalcifications are used to classify the region as malignant or benign. These algorithms are verified by real experiments.

Medical imaging workstations need capabilities for defining a Region Of Interest (ROI) within an image, so that it can be processed for visualization or quantitative measurements. The purpose of this work was to develop a generic, object-oriented toolkit which allows creation, modification, querying and traversal of an ROI. In the object-oriented design, an ROI class is defined with associated methods. An ROI object consists of a set of 3-D points represented using interval- end-points. The ROI toolkit has been developed using C++, in conjunction with patient browser and image object navigation (display) toolkits, to serve as the foundation for thoracic and neuro-radiology workstations. Because of the object-oriented design and generic representation of the ROI, this toolkit can easily facilitate standard and advanced image processing methods.

The digital breast imaging teaching file developed during the last two years in our laboratory has been used successfully at UCSF (University of California, San Francisco) as a routine teaching tool for training radiology residents and fellows in mammography. Building on this success, we have ported the teaching file from an old Pixar imaging/Sun SPARC 470 display system to our newly designed telemammography display workstation (Ultra SPARC 2 platform with two DOME Md5/SBX display boards). The old Pixar/Sun 470 system, although adequate for fast and high-resolution image display, is 4- year-old technology, expensive to maintain and difficult to upgrade. The new display workstation is more cost-effective and is also compatible with the digital image format from a full-field direct digital mammography system. The digital teaching file is built on a sophisticated computer-aided instruction (CAI) model, which simulates the management sequences used in imaging interpretation and work-up. Each user can be prompted to respond by making his/her own observations, assessments, and work-up decisions as well as the marking of image abnormalities. This effectively replaces the traditional 'show-and-tell' teaching file experience with an interactive, response-driven type of instruction.

The computing architecture and software developed to acquire and evaluate images from Cuban Giroimag MRI machines are presented. Giroimags computing systems are based on networked PCs forming a client/server architecture. The main principle applied in the system development is to take advantage of network computing and the power of modern personal computers. The goal proposed was to develop an open system, which permits to manipulate Giroimags from any networked computer and at the same time to give users all the possibilities to access every Giroimag scanning feature. The system developed eases the manipulation of Giroimag machines for image scanning and the resulting images could be accessed from every computer connected to the network. The proposed architecture could be easily integrated to Internet for remote image evaluation.

In recent years, CT fluoroscopy has been used for interventional procedures like needle biopsy, micro coil implantation for preoperative marking, and the guidance during ethanol injection. It takes about from 30sec to more than 60sec in those procedures. The radiation dose to a patient and a physician's hands has increased relatively to the general CT examination. And a patient would have to be exposed to a relatively large amount of x-rays even if it was desired to display only local region of interest. We have studied the local scan which limits an exposure field to avoid unnecessary x-ray, and have performed the basic experiments of the local scan using commercial scanner to compare the radiation dose with the conventional scan. And the absorbed dose estimation by the computer simulation was presented. It was proved quantitatively that the local measurements could significantly reduce the radiation dose. We have proposed a simple algorithm to recover the discontinuity of the locally measured data. It was shown that the artifacts caused by the boundaries of the exposure field can be eliminated by the algorithm.

The development of an efficient parallel hardware architecture suitable for CCD-mosaic digital mammography has been accomplished. This paper presents this architecture including both the analog and digital portions of the imaging hardware. A two dimensional array of CCD sensors are used to capture the mammographic image synchronously and simultaneously. Each CCD's analog signal is converted to a 12 bits/pixel digital value through an array of high speed analog-to-digital converters. A parallel array of mesh connected TMS320C40 DSP processors then takes in the digital image data simultaneously. The DSP's are used to precisely register the mosaic of individual images to form the final large format digital mammogram. Also, they are used to control CCD characteristics and parallel data transport to the viewing workstation. One master DSP is located on the workstation's PCI bus which controls the parallel DSP array and collects compressed image data through a 60MB/s port. Since all computations are performed in parallel using local memory on each DSP, the overall acquisition, image registration, and transmission to display of the final mammogram is performed in less than 30 seconds. This allows the physician to perform a preliminary observation of the patient's mammogram.

Convolution is a fundamental operation to many image processing algorithms and applications. One such algorithm is unsharp masking, which is widely used in medical imaging. A major component in unsharp masking is the computation of a lowpass-filtered image, e.g., via generalized convolution with a Gaussian filter or via specialized convolution with a boxcar filter. Generalized convolution is computationally expensive, e.g., convolution with a 3 X 3 kernel on a 512 X 512 image takes 1.45 sec on SUN SparcStation 20/71. In order to achieve faster computation in convolution, hardwired solutions with ASICs and/or fixed-function chips with little programmability have been traditionally used. The disadvantages associated with hardwired implementations are that they are rigid, uni-functional and not upgradable. Our approach has been programmable convolution, which is flexible, multi-functional, easily-upgradable and has a performance comparable to the hardwired implementations. This paper describes efficient software implementations of both generalized and boxcar convolution on a programmable multimedia processor, the Texas Instruments TMS320C80, also known as Multimedia Video Processor (MVP). Using the MVP's advanced digital signal processors (ADSPs), instruction-level parallelism and intelligent input/output interface, we have been able to significantly improve the performance of both generalized and boxcar convolution. For a 512 X 512 8-bit image, generalized convolution takes 19.5 ms for a 3 X 3 kernel. While the boxcar convolution has similar performance for a 3 X 3 kernel, the performance improvement by a factor of up to 13 has been achieved for large-size kernels such as 21 X 21. Our implementation of convolution algorithms on programmable mediaprocessor clearly demonstrates the feasibility of software-based approach.

The EMERALD Project has the objective to develop and exploit multimedia services for medical imaging environments over broadband networks. The partners include hospitals, technical universities, research organizations and industrial companies of the medical imaging scene. The project has focused on the exploitation phase (2 years), although a thorough user requirement analysis was performed, followed by the well-known specification and development activities. The development of the technical platform offering the services was based on an existing ATM Telemedicine platform developed earlier by GBT- UPM, one of the EMERALD partners. The architecture and the services were enhanced or newly developed based on the generic concept of the Open Medical Imaging Telemedicine Platform, which is also presented in this paper.

CT/MR scan data presentation (layer by layer) strongly resembles the slice data format used to drive the new manufacturing paradigm of rapid prototyping (RP). Encouraged by the similarity, computer translators were developed to convert the CT/MR data from their virtual reality into a format whereby RP could be used to produce anatomically accurate physical models ('Real Virtuality'). The precision of RP technology is driving the refinement of the algorithms for locating surfaces and features from the CT/MR data. The RP models have been found to have particular value in many applications, such as diagnostics, treatment planning, procedure practice/simulation, and consultation. Inaccuracies in the RP models result from the limited resolution of the CT/MR data and limitations with current data interpretation algorithms. This paper will illustrate current applications for these physical medical images and research efforts into generating more accurate surface data.