Kelby K. Chan

Visualization and volumetric compression

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

Radiological image compression using full-frame cosine transform with adaptive bit-allocation

We report a new bit-allocation scheme based on the full-frame cosine transform for radiological image compression. The new technique differs from a previously reported method in its use of a two-dimensional bit-allocation table to encode the compression data. This allows for an improved treatment of high frequency components in the transform domain. Consequently, it has the capability of faithfully reproducing limited numbers of high-contrast sharp edges in the image. Previously reported artifacts, induced in the reconstructed image by sharp edges in the original, have been eliminated. Experiments with 10 radiological chest images show almost no perceivable degradation in the reconstructed image at compression ratios below 10:1. Image quality at a fixed compression ratio is, in every case, comparable or superior to results using the old method. Furthermore, the new algorithm lends itself to hardware implementations that are both simple and fast.

Performance characteristics of an ultrafast network for PACS

Three difficult problems in making picture archiving and communication systems (PACS) a clinical reality in radiology are image archiving, very high-resolution display stations, and high-speed networking. This paper considers high-speed image transmission through a high- capacity network. Several commercially available high-speed networks were tested over the past year. Only one of these networks (UltraNet) has adequate throughput and capacity potential necessary for the PACS used in the test. The focus of this experiment is to determine the throughput and capacity characteristics of this star topology networking scheme as relates to the operation of a PACS in the clinical environment. A large-scale test was performed to gauge network memory-to-memory performance for three networking configurations modeling those in a PACS: duplex, parallel and relay. Ten computers used in the PACS (Sun 3 and 4 class computers) were connected with UltraNet for the test. For point-to-point throughput (half-duplex model) the network delivers up to 3.1 megabytes/second (MBps) for Sun 3/computers and 4.7 MBps for the Sun SparcServer 490. As regards capacity considerations (parallel model), five parallel image transfer processes generated a maximum of 13.9 MBps through the network. Only a slight degradation in individual process throughput was observed (1.4%). With regard to shared access to high-contention resources on the PACS network (e.g., archive servers), this network demonstrated equal sharing of server networking capacity between various client computers (relay model). For disk-to-disk performance measurement under loaded clinical conditions between two SparcServer 490s, the overall average transfer rate was found to be 1.125 +/- 0.257 MBps, while the average network transfer rate of 3.632 +/- 1.542 MBps was determined. This compares to an average overall transfer rate of 0.389 +/- 0.061 MBps and average network transfer rate under similar conditions of 0.568 +/- 0.060 MBps using Ethernet. For disk-to-memory transfer from a parallel transfer disk (PTD) on a Sun 4/470 to the 2K display frame buffer on a Sun 4/370, the PTD portion of the transfer required 1.2 seconds (6.7 MBps) and the network portion required 1.04 seconds (7.7 MBps) for an overall transfer rate of 3.6 MBps.

Three-dimensional transform compression of images from dynamic studies

Transform based compression methods achieve their effect by taking advantage of the correlations between adjacent pLtels in an image. The increasing use of three-dimensional imaging studies in radiology requires new techniques for image compression. For time-sequenced studies such as digital subtraction angiography, pixels are correlated between images, as well as within an image. By using three-dimensional cosine transforms, correlations in time as well as space can be exploited for image compression. Sequences of up to eight 512 x 512 x 8-bit images were compressed using a single full volume three-dimensional cosine transform, followed by quantization and bit-allocation. The quantization process is a uniform thresholding type and an adaptive three-dimensional bit-allocation table is used. The resultant image fidelity vs. compression ratio was shown to be superior to that achieved by compressing each image individually.

Automatic Acquisition of CT, MR, and US images for PACS

Existing radiological imaging devices are usually not designed for contemporary network communication. Thus special developments are necessary for their integration into a Picture Archiving and Communication System (PACS). Requirements and concepts for such developments are discussed demonstrating automatic data acquisition from computed tomography (CT), magnetic resonance imaging (MR), and Ultrasound diagnostic (US) systems at the University of California Los Angeles (UCLA). The operation of the data transmission depends on the networking capabilities of the imaging device and the accessibility of its internal data structure. While the main frame computers of advanced CT and MR scanners allow an automatic on-line data transfer during clinical operation current ultrasound devices need additional modules for digitizing and transmission of image data. Depending on the system design of the imaging device the on-line acquisition can interfere with the clinical operation whereas an off-line data transfer compromises the timely performance of the image communication system.

Systems integration for PACS

A successful PACS (Picture Archiving and Communications System) implementation requires an eclectic integration of a number of key technologies. Among these are equipment interfaces, communications, storage, and display. Coincident with this, the software architecture must support a distributed system of heterogeneous structures, provide for protocol and format conversions to a unified system standard, be scalable to accommodate expansion, and provide a measure of fault tolerance. In this paper we survey the current state of the UCLA PACS components and architecture.

Data Storage and Compression

The PACS data files migrate and reside among a variety of storage devices during their life-cycle. High-capacity Winchester disks are used ubiquitously. High-speed disks such as parallel transfer disks and disk arrays are often configured in the diagnostic display workstations. Optical disk jukeboxes provide the central archive. This paper reviews the current storage technologies and also discusses the data compression techniques applicable to the radiological images.

Multiple communication networks for a radiological PACS

The authors have implemented a communication network connecting multiple buildings for their picture archiving and communication system (PACS) in the Radiology Department at UCLA. The network consists of three types of local area networks (LANs) and a 1.0-km fiber-optic link connecting the outpatient and inpatient facilities. Images from radiologic imaging devices (4 CT scanners, 5 MR scanners, 4 CR units and 5 film digitizers) are transmitted to the acquisition computers via the Ethernet LAN. The fiber distributed data interface (FDDI) LAN then provides data communication among the cluster controllers, the acquisition computers, and the database servers. A 1-gigabit UltraNet LAN is used to route images from the cluster controllers to remote display workstations. All inter-building connections are through fiber-optic cables. Among these multiple networks, Ethernet offers multi-access to the multimodal PACS in image acquisition, FDDI controls a fast data flow so that all acquired images have a shorter residence time on local disks, and UltraNet provides high-speed transfer of images from the cluster controllers to the display workstations. The three-tiered functionality of Ethernet, FDDI, and UltraNet eliminates network traffic bottlenecks and hence provides high performance in image communication. The delay time of a 2K X 2K X 8-bit CR image (4 MBytes) from acquisition to display is less than 5 minutes. In addition, the standard Ethernet serves as a backup to guarantee network connectivity of the entire PACS.

On-line acquisition of CT and MRI studies from multiple scanners

Currently, three GE CT9800 and three GE Signa Advanced scanners are connected to the picture archiving and communication system (PACS) at the University of California Los Angeles (UCLA) Medical Center. While the MRI systems are equipped with a standard Ethernet interface, special development efforts were necessary in order to utilize the data port of the CT system for an automatic data transmission into the PACS network. For both modalities designated computer stations accommodate the acquisition of the image data through a special protocol and transmit the data in UCLA PACS standard format to the central controller for archiving and distribution to the display stations. The MRI images are acquired as they are generated through a capture process of the acquisition station that has direct access to the internal database of the MRI system. The transmission of the CT image data, however, is governed by the CT system computer and is limited to a low priority process in between clinical operations. Average MRI studies become available for review within thirty minutes, whereas CT studies are delayed up to two to three hours depending on the load of the CT system.

Development of an image segmentation and registration algorithm on a SIMD parallel processor

Currently, more sophisticated image processing operations are not used on clinical workstations or CR (computed radiography) systems. One reason for this is the excessive time required to perform the image processing. One parallelism approach for improving image processing speed is to utilize a single instruction stream - multiple data stream (SIMD) computer. This paper presents the preliminary results for two classes of image processing algorithms which are useful in radiological applications: segmentation and registration.