Up: Title Page Contents: Conference Page

Text

Advanced imaging technologies are fast becoming integral to medical education, medical training, diagnosis and surgery simulation. Other advances in telecommunications will provide the necessary infrastructure for real-time transmission and interaction with 3-D images for teaching, diagnosis and consultation purposes in support of virtual collaboratories, virtual hospitals and electronic housecalls. One impediment to these applications has been the cost of graphics workstations required for visualization and manipulation of the high-fidelity, large database 3-D images needed by educators and medical practitioners. Such workstations and virtual environment hardware for stereoscopic imaging with manipulation capability are typically high-end machines with equally high price tags. We have been working to show the practicality of high-fidelity imaging on low cost machines. This report documents our imaging of visible human and other, patient-specific data on PC platforms. While current PC computer technology does not permit manual rotation or other manipulation of high fidelity stereoscopic images at the desired minimum of 15 frames/sec, it does permit them to be displayed and rotated within a fraction of a second to several seconds, with the actual time dependent upon the size of the data set. This indicates that within a year or two high fidelity images can be sent over a high speed internet for diagnosis and consultation, to and from remote sites using a multiplicity of platforms.

For this research, the Visible Human female CT data set and patient-specific data from a variety of sources were collected. These included computed tomography (CT), magnetic resonance (MR) and echocardiography. After formatting scans with DICOM, volumetric segmentation was followed by mesh generation that took advantage of every detail in a 1 mm slice. The software for these processes was developed in the NASA Ames Research Center’s Biocomputation Center. For 1 mm scans, resolution is currently 1 mm in the z direction and .3 to .5 mm in the x-y directions. For visualization, the reconstructions are displayed using the SGI OpenGL graphics library. Surfaces can be visualized as shaded solids, or as semi-transparent if one wishes to see features of objects under a surface. The images, once reconstructed, can be transected from any angle desired by the use of clipping planes. For real time rendering during manipulation, changes in the x, y and z directions must all be computed, increasing the computational load. Stereoscopic imaging requires a doubling of the data set to display left-right eye images simultaneously, and a corresponding increase in computational load. Thus, stereoscopic high-fidelity visualizations for virtual environment surgery simulators and realistic manipulations of the images are currently limited by the size of the data sets.

Our stereoscopic images were originally visualized on a virtual environment Immersive WorkBench (Fakespace) using a Silicon Graphics (SGI) Onyx graphics workstation, a transmitter and CrystalEyes (StereoGraphics, Inc).The images can be rotated or cut into using keys on the keyboard or a stylus and special software developed in the Biocomputation Center. The cost of this kind of hardware for developing virtual environment technologies can be one hundred thousand to several hundred thousand dollars. More recently, we have visualized our stereoscopic images on a Gateway 2000 PC with a 333 MHz Pentium II processor, 192 MB of system memory and 8 MB of video memory, using a Diamond FirreGL 1000 computer graphics card and a ViewSonic PT813 monitor. The entire system cost ~$3,500. The images displayed on this system can be rotated using right and left keys on the keyboard, or tilted using up and down keys. An alternative method is to employ a joystick.

The images visualized ranged in size from ~ 30,000 polygons (<1MB of data) in the case of a small vascular tree to 4 million polygons for a face and skull of a youth (~195 MB). Here we concentrate on skeletal (skull) and heart data taken from the Visible Human Female and from patients. The heart contains ~ 300,000 polygons (~15 MB), the skull of the Visible Human Female has two million polygons (~100 MB), and the infant skull, 1.2 million polygons (~84 MB). Examples of skulls reconstructed from the Visible Human Female CT data set and from a patient‘s CT scan, visualized on a PC as well as on the Immersive WorkBench, are shown in Figure 1 and Figure 2. These data sets are too large for real-time manipulation using conventional workstations. On the WorkBench, the baby’s skull can be rotated at about 1 frame/sec and the Female skull at about 2 frames/sec. On the PC, rotation times are lengthened.

The baby skull can be rotated at about 1 frame/8 sec and the female skull, one frame/15 sec. In contrast, the small data set for the vascular tree can be visualized stereoscopically and manipulated in real time on a PC.

The size of the data sets also limits interaction with the images when they are transmitted across the Internet to and from remote sites. However, the bandwidth of the networks provided by federal and commercial sources are constantly increasing. This will mean that virtual collaboratories and virtual hospitals spanning the globe could become commonplace within 3-5 years, assuming that graphics and transmission capabilities are ready. Additionally, low-Earth orbiting satellites will soon be in place to provide digital, high bandwidth, no-lag communications. This will make it possible to utilize transportable medical diagnostic and satellite transmission/reception equipment to reach patients at any remote site around the world. The graphics software we are developing and implementing will prove useful in these environments. Executable code will be made available as the research continues.

While the research we are engaged in has spinoffs for the medical and education communities, it is also important to NASA. The heart is known to undergo changes in weightlessness. It will be necessary to have methods in place for frequent heart monitoring in the space environment, both on space station and on spacecraft traveling outside of Earth’s gravity to the moon or to Mars. The most efficient monitoring can be carried out using echocardiography with immediate digitization of data and formatting, followed by 3-D imaging using a PC. The images can be studied by a physician in space, and relayed to the ground for consultation. Telemedicine capabilities will play a substantial role.

The Visible Human data sets have been important in the development of imaging technologies such as those briefly described here. They provided excellent resources of entire male and female bodies for imaging research. Such complete data sets are not obtainable from patients. The problems in differing fields of view employed during collection of CT data from the head of the Visible Human Male, clipping of the nose in the Visible Human Female CT data set, and brain artifacts in both the male and the female are far outweighed by the value of complete data sets. Of further value are CT and MRI scans, and physical sections, of the same individuals. These data sets are a reference for tracing the origins and insertions of muscles, for showing the relationships between blood vessels and nerves, and for demonstrating the 3-D relationships between body parts in head, limbs and trunk. The data sets provide a roadmap to the future teaching of morphology, whether for biology or medicine, and stimulate creative approaches to imaging of 3-D data. They have some usefulness in devising surgery simulators, where the Visible Human data sets can be used during software development. However, for planning of specific surgeries, or for consultation purposes, the real patient’s data must be used. High fidelity, interactive imaging using the least expensive equipment is a goal in this endeavor.

Up: Title Page Contents: Conference Page