| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Computers in Radiology |
1 Both authors: Department of Radiology, Stanford University Medical Center, MC 5105, 300 Pasteur Dr., Rm. S-056, Stanford, CA 94305.
Received August 21, 2000;
accepted after revision October 3, 2000.
Address correspondence to C. F. Beaulieu.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CONCLUSION. The mean attenuation of peripheral tendon (
100 H)
is distinctly higher than that of muscle (60 H) enabling high-quality
volume rendering of muscle—tendon—bone relationships with
unenhanced CT. High-frequency (bone) CT reconstruction algorithms commonly
used for extremity CT produce approximately twofold higher image noise and
inferior three-dimensional renderings compared with those based on less noisy
standard or soft-tissue reconstruction algorithms. These concepts can be used
to uniquely reveal tendon—muscle—bone relationships for clinical,
scientific, and educational purposes.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1000
H), attenuation differences among structures such as tendon and muscle are at
least an order of magnitude smaller and make 3D rendering difficult. In
addition, shaded-surface displays, previously the most commonly used technique
for 3D rendering [1], are
limited in that all structures above the selected threshold are rendered as a
single object and make distinction between bone and soft-tissue structures
impossible. Technical advances in CT and the emergence of volume-rendering
computer graphics [2] now make
simultaneous display of soft tissues and bone feasible. The resultant images
are unique. The purpose of this study was to quantify the attenuation values
of the relevant structures on unenhanced helical CT and to develop specialized
opacity transforms for high-quality 3D volume rendering of
tendon—muscle—bone relationships. |
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Attenuation Measurements
We initially evaluated CT data from eight studies primarily represented by
images reconstructed with the bone kernel. After recognizing the relatively
higher image noise and its resultant deleterious effect on 3D reconstructions,
we concentrated measurements on wrist studies from two subjects in which data
from both bone and standard reconstructions were available. Attenuation
measurements were made with VoxelView 2.5 software (Vital Images, Minneapolis,
MN) on the original axial CT data displaying the full 12 bits of grayscale
resolution (-1000 to +3096 H). Attenuation values were recorded on a
point-by-point basis from bone, muscle, and tendon structures represented in
the images. In applying region-of-interest measurements with circular regions,
we found that the small cross-sections of tendon made measurements less
reliable than with point measurements. For bone, we concentrated measurements
on cortical bone for the radius, ulna, and metacarpals, in which a distinct
cortex was visible. Primarily cancellous bones such as the carpals were also
measured, though their mean attenuation tended to be lower than that of
cortical bone because of interspersed marrow fat among the trabeculae. Muscle
measurements were performed on larger muscle bellies such as the pronator
quadratus. Tendon attenuation measurements were made in multiple tendons
within each data set. We avoided measurements in areas in which beamhardening
artifact or patterned noise had an obvious spurious effect on the results. For
the data sets in which both bone and standard reconstruction images were
available, measurements on similar areas of bone, muscle, and tendon were
performed. From a pool of approximately 3000 measurements, mean attenuation
and standard deviations were computed. In addition, we performed anecdotal
measurements on cortical bone, muscle, and tendon around the ankle, elbow,
fingers, and knee. Whereas no systematic regional or intersubject differences
were observed in the peripheral extremities, images from central areas such as
the hips or shoulders did not usually show discrete tendons. Relatively larger
muscle bellies, compared with their tendons, and possibly higher image noise
were thought to be the primary factors accounting for this difference.
Three-Dimensional Volume Rendering
Volume renderings were produced with Voxel-View. Although we have the most
experience with this software because it is available in our laboratory, other
commercial volume-rendering software such as Vitrea (Vital Images), Advantage
Windows (General Electric Medical Systems), Virtuoso (Siemens Medical Systems,
Munich, Germany), and Zio (Zio Software, Tokyo, Japan) should be similarly
capable. VoxelView allows a wide range of control over rendering parameters
needed to depict the relatively small attenuation differences between tendon
and muscle and incorporates a special lighting model to produce high-quality
renderings. Given our desire to use the lighting model and to depict
relatively subtle (40 H) differential attenuation between muscle and
tendon, we took advantage of the capacity to remap the data from 12-bit to
8-bit gray-scale resolution. In this process, we truncated the original CT
data between -200 and 1024 H and mapped the remaining range to 8 bits (256
levels) of gray scale. In our application, one advantage of this remapping is
that finer control over the opacity transform is possible, allowing emphasis
between subtle attenuation differences. A second feature of this remapping is
that when the lighting model is applied, the available second byte of
VoxelView data is used for a "fast" lighting model. Because this
model uses the gray-scale information in the image to generate lighting
vectors, gray-scale information itself is obscured. The means to retrieve this
density information is to apply a color lookup table that systematically maps
gray scale to hue (Fig. 1). In
constructing the color map, we selected hues that produced renderings
depicting muscle and tendon as reddish and bone as white. These colors most
closely approximated those that occur naturally in human tissue.
|
After tabulation of the mean values and ranges of attenuation of structures of interest, we created customized lookup tables to serve as opacity transforms for volume rendering. These tables were designed to make bone maximally opaque, tendon moderately opaque, and muscle semitransparent. Fat and air-attenuation ranges were rendered fully transparent. In regions of the attenuation histogram between densities of interest, we mathematically interpolated opacity values to create bell-shaped opacity transforms for each muscle, tendon, and bone.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Figure 1 shows an axial
image of the wrist, its corresponding attenuation histogram, and the custom
opacity transform we implemented. Tendon has visibly higher attenuation than
that of muscle. The fact that the attenuation difference is small, however, is
reflected in the histogram in that the ranges representing muscle and tendon
overlap and, thus, do not create individual peaks. But by taking advantage of
the quantitative attenuation data, an opacity transform emphasizing each of
these voxel subsets could be created. For example, the second peak in the
opacity transform is centered about the mean attenuation of tendon and allows
their display as distinct structures relative to less opaque muscle (first
peak) and bone (high-opacity portion at all attenuation values above 150
H). Compared with images reconstructed with standard algorithm, data
reconstructed with the bone algorithm had attenuation histograms in which
there was even less distinction between muscle and tendon, making application
of our specialized opacity transforms less effective in showing muscle,
tendon, and bone simultaneously.
Visual differences between 3D images resulting from data reconstructed with standard and bone kernels are illustrated in Figure 2A,2B. In Figure 2A, data were reconstructed with the standard algorithm and resulted in lower noise and less overlap between the attenuation histograms of muscle and tendon. In this reconstruction, muscle can be displayed as semitransparent, tendon as relatively opaque, and bone as fully opaque. Whereas the lower resolution standard or soft-tissue algorithms decrease spatial detail and reduce image noise, at the acquisition settings we used for clinical extremity CT, this kernel produces substantially more effective 3D renderings. In Figure 2B, data were reconstructed with the bone algorithm, and the rendering was grainy and made distinction between tendon and muscle difficult.
|
|
With appropriate source data, our rendering method can also be applied to other body regions. Figure 3 shows application to unenhanced CT of the ankle. Although extensive measurements across multiple joints and patients have not yet been done, we generally observed that the attenuation of peripheral tendon is relatively constant, so that our rendering methods should be broadly applicable. Note that in Figure 3, muscles have been rendered relatively more opaque than those on the wrist images shown in Figure 2A,2B. Blood vessels, which have similar attenuation to that of muscle, are also depicted in the 3D volume renderings.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Some radiologists and a significant number of orthopedic surgeons have embraced 3D display of CT data as a means to visualize bony relationships in operative planning. Because of previously limited computing power, most studies have used shaded-surface—display computer graphics, in which the bone surface is derived by thresholding, eliminating voxels below a selected attenuation threshold. When a bone-attenuation threshold is selected, muscle and tendon structures are effectively discraded from the data. With thresholds lower than bone, some surface displays of tendons in the area of the wrist and ankle have been reported [1, 7]. A limitation of this method, however, is that the binary nature of shaded-surface displays eliminates the ability to retain gray-scale variations among the structures above the threshold, so all structures are depicted as monochromatic.
Advances in software and graphics hardware have led to the development of more sophisticated methods of data display in the form of continuous volume rendering [2, 8, 9]. Over the past 10 years, volume rendering has evolved from a slow impractical labor-intensive method to a viable clinical tool [9]. Volume rendering of CT scans has practical musculoskeletal applications [10], although, to our knowledge, it has not yet been applied to the diagnosis and characterization of tendon injury. The key advantage of volume rendering is that spatial relationships between muscle, tendon, and bone can be simultaneously displayed while preserving inherent gray-scale information. Among a growing number of volume-rendering systems programs now commercially available, there are variations in methodology for applying opacity transforms, color mapping, and generation of lighting models. In the system we used, variations in hue (color), according to the attenuation histogram, are used to retain CT density information. This point is important because although the colors themselves are chosen by the user, their assignments are systematic and lend meaning to variations in hue as they map directly to variations in CT attenuation.
Generally, clinical extremity CT uses a "bone," "sharp," or "ultrahigh" reconstruction kernel, the specific nomenclature of which varies among scanner manufacturers. These kernels produce the necessary high-frequency spatial information leading to sharpness and the ability to diagnose subtle abnormalities. For a given CT acquisition, the disadvantage of these kernels relative to those used, for example in abdominal imaging (standard or softtissue kernel), is that high-frequency kernels generate images with higher noise (Table 1 and Fig. 2A,2B). The implication of relatively higher noise is that for structures with a given mean attenuation value (the mean itself being little affected by the reconstruction kernel), more noise broadens the distribution of attenuation values around the mean. As a result, with images from the high-frequency kernel, it is difficult to selectively show low-contrast structures, which are separated by only small mean-attenuation differences. It is thus not surprising that we found that 3D renderings derived from bone kernel source images provided inferior low-contrast structure delineation and less smooth images than those with the standard algorithm (Fig. 2A,2B). Because the source images with the standard reconstruction kernel have lower spatial resolution than that of bone kernel images, we now reconstruct both standard and bone kernel images for extremity CT in which 3D rendering is desired. Bone kernel images (axial and multiplanar reformations) are used for the primary diagnosis, and standard images are used only for 3D imaging. In a filmless environment, the only expense in generating extra images is scanner reconstruction time and image storage space.
Given the novelty of the display method shown, we have relatively little direct evidence of the clinical efficacy of the postprocessing performed. Anecdotally, orthopedic and hand surgeons have indicated that the displays are useful to them in understanding complex spatial relationships between tendons and adjacent fractures. Potential direct uses include the display of disrupted or displaced tendons that may occur with displaced distal radius fractures. In calcaneal body fractures, entrapment of the peroneal tendon could be directly visualized. In planning tendon transfers for developmental anomalies, we also believe that these types of displays will help determine if anticipated procedures will be feasible. Finally, significant interest in the high-quality CT data and displays comes from biomechanical engineers, who expect to take advantage of the techniques in mathematic modeling of forces transmitted across joints in the pathogenesis of degenerative arthritis, for example.
Advantages of unenhanced CT with volume rendering are that it uses widely available CT technology, that it does not require IV contrast material, and that data acquisition can be performed in a short time. Although volume rendering and a suitable workstation are necessary, these systems continue to become more powerful and less expensive, so cost and processing time are unlikely to be limiting factors. Limitations primarily are that the technique has not had significant clinical testing to determine its incremental value, as previously discussed. When we compare this method with MR imaging, there is no question that the depiction of soft-tissue structures themselves is superior on MR imaging. We have no information yet on whether relatively subtle intrasubstance or peritendinous abnormalities will be appreciable on unenhanced CT. In addition, concurrent edema or hemorrhage around a fracture site might obscure the inherent attenuation differences we showed between muscle and tendon. One might also argue that similar displays could be achieved with 3D MR imaging data. In our experience, it is possible to create useful 3D volume renderings of MR data, but the fact that signal intensities between bone and soft-tissue structures vary with the pulse sequence makes processing more difficult than with CT data. Another factor that compounds the difficulty in 3D rendering of MR data is that absolute signal intensities vary across images as a result of sensitivity profiles of the radiofrequency coil. This variation makes any "standardized" remapping of signal intensities to opacity transforms impossible. With CT, regional variations in density may occur because of artifacts such as beam hardening, but in general, there is consistency between subjects and body regions. With highquality source CT data and the use of an opacity curve similar to ours as a starting point, we have been successful in applying our rendering techniques to other peripheral joints in time periods of less than 5 min, suggesting feasibility in a clinical environment. Although we used an offline workstation, the continued evolution of scanner consoles and picture archiving and communication system (PACS) should enable direct production of similar images in the near future.
In conclusion, we have shown that unenhanced extremity CT scans show small, but distinct, attenuation differences between muscle and tendon. By taking advantage of these differences and appreciating the impact of CT reconstruction algorithms on volume-rendering performance, we have illustrated how to create unique 3D depictions of musculoskeletal structures. Because both the CT and computer technologies underlying these techniques are now widely available, we expect that further application will show that the displays are of clinical, scientific, and educational value.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  K. Ohashi, J. M. Restrepo, G. Y. El-Khoury, and K. S. Berbaum Peroneal Tendon Subluxation and Dislocation: Detection on Volume-rendered Images--Initial Experience Radiology, January 1, 2007; 242(1): 252 - 257. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. H. Choplin, K. A. Buckwalter, J. Rydberg, and J. M. Farber CT with 3D Rendering of the Tendons of the Foot and Ankle: Technique, Normal Anatomy, and Disease RadioGraphics, March 1, 2004; 24(2): 343 - 356. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Y. El-Khoury, D. L. Bennett, and G. J. Ondr Multidetector-Row Computed Tomography J. Am. Acad. Ortho. Surg., January 1, 2004; 12(1): 1 - 5. [Full Text] [PDF]
|
|
|
|
 |
|
 K. Ohashi, G. Y. El-Khoury, and D. L. Bennett MDCT of Tendon Abnormalities Using Volume-Rendered Images Am. J. Roentgenol., January 1, 2004; 182(1): 161 - 165. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
