| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Technical Developments |
1 From the Department of Radiology, Stanford University School of Medicine, 300 Pasteur Dr, Stanford, CA 94305-5105. From the 2001 RSNA scientific assembly. Received April 7, 2002; revision requested June 11; final revision received January 18, 2003; accepted February 10. Supported by National Institutes of Health grants 5RO1HL58915 and 1RO1HL67194. Address correspondence to G.D.R. (e-mail: grubin@stanford.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
Supplemental material: radiology.rsnajnls.org/cgi/content/full/2291020370/DC1
© RSNA, 2003
Index terms: Computed tomography (CT), angiography, 9*.129172 • Computed tomography (CT), maximum intensity projection, 9*.12917
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
To produce MIPs, a viewing angle is chosen to define the projection plane. Parallel rays are then cast from the projection plane through the stack of reconstructed sections that make up the data volume, and the maximum intensity encountered along each ray is placed into the projection plane to construct the MIP. Vessels have higher contrast intensity values than those for soft tissue; therefore, the MIP shows a projected two-dimensional view of the vessels as seen from the center of the projection plane (3). Since some information is lost in the conversion from three to two dimensions, MIPs can be computed from many viewing angles and shown in a cine loop to convey the three-dimensional anatomy of the vessels (1,3).
The contrast in MIPs decreases with increasing projected volume (MIP thickness) because the probability that the maximum value encountered in the background will match or exceed the vessel intensity increases with MIP thickness. Although MIPs exhibit an increased contrast-to-noise ratio compared with that of source images, predominantly as a result of decreased noise (8), the reduced contrast between vessels and background can result in artifacts. This effect can lead to the disappearance of vascular features that have intensities only as great as the intensity of the background. Therefore, small vessels, which have decreased intensity as a result of volume averaging, can become invisible. The edges of larger vessels, which are less intense than the vessel center because of volume averaging, may be obscured, which leads to apparent vessel narrowing (7). High-grade stenoses may be overestimated on MIPs and appear as segmental vessel occlusions (9,10).
Regions of interest (ROIs) can be defined around vessels to limit the MIP thickness, thereby improving contrast in the MIP. In CT angiography, this method also allows the exclusion of high-attenuating bone that otherwise could overlap and obscure the vessels. A rectangular oblique plane can be easily specified and thickened to enclose a cuboidal ROI that can be used to produce conventional rectangular-slab MIPs, which are also known as thin-slab MIPs. In regions of complex and tortuous anatomy and for certain viewing angles, however, cuboidal ROIs cannot maximally exclude bone (1) and may include excessive soft tissue. Usually, separate cuboidal ROIs have to be specified for each vessel of interest, which increases the number of MIP reconstructions per study. Alternatively, manual section-by-section editing can be performed to draw ROIs around structures to exclude or include them, but this is tedious, may not be reproducible, and may be susceptible to tracing errors (1).
We developed and evaluated a semiautomated method for producing MIPs that is based on an automatically specified branching curved-slab MIP that adaptively encloses vessels of interest while excluding bone and surrounding soft tissue.
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
Next, a list of centerline points is obtained by sampling the median centerline at subvoxel intervals. At every point in this list, we determine the thickness of the vessel in the view direction at that point by using an adaptive threshold. From the curved median centerline, we then project a curved plane perpendicular to the viewing angle. By thickening this plane, a curved slab is defined. By using the calculated vessel thickness along the centerline, we vary the thickness of the slab along its length to enclose the minimum volume necessary to include the vessel (Fig 1; Movie 1, radiology.rsnajnls.org/cgi/content/full/2291020370/DC1). A minimum slab thickness of 4 voxels is enforced to ensure that areas of tight stenoses, which might have a zero thickness measurement, will be fully included in the curved slab. This curved slab is then used as the ROI for subsequent calculation of the curved-slab MIP.
|
Rectangular-slab MIPs are usually produced to depict one vessel and its branches or in some cases only part of a tortuous vessel. This results in multiple MIPs that have to be computed, reviewed, and stored. To reduce the number of images, our algorithm can produce composite curved-slab MIPs that include many vessels coursing through different parts of the body. To effect this, the user specifies the starts and end points of all vessels to be processed. Median centerlines are produced as described previously. After this, the user is able to select the viewing direction and which vessels will be included in a particular composite curved-slab MIP. Curved slabs are then defined for the selected vessels. To avoid overlap between the curved slabs from adjacent vessels, the width of the curved slab is also varied along the length of the vessel by using a similar adaptive threshold algorithm. The resulting composite curved slab excludes bone and extraneous soft tissue. The composite curved-slab MIP is then calculated by means of this composite curved slab.
We implemented our method with an offline workstation (Windows [Microsoft, Redmond, Wash] with two 500-MHz Pentium III processors [Intel, Santa Clara, Calif]) and 1 GB of random access memory. Because our method was not specifically optimized for a multiprocessor system, only one processor was used.
Validation
To evaluate our algorithm, we selected four consecutive patients (three men and one woman; mean age, 64 years; age range, 56–74 years) with abdominal aortic aneurysms who underwent CT angiography as part of their clinical evaluation. We were granted institutional review board approval to use the images from these studies. Informed consent was not required because patient images were anonymous. This allowed us to evaluate our technique in the presence of heterogeneous soft-tissue and contrast intensities (8), as well as the non-Gaussian correlated noise that is seen in patient data sets. Standard rectangular-slab MIPs were produced for these patients by radiology technologists who were experts in three-dimensional imaging by using a clinical workstation (GE Medical Systems, Milwaukee, Wis) as part of their standard clinical work-up. Rectangular-slab MIPs produced for these patients included nine arteries (aorta, celiac trunk, and hepatic, splenic, superior mesenteric, bilateral renal and bilateral common iliac arteries) from each patient. Clinical images were produced with a cuboidal ROI that was visually estimated by the radiology technologists to be the minimum needed to enclose the vessel. Subsequently, the CT angiographic data sets were transferred to our offline workstation. Vessel starts and end points were manually input by one of the authors (R.R.), and curved-slab MIPs were produced automatically for the same arteries and orientations used for clinical rectangular-slab MIPs (Fig 2). The time required to produce images by both means was recorded.
|
) of the measured values. Contrast-to-noise ratio (CNR) was calculated as follows:
|
V and B are their respective SDs (noise). Weighting factors (HV and HB) based on the relative size of the samples were used. This adjusted for any differences in the sample size used to calculate noise values. For example, if the number of background samples was twice as large as the number of vessel samples, HB would be 2/3 and HV would be 1/3 (8). For each artery imaged with both curved- and rectangular-slab MIPs, the full width at half maximum of vessel intensity was measured automatically from the intensity profile along a supersampled trilinearly interpolated line that was perpendicular to the median centerline of the vessel. The threshold was set as the midpoint between the mean background intensity and the maximum intensity of the profile. The distance between the two intersection points of the threshold and the intensity profile was recorded as the full width at half maximum (3). This measurement was repeated automatically along the centerline of each vessel, with the distance between measurements set to half the minimum voxel dimension.
In addition, composite curved-slab MIPs were produced for each patient through all arteries that were compared, and the number of rectangular-slab MIPs and composite curved-slab MIPs required to depict all arteries was recorded.
Statistical Analysis
We tested seven hypotheses to determine the statistical significance of differences between the measurements. Hypothesis 1: There is no difference in central vessel intensities between curved- and rectangular-slab MIPs. Hypothesis 2: The curved slabs used for curved-slab MIPs have smaller volumes compared with those of the cuboidal ROIs used for rectangular-slab MIPs. Hypothesis 3: The background intensity in curved-slab MIPs is lower than that in rectangular-slab MIPs. Hypothesis 4: Arterial contrast in curved-slab MIPs is higher than that in rectangular-slab MIPs. Hypothesis 5: Background noise in curved-slab MIPs is higher than that in rectangular-slab MIPs. Hypothesis 6: Contrast-to-noise ratio in curved-slab MIPs is higher than that in rectangular-slab MIPs. Hypothesis 7: Apparent vessel width in rectangular-slab MIPs is reduced compared with that in curved-slab MIPs.
A two-tailed t test was used to test hypothesis 1, and one-tailed t tests were used to test the other hypotheses. Differences with a P value of less than .05 were considered to be statistically significant. In addition, differences in contrast and apparent vessel narrowing between curved- and rectangular-slab MIPs were correlated with vessel width as calculated along the vessel lengths by means of linear regression.
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
Image Contrast
We calculated comparison data as follows: To measure the mean central vessel intensity, we sampled the centerline in each curved-slab MIP (one image for each of the nine arteries in each of the four patients) at a rate of approximately two samples per millimeter, which resulted in a mean 272 samples per artery ± 132. Similarly, to measure the mean background vessel intensity, we sampled along each edge of each artery at the same rate, which resulted in a mean 278 samples per artery ± 113. We repeated these measurements at the identical locations in the rectangular-slab MIPs. The intensity values measured at central vessel points in rectangular-slab MIPs were usually identical to those measured at the corresponding points in curved-slab MIPs, with only some values that were 1–2 HU lower or higher. Consequently, mean central vessel intensity was 362.8 HU ± 74.0 (range, 239.4–470.9 HU) in curved-slab MIPs compared with an almost identical mean central vessel intensity of 362.8 HU ± 74.1 (range, 238.5–472.5 HU) in rectangular-slab MIPs. There was no significant difference in the central vessel intensities between curved- and rectangular-slab MIPs (hypothesis 1, P = .98).
On the other hand, mean background intensity was 209.1 HU ± 67.2 in rectangular-slab MIPs compared with a mean background intensity of 146.7 HU ± 44.6 in curved-slab MIPs. This represented a mean decrease of 62.4 HU ± 25.1 in the background intensity in curved-slab MIPs compared with that in the corresponding rectangular-slab MIPs (hypothesis 3, P < .01). The decrease in background intensity ranged from 0 to 680 HU. The instances where there were large decreases in background intensity were a result of exclusion in the curved-slab MIP of bone structures from the edges of vessels. This type of exclusion occurred nine times in one vessel and once each in four other vessels. When these areas were removed from the data, the mean decrease in background intensity was still 55.1 HU ± 15.3. This difference was significant (hypothesis 3, P < .01).
Arterial contrast improved from 153.7 HU ± 57.3 in rectangular-slab MIPs to 216.1 HU ± 25.1 in curved-slab MIPs (hypothesis 4). This difference was significant at P less than .01 when compared by means of a paired t test. Since there was no difference in central vessel intensity, the decrease in background intensity was almost solely responsible for this improvement in contrast. The mean diameter of all studied arteries was 12.6 mm ± 5.3. The R2 value for the correlation between vessel diameter and increase in arterial contrast between curved- and rectangular-slab MIPs was 0.125.
In rectangular-slab MIPs, mean arterial noise was 24.8 HU ± 10.3. Values for arterial noise in curved-slab MIPs were almost identical since the central vessel intensities did not change appreciably. Background noise increased from 84.6 HU ± 60.9 in rectangular-slab MIPs to 101.7 HU ± 30.8 in curved-slab MIPs (hypothesis 5, P < .05). However, there was a 10% increase in contrast-to-noise ratio from 2.01 ± 0.73 in rectangular-slab MIPs to 2.20 ± 0.36 in curved-slab MIPs. This difference was statistically significant (hypothesis 6, P < .05).
Apparent Vessel Width
For the calculation of vessel width, a mean 591 automatic measurements ± 276 of full width at half maximum were performed along the centerline of each artery (approximately four samples per millimeter) for all arteries in all patients. Vessel width in rectangular-slab MIPs was reduced by 0.12–1.09 mm compared with that in curved-slab MIPs, with a mean decrease of 0.53 mm ± 0.34 (hypothesis 7, P < .01) (Fig 2). In comparison, mean in-plane voxel dimensions were 0.55 x 0.55 mm, and the mean through-plane voxel dimension was 1.38 mm. There was no significant difference in values between different vessels, although vessels that coursed mostly in plane (the celiac trunk and renal arteries) had the largest mean apparent vessel narrowing (0.64 mm ± 0.62 and 0.44 mm ± 0.40, respectively). The R2 value for the correlation between vessel diameter and decrease in apparent vessel width in rectangular- versus curved-slab MIPs was 0.112.
Processing Time
Median centerline extraction before creation of a curved-slab MIP required 1.0 minutes ± 0.18 of user interaction and 3.1 minutes ± 0.5 of computer processing time, but this was done only once per patient. Automated specification of the curved slabs required a mean 9 seconds ± 3 of computer processing time per patient (1.5 seconds per vessel). Therefore, curved-slab MIP creation required 4.25 minutes ± 0.73 per patient compared with 2.5 minutes ± 0.4 of manual interaction required by the radiology technologists to produce rectangular-slab MIPs.
Number of Images
To adequately depict the nine arteries in the four patients, the radiology technologists required eight rectangular-slab MIPs each in three patients and seven rectangular-slab MIPs in one patient, which gives a mean 7.75 rectangular-slab MIPs per patient. Most of the arteries were imaged with one rectangular-slab MIP each in the following orientations: one hepatic coronal MIP, one splenic coronal MIP, one superior mesenteric and celiac sagittal MIP, two renal sagittal MIPs, one aortic sagittal MIP, and two iliac sagittal MIPs. Some images included more than one artery, and some arteries required more than one rectangular-slab MIP. The short celiac trunk was always included with the superior mesenteric artery in sagittal MIPs, and both renal arteries were included in three coronal MIPs. Five tortuous arteries (two hepatic arteries, one splenic artery, and two renal arteries) could not be included within a single cuboidal ROI and required two rectangular-slab MIPs each. The splenic artery was not imaged in two patients and the hepatic artery was not imaged in one patient since they were not clinically indicated. The radiology technologists included only the common iliac artery in each iliac sagittal MIP. Aortic coronal MIPs could not be produced because the spine could not be excluded adequately with a cuboidal ROI.
Composite curved-slab MIPs were produced in sagittal and coronal orientations. The iliac arteries were included up to the limits of the scan range. Vessels that overlapped in the viewing direction could not be included in the same sagittal or coronal composite curved-slab MIPs. For example, the superior mesenteric arteries overlapped the aorta in the coronal viewing direction and were included only in the sagittal viewing direction. The coronal image therefore included the aorta, the celiac trunk and its branches, the renal arteries, and the common and external iliac arteries (Fig 3, A). Similarly, the iliac arteries tended to overlap when viewed from the sagittal viewing direction. To provide adequate visualization of the iliac arteries, two composite curved-slab MIPs were always required in the sagittal orientation. Therefore, each sagittal image included the aorta, celiac trunk, superior mesenteric artery, and either the left or right common external and internal iliac arteries (Fig 3, B and C). Therefore, to display all vessels, three composite curved-slab MIPs were required per patient.
|
|
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
The reduction in vessel narrowing seen in our study would be sufficient to be clinically important, especially in the recognition of total stenosis versus residual luminal patency. The actual amount of apparent vessel narrowing can be increased locally in rectangular-slab MIPs (eg, if high-intensity soft tissue overlaps the vessel), which probably causes reductions in apparent vessel narrowing of more than 1 mm (Fig 2). Variable distribution of the excluded high-intensity soft tissue surrounding the vessels may explain the lack of correlation between vessel width, as well as the increases in contrast or apparent vessel narrowing.
All patients in our study had good luminal contrast enhancement, but this is not always the case. Problems such as high-grade stenoses, volume averaging, and intravoxel dephasing effects can decrease luminal contrast. Findings in previous studies (7) show that apparent vessel narrowing is nonlinearly dependent on luminal contrast. The increased contrast in curved-slab MIPs is likely to have additional use in these situations.
In CT angiography, curved-slab MIP has an additional advantage, as it tends to include fewer bone projections near the edge of the vessel, which decreases the chance that adjacent bone will obscure vessel features. This ability to adaptively exclude bone in the view direction allows rendering of 360° curved-slab MIPs of the vessel without manual interaction. Calculation of the median centerline constitutes the major time component in curved-slab MIP creation. The median centerline can also be used to quantify the aorta and its branches (12) and to automatically produce curved-planar reformats (13), which would potentially decrease the total time for postprocessing per patient.
Composite curved-slab MIPs reduce the number of images required to depict vessels compared with that with rectangular-slab MIPs by increasing the number and extent of arteries included in each image. This may reduce the time required to review a study and may facilitate communication with referring clinicians.
Even though findings in our small study show significant advantages to the use of the curved-slab MIP, a larger study is needed to establish the clinical effect of this method. For example, the small increase in noise seen with our technique may not be acceptable if the source data set is excessively noisy. Extensive circumferential calcium in the vessel wall in CT angiograms may preclude assessment of the vessel lumen with MIPs (14,15). Our algorithm does not address these problems. Where curved-slab MIPs are unsuitable, however, our interface allows the user to switch to a curved-planar reformat or to the primary reconstructed images centered on the area to be assessed.
In conclusion, curved-slab MIP increases overall arterial contrast to background compared with that with rectangular-slab MIP as a result of reduced background intensity and exhibits a corresponding reduction in apparent vessel narrowing that may reduce overestimation of stenoses and disappearance of low-intensity vessels. Our method enables multiple vessels to be included in each image without including excessive amounts of soft tissue and high-attenuating bone. This method has the potential to reduce the time required to assess vasculature with MIPs.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: MIP = maximum intensity projection, ROI = region of interest
Author contributions: Guarantor of integrity of entire study, G.D.R.; study concepts and design, S.N., G.D.R., R.R.; literature research, R.R., S.N.; clinical and experimental studies, R.R., S.N., G.D.R.; data acquisition, R.R.; data analysis/interpretation, R.R., S.N., G.D.R.; statistical analysis, R.R., S.N., G.D.R.; manuscript preparation, definition of intellectual content, editing, revision/review, and final version approval, R.R., S.N., G.D.R.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. Raman, B. Raman, S. Napel, and G. D. Rubin Semiautomated Quantification of the Mass and Distribution of Vascular Calcification with Multidetector CT: Method and Evaluation Radiology, April 1, 2008; 247(1): 241 - 250. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. C. Dalrymple, S. R. Prasad, M. W. Freckleton, and K. N. Chintapalli Informatics in Radiology (infoRAD): Introduction to the Language of Three-dimensional Imaging with Multidetector CT RadioGraphics, September 1, 2005; 25(5): 1409 - 1428. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
