| 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, S-072B, Stanford, CA 94305. Received January 8, 2007; revision requested March 1; revision received July 30; accepted August 27; final version accepted September 28. Supported by National Institutes of Health grants 1RO1HL58915 and 1RO1HL67194. Address correspondence to G.D.R. (e-mail: grubin{at}stanford.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
© RSNA, 2008
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
The speed and coverage of multi–detector row computed tomography (CT) allows imaging of most of the arterial system in a single scan (11–13), presenting a unique opportunity to quantify arterial calcification as a measure of atherosclerotic burden. Determining the quantity and distribution of calcium in each blood vessel could allow the investigation of the relationships between the distribution of arterial calcium and a variety of behavioral, physiologic, pathologic, and genotypic conditions.
Among the methods that have been proposed for quantifying coronary calcium, quantification of the mass of calcium is reported to be more accurate and less variable than the classic Agatston score or the volume score (14–18). Reported methods for the quantification of coronary calcium mass have been both threshold specific and scan protocol specific (17,18). For example, one method requires separate calibration scans by using a complex phantom that simulates small calcium fragments. Calibration is performed for each scanner and scan protocol and is used to obtain comparable measurements across different scanners and protocols. With the calibration scans, a correction factor is also calculated for each level of arterial contrast enhancement following intravenous contrast medium administration and is used to obtain comparable measurements that are independent of the level of arterial contrast enhancement (17). In addition, although these methods allow total mass quantification per vessel and per scan, they do not provide a detailed analysis of the size and distribution patterns of the calcium fragments.
Moreover, these methods for coronary arterial calcium quantification require the manual identification of individual calcium fragments. Thus, scaling these methods to the extracoronary arteries presents a challenge of practicality. CT studies currently performed for the quantification of calcification of the coronary arteries include less than 100 sections and take 10–15 minutes to process manually. Systemic arterial CT angiograms currently consist of hundreds to more than 1000 sections and, therefore, require substantially longer to process.
Our study purpose was to prospectively validate in phantoms and to prospectively and retrospectively evaluate in volunteers and patients, respectively, a fast semiautomated method we developed for quantification of the true mass and distribution of calcium in the systemic arteries, without the need for separate calibration scans for each level of arterial contrast enhancement and without the need for calibration with complex phantoms.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
Selection of arteries of interest.—Our algorithm first requires the user to identify the start and end points of vessels to be processed. This serves two purposes. It enables the bones to be excluded by using a previously developed algorithm that identifies and deletes structures that possess a shape and density distribution characteristic of bone (19). It then extracts median centerlines of the vessels of interest by using a previously published method (20) that has been shown in subsequent studies to be reliable (21,22). From the user-defined seed points, this algorithm automatically produces an initial branching path on the surface of the artery, from the start to the end points. The path then undergoes midline extraction and smoothing to produce branched median centerlines through the selected arteries.
Vessel segmentation.—From the centerlines, we define the volume occupied by the selected arteries, excluding other vessels and surrounding soft tissue. To accomplish this task, a list of centerline points is obtained by sampling the median centerlines at subvoxel intervals. The perpendicular vessel cross section through each of these points is then segmented by using the point as a seed for region growing, which is based on an adaptive threshold level computed from the voxel attenuation statistics near the seed point. The average size and shape of the previous five segmentations are used to constrain the subsequent segmentation from extending into adjacent veins or arterial branches. The segmented cross sections are then combined to obtain one composite vessel segmentation. To ensure inclusion of the wall of the vessel in the segmentation, all voxels within 10 mm of the surface of the segmentation that are not classified as bone (the "vessel surround") are added to the segmentation (Fig 1). Because the bony structures have already been excluded by a previous processing step, bone that is within 10 mm of the vessel is not included in the segmentation. Calcium fragments that are within 10 mm of the surface of the segmentation are included in the segmentation.
|
In the process of connected-component analysis, these fragments are resegmented by using the same calcium detection threshold level to include the whole fragment. For this resegmentation step, voxels that are outside the vessel segmentation, higher than the calcium detection threshold level, and in continuity with the partially segmented fragment are also included as part of the fragment. Because only voxels higher than the calcium detection threshold level are selected, no soft-tissue voxels or luminal voxels are segmented as part of the calcium fragment. The user is then prompted to review the detected calcium fragments; delete any erroneously included fragments such as venous phleboliths, osteophytes, or dystrophic extravascular soft-tissue calcifications; and also add any fragments missed by the algorithm.
Mass quantification.—Mass quantification by using our algorithm is designed to be independent of the calcium detection threshold level and is the result of an extrapolation of the brightness-volume product (BVP) of individual calcium fragments at a series of attenuation threshold levels. The BVP of a fragment is defined as the average attenuation of the voxels in the fragment in Hounsfield units multiplied by the volume of the fragment in cubic millimeters. Since the average Hounsfield unit value of the fragment is used to calculate BVP, the various densities of calcium that compose the calcium fragment all contribute to the calculation of the average Hounsfield unit value. Therefore, the heterogeneity of the fragment does not affect the calculation of the BVP.
The following is performed for each fragment. An initial BVP is calculated at the calcium detection threshold value that is higher than the attenuation of surrounding soft tissue and contrast medium. This calculation excludes all soft-tissue voxels from the segmentation of the fragment. Then, the threshold level is increased iteratively by 10 HU, and the BVP is recalculated at each threshold level. This procedure is conducted until the BVP is 10% of the initial BVP. From the series of BVP measurements thus obtained, a best-fit line is calculated and extrapolated to calculate the theoretic BVP that would be obtained at a threshold level of 0 HU (Fig 2). At least five BVP measurements are used to calculate the slope of the best-fit line. If fewer than five measurements are obtained, the fragment is marked as unquantifiable.
|
To calculate the fragment mass from BVP that would be obtained at a threshold level of 0 HU, we used four homogeneous cylindric standards of known calcium density of 0 mg/mL (water attenuation), 50 mg/mL, 100 mg/mL, and 200 mg/mL. These were placed under the patient and included in the scan field. The densities of the standards expressed in milligrams per cubic millimeter were plotted against the attenuation of each standard in Hounsfield units. A line of best fit was computed by using these four measurements. The gradient G of the line of best fit was used to derive the mass of each fragment from the BVP that would be obtained at a threshold level of 0 HU (BVP0). Mass (in milligrams) was calculated as follows: G[mg/(HU · mm3)] · BVP0(HU · mm3).
This method of derivation of mass has two major elements that potentially reduce error and improve precision and simplicity. First, BVP is not calculated at any one threshold level; it is calculated at many threshold levels. We extrapolated these measurements to calculate the BVP at a threshold level of 0 HU (water attenuation). Second, simple in-scan density standards are used to calculate the gradient G. There is no need to obtain separate calibration scans by using complex phantoms that simulate small calcium fragments and there is no need to repeat these calibration scans for a range of arterial contrast enhancement levels. The algorithm is therefore designed to be both threshold level and protocol independent.
Calcium distribution.—For analysis of arterial calcium distribution, radial rays are projected from the median centerline to produce a flattened image map of the vascular surface, with the positions and shapes of the calcium fragments as seen from the centerline superimposed on the image (Fig 3). The first radial ray is projected in an anteroposterior orientation, and the subsequent rays are projected 360° around the center of the vessel, in a clockwise direction. The surface area of the vessel is calculated from the length and angular separation of the radial rays when they intersect with the luminal surface of the vessel.
|
Validation
Physical phantom studies.—Because there is no noninvasive reference standard available for live subjects, we used a physical phantom to verify accuracy. Physical phantoms also allowed us to carefully control calcium mass. Three homogeneous solids with elemental calcium densities of 0.15, 0.30, and 0.45 mg/mm3 were created with calcium oxide and an epoxy resin base. The materials were mixed thoroughly before curing and were constantly shaken during the curing process to ensure homogeneity. Three cylindric standards were machined from each of the homogeneous solids. A fourth standard was made from epoxy resin that contained no calcium (attenuation, 40 HU).
To simulate the irregular fragment geometries encountered in vivo, irregular fragments were randomly chipped from the three solids. Twenty-four fragments representing a range in mass of 0.4–50 mg were selected to obtain a spectrum of mass value. The smallest fragment that could be accurately measured with our scale was 0.4 mg. The mean mass of all fragments was 15.12 mg. The fragments were then placed on the walls of four aortic phantoms. The walls of these cylindric phantoms were composed of low-density polyethylene (attenuation, –50 HU) with a thickness of less than 0.5 mm and were manufactured with a diameter of 35 mm. The phantoms were filled alternately with water and then with a dilute iodine solution (1:30 dilution of iohexol [Omnipaque; GE Healthcare, Milwaukee, Wis], 300 mg/mL, 200–250 HU). The phantoms were sealed at both ends with cork (attenuation, –800 HU) and made watertight by using polymer glue (attenuation, –50 HU).
The four phantoms and four standards were placed in a 25-cm-diameter water bath, with the phantoms positioned near the isocenter of the CT gantry (Fig 4). The standards were placed at the bottom of the phantom, adjacent to the CT table, to mimic the geometry used for patient scanning and, thereby, to generate similar artifact-causing conditions, such as scatter and beam hardening, that occur during patient scanning. The water bath was treated with a small amount of silicone solution to decrease bubble formation.
|
Prospective scans.—We quantified the precision of our algorithm and compared it with the traditional Agatston score by conducting a prospective study in volunteers who reported no history of vascular disease or symptoms attributable to vascular disease. According to a protocol approved by our institutional review board, 21 asymptomatic volunteers (11 men, 10 women; mean age, 60 years; range, 50–80 years) were recruited among employees and families of employees at Stanford University, Stanford, Calif. Identification information of volunteers was removed from all images to achieve compliance with the Health Insurance Portability and Accountability Act (HIPAA). The risks associated with the radiation from the CT scans were discussed with volunteers, and full informed consent was obtained. Because arterial calcium tends to develop 10 years later in women than it does in men, the women were selected to be 10 years older than the men to increase the likelihood that arterial calcium would be present in most of the subjects.
Each volunteer was positioned supine on the CT table over a body-contoured phantom containing three calcium hydroxyapatite reference samples (phantom density of 50, 100, and 200 mg/mL) and one water-equivalent reference (Image Analysis, Columbia, Ky). In the first 10 volunteers, two unenhanced CT scans were obtained, followed by the acquisition of a single contrast material–enhanced CT scan. In the subsequent 11 volunteers, one unenhanced CT scan was obtained, followed by the acquisition of two contrast-enhanced CT scans, to enable quantification of interscan variability on both unenhanced and contrast-enhanced scans. Depending on subject weight, 35–60 mL of nonionic iodinated contrast medium (concentration, 300 milligrams of iodine per milliliter) was delivered through an antecubital intravenous catheter, with the goal of achieving uniform arterial opacification during the entire scan duration.
On the basis of preliminary work performed by one of the authors (G.D.R.), a biphasic injection was delivered with an initial 5-second phase at a flow rate of 2–3 mL/sec and a subsequent 15–20-second phase at a flow rate of 1–2 mL/sec. Contrast-enhanced CT scans were triggered by using a bolus-monitoring algorithm. Once the enhancement in the aortic arch reached 100 HU, helical scanning was initiated with 16 detector rows and 1.25-mm section thickness (16 x 1.25), pitch of 1.3, 0.5-second rotation, 120 kV, and 100 mA, which resulted in a CT dose index of 5.34 mGy and a radiation dose of 5.9 mSv. The scan was initiated at the angle of the mandible and extended inferiorly to the proximal tibial diaphysis. Sections were reconstructed at 0.7-mm intervals, corresponding to approximately 50% of the full width at half maximum of the section sensitivity profile. These scans were used to quantify calcium in the extracoronary arteries.
In addition, a prospectively electrocardiographically triggered breath-hold CT scan was obtained through the heart with 16 x 1.25, which was electrocardiographically gated to 80% of the R-R interval. The scan was obtained with an x-ray tube potential and current of 120 kV and 100 mA, respectively, and a gantry rotation of 0.5 second, resulting in a CT dose index of 5.8 mGy and a radiation dose of 1.05 mSv. The data were reconstructed by using a half-scan reconstruction algorithm to give an effective temporal resolution of 250 msec per section. These scans were used to assess coronary calcium. To assess intrasubject variability of coronary artery calcium measurements, scanning was repeated once. The total dose to each volunteer from all five scans by using our low-dose protocol was approximately 19.8 mSv. This dose was smaller than the estimated dose of 27.5 mSv for a nonelectrocardiographically gated CT angiogram of the same territory by using our standard diagnostic protocol, which includes acquisition of an unenhanced scan followed by that of a contrast-enhanced scan.
On these scans, calcium mass was quantified by using our algorithm, and the Agatston score was calculated by using the published method (23). Calcium was quantified in the coronary arteries, the carotid arteries, the thoracoabdominal aorta, the renal arteries, the common and external iliac arteries, the femoral arteries, and the popliteal arteries. To simplify reporting, results were coalesced into the coronary and extracoronary arterial beds.
Retrospective patient assessment: feasibility in patients with disease.—This study was conducted with an approved protocol of our institutional review board; informed consent was waived, as patient images were collected retrospectively and identification information was removed to achieve compliance with HIPAA. To demonstrate the feasibility of our algorithm in patient data sets, we quantified the amount and distribution of abdominal aortic and iliac calcium on CT angiograms in 15 patients (nine men, six women; mean age, 52 years). Ten consecutive patients (mean age, 61 years) underwent CT angiography for preoperative evaluation of aortic aneurysms, and five consecutive patients (mean age, 30 years) underwent CT angiography to rule out traumatic aortic injury or aortic dissection and did not have known clinical manifestations of systemic atherosclerosis.
For all scan protocols, section thickness was 1.25 mm and section spacing was 0.8 mm, with 120 kV and 390–440 mA. In all cases, 100–150 mL of iodinated contrast medium (iohexol), 350 mg of iodine per milliliter, was injected at a rate of 4–5 mL/sec. All CT angiograms encompassed the entirety of the chest, abdomen, and pelvis. The mean aortic attenuation within the CT scans was 220 HU (range, 150–270 HU).
Because the scans were retrospectively chosen, calcium standards were not included in the scan field. As a result, the mass conversion factor was estimated from the density standards on other scans that were obtained with a similar scan protocol. These scans were obtained by searching our picture archiving and communication system for scans from other patients that had density standards included as part of other related research projects. These scans were obtained with the same scanner and the same scanning protocol and were used only to calculate calibration factors. Calibration was not performed on a section-by-section basis because the density standards were shorter than the scan range. Identification information was removed from all scans to meet HIPAA requirements.
Calcium mass statistics in patient studies were calculated per patient and per anatomic vascular segment by a single operator (R.R.), who is a radiology resident with 6 years of experience in cardiovascular CT research. The thoracic aorta was divided into three segments: ascending, arch, and descending. The abdominal aorta was divided into four segments, demarcated by the origins of the celiac artery, the most inferior renal artery, the inferior mesenteric artery, and the aortic bifurcation. In the pelvis, the left and right common and external iliac arteries were analyzed independently. The calcium content of each segment was determined, with and without normalization, by the total surface area of the segment.
The user interaction time, computer processing time, and review time required to analyze these images was recorded. To assess measurement precision relative to user-selected inputs, for each patient study, the effect of variation in user-selected points was simulated by selecting points within the volume of a sphere with a radius equal to the vessel radius, centered at each initially selected user point. The points were chosen along the radial direction, with each set of radial points 60° from the next set and spaced 1 mm apart in the radial direction. In so doing, we generated between 60 and 220 points, depending on the size of the vessel, and simulated variation in user input; the calcium quantification algorithm was repeated for each of these points. The number and mass of fragments detected by the algorithm for each pair of start and end points was recorded, and results were compared to quantify the effect of variation in the initial user inputs.
Statistical Analysis
In phantom studies, the relationship between the measured and actual mass of each fragment was quantified by using linear regression. Error was quantified in milligrams per fragment and as a percentage of fragment mass. The variability of mass measurement between unenhanced and contrast-enhanced scans was quantified in milligrams per fragment and as a percentage of fragment mass. Because fragment masses were not normally distributed, Wilcoxon signed rank tests were used to test the hypothesis that there was no difference between error on unenhanced scans and error on contrast-enhanced scans. A Wilcoxon signed rank test was also used to test the hypothesis that there was no difference in measured mass on unenhanced versus contrast-enhanced scans.
A two-tailed unpaired t test was used to test the hypothesis that there was no difference in percentage variability between small (mass, <4 mg) and large (mass, >4 mg) fragments. A large fragment was defined as a fragment with a mass of more than 4 mg because approximately one-half of the fragments (13) had actual masses of more than 4 mg, and the other half (11) had actual masses of less than 4 mg. Measurement bias was quantified by determining the signed error and signed percentage error for fragments on both the unenhanced and contrast-enhanced scans.
In prospective volunteer studies, interscan variability between paired unenhanced scans, between paired contrast-enhanced scans, and between paired unenhanced and contrast-enhanced scans for calcium mass was quantified in milligrams and as a percentage of mean calcium mass. For Agatston scores, this interscan variability was quantified in units and as a percentage of mean Agatston score. A two-tailed unpaired t test was used to test the hypothesis that there was no difference in percentage interscan variability for Agatston scores and for calcium mass.
In retrospective patient studies, the mean calcium content of each vascular segment was compared with the calcium content of the other segments by using a randomized block analysis, with each patient as a block. The null hypothesis was that the mean calcium content of each vascular segment was equal. Post hoc analyses were performed by using the Bonferroni method to compare the mean calcium content of each vascular segment with the calcium content of the other segments to test the hypothesis that the calcium content was significantly greater in one segment. Both these analyses were repeated after calcium mass was normalized to the surface area of each segment. For analysis of the effect of variation in user-selected points, error was quantified in milligrams per fragment and per scan. The statistical package (Analyze-It, version 1.73; Analyze-It Software, Leeds, England) for spreadsheet software (Excel; Microsoft) and statistical software (SPSS; SPSS, Chicago, Ill) was used for all statistical tests.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
All fragments were detected and quantified on the unenhanced scans. On the contrast-enhanced scans, six small fragments (mean mass, 1.08 mg ± 0.47; maximum mass, 1.50 mg) were not identified because their maximum Hounsfield unit value was lower than the mean attenuation of the opacified lumen of the phantom (attenuation, 275 HU). For statistical analysis, fragments that could not be detected were recorded as having a measurement error equal to the mass of the fragment. When we compared mass measurements between unenhanced and contrast-enhanced scans, the mean absolute variability for the 18 fragments detected on both unenhanced and contrast-enhanced scans was 1.8 mg ± 2.0 (mean percentage variability, 13.0% ± 7.2). The difference between mass measured by using unenhanced scans and that measured by using contrast-enhanced scans was not significant (P = .35). Mean absolute variability in mass measured between unenhanced and contrast-enhanced scans for fragments more than 4 mg was 2.3 mg ± 2.2 (mean percentage variability, 11.9% ± 7.1), and mean absolute variability for fragments less than 4 mg was 0.5 mg ± 0.3 (mean percentage variability, 16.1% ± 7.3). The difference in percentage measurement variability between large and small fragments was not significant (P = .28).
Mean signed error for measured versus actual mass for both unenhanced and contrast-enhanced scans was 0.35 mg ± 2.0 (mean percentage signed error, 0.61% ± 10.9). The mean signed error for measured versus actual mass for the unenhanced scans and the contrast-enhanced scans was 0.56 mg ± 1.6 (mean percentage signed error, 2.6% ± 10.2) and 0.08 mg ± 2.4 (mean percentage signed error, –2.1% ± 11.5), respectively. Fragments more than 4 mg in mass had a mean signed error of 0.64 mg ± 2.5 (mean percentage signed error, 2.0% ± 9.7), whereas fragments less than 4 mg in mass had a mean signed error of –0.10 mg ± 0.32 (mean percentage signed error, 1.6% ± 12.7). All confidence intervals for the signed errors included zero. Thus, a bias toward over- or underestimation of the mass was not found. The difference in percentage signed error between large and small fragments was not significant (P = .33).
Prospective Patient Scans
Five patients had no detectable extracoronary calcium on all scans, and three patients had no detectable coronary calcium on all scans. These patients were excluded from the analysis for interscan variability. The 16 patients with detectable extracoronary calcium had a mean calcium mass of 472.9 mg ± 688 and a mean Agatston Score of 1633 units ± 2276. Unenhanced scans had a mean absolute interscan variability of 9.6 mg ± 8.6 (mean percentage interscan variability, 5.6% ± 5.9). The Agatston score had a significantly higher mean variability of 129 units ± 184 (mean percentage variability, 11.6% ± 7.9; P < .01). Contrast-enhanced scans had a mean variability of 12.3 mg ± 9.3 (mean percentage variability, 5.7% ± 6.2), whereas the Agatston score had a significantly higher variability of 167 units ± 156 (mean percentage variability, 26.6% ± 34.8; P < .01). Unenhanced versus contrast-enhanced scans had a mean variability of 24.9 mg ± 33.4 (mean percentage variability, 6.3% ± 3.5), whereas the Agatston score had a significantly higher mean variability of 991 units ± 1820 (mean percentage variability, 45.2% ± 21.1; P < .01). The 18 patients with detectable coronary calcium had a mean calcium mass of 33.6 mg ± 87.5 and a mean Agatston score of 201 units ± 492. The mean variability in coronary calcium was 5.5 mg ± 14.2 (mean percentage variability, 10.9% ± 9.3), whereas the Agatston score had a significantly higher mean variability of 27.7 units ± 56.5 (mean percentage variability, 23.7% ± 32.3; P < .01).
Retrospective Patient Scans
Calcium mass.—On the five scans in patients without symptomatic atherosclerosis, a mean of two fragments (mean mass, 0.7 mg; range, 0–1.8 mg) of calcium were detected. No calcium was detected in two patients. Of a total of 10 fragments detected in the three remaining patients, seven fragments (mean mass, 2.46 mg) were located in the infrarenal aorta, two fragments (mean mass, 0.71 mg) were located at the common iliac artery bifurcations, and one fragment (mass, 0.33 mg) was located at the inferior aspect of the arch of the aorta. In patients with aortic aneurysms, the mean number of detected fragments was 21 (range, 6–254), the mean fragment size was 23.1 mm3 (mean mass, 15.3 mg), and the mean total calcium mass per patient was 321.3 mg (range, 45–1443 mg).
Calcium surface and mass distribution.—In patients with aortic aneurysms, 1.45% (range, 0.1%–3.3%) of the total surface area of the assessed arterial walls was calcified (Table). The abdominal aorta as a whole and the infrarenal aorta had significantly greater calcium than all other segments (P < .05). The subsegment of the infrarenal aorta between the inferior mesenteric artery and the aortic bifurcation also possessed significantly more calcium than all other segments (P < .01). There was no significant difference in the calcium content among all other segments and no significant difference between the calcium content of the right and left common and external iliac arteries. When the calcium content of each segment was normalized to the surface area of the segment, the results did not change.
|
Dependence on user input.—The mean aortic radius at the user-defined start points was 12.1 mm, and the mean radius of the external iliac arteries at the user-defined end points was 3.4 mm. The measured mass of all detected fragments was identical when user points were varied and the calcium quantification algorithm was repeated. However, because the user-selected start and end points were varied within a spheric area, two fragments were missed in a proportion of the repeated computer runs of the algorithm. Specifically, one fragment (6.14 mg) was 4 mm proximal to the initial user-selected start point in the abdominal aorta and was missed when the start point was distal to the fragment. This occurred in 32.8% of the repeated computer runs. The other fragment (3.82 mg) was 5 mm distal to the user-selected end point in the external iliac artery and was missed when the end point was proximal to the fragment. This occurred in 71.4% of the repeated computer runs.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
Our algorithm has limitations. When a high level of arterial contrast enhancement is present, small fragments of calcium may have a maximum Hounsfield unit value that is less than the level of arterial contrast enhancement and may therefore be excluded from the analysis. Thus, we recommend maintaining luminal opacification at reasonable levels. If the amount of noise in the scan is high, the calcium detection threshold level will be much higher than the level of arterial contrast enhancement. As a result, small low-attenuation calcium fragments may be missed. Calcium fragments that are completely outside the segmentation will be missed by the automated segmentation. This is most likely to occur in aortic aneurysms with substantial mural thrombus. However, these fragments should be included once the segmentation is reviewed by an experienced observer prior to analysis. If the calcium fragment is at least partially enclosed by the segmentation, it will be detected in the initial stages of our algorithm. The subsequent calculation of BVP will include the whole fragment, including the part that is outside the segmentation. Calcium fragments that are proximal to the initial start point entered by the user or distal to the end points will be missed by our algorithm. Therefore, care must be taken by the operator to assure that the start and end points encompass all mural calcium. Our algorithm may erroneously include some calcium that is located near the ostia of branch vessels that originate in the vessels of interest. Finally, as with all other algorithms, our method is dependent on the amount of error in the fabrication of the mass standards used to convert BVP to calcium mass.
Our method for automated quantification of calcium in the systemic arteries produced accurate and reproducible measurements on scans in phantoms. In a prospective study in a limited set of 21 asymptomatic volunteers, we demonstrated that this method exhibits a low variability on both unenhanced and contrast-enhanced scans. We demonstrated its feasibility for rapid quantification of calcium mass and distribution in patients with and without known vascular disease. These results are preliminary, but our algorithm showed promise for use as a tool to facilitate studies of the relationship between the quantity and distribution of systemic arterial calcification in a variety of patient subtypes.
|
ADVANCE IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
|
IMPLICATION FOR PATIENT CARE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: BVP = brightness-volume product • HIPAA = Health Insurance Portability and Accountability Act
Author contributions: Guarantor of integrity of entire study, G.D.R.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, all authors; clinical and experimental studies, all authors; statistical analysis, all authors; and manuscript editing, all authors
Authors stated no financial relationship to disclose.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCE IN KNOWLEDGEIMPLICATION FOR PATIENT CAREReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
