Carotid Disease: Automated Analysis with Cardiac-gated Three-dimensional US桾echnique and Preliminary Results1
1 From the Departments of Radiology (S.N., B.A.R., R.B.J.), Applied Physics (H.X.), and Medicine (D.S.P.), Stanford University School of Medicine, Lucas Magnetic Resonance Imaging and Spectroscopy [MRS] Center P-287, Stanford, CA 94305-5488; Acuson, Mountain View, Calif (T.S.S.); and Department of Biomedical Engineering, University of Virginia, Charlottesvile (J.A.H.). From the 2000 RSNA scientific assembly. Received January 29, 2001; revision requested March 7; revision received June 29; accepted July 16. Supported in part by a grant from Acuson. B.A.R. supported by the 1998 Cesare Gianturco/RSNA Fellow Support by the Cook Group. A
ABSTRACT
Automatic analysis was performed of four-dimensional ultrasonographic (US) data in the carotid artery. The data, which were acquired in 31 subjects (eight healthy volunteers and 23 patients) by using a US scanner fitted with a special probe, were successfully processed. Acquisition time averaged 12 minutes. Data for all healthy volunteers (n = 8) and patients with complete occlusions (n = 3) were correctly classified. Data for two of the 12 patients with mild to severe (but not occlusive) disease were misclassified by one category.
Index terms: Carotid arteries, flow dynamics, 172.12984 • Carotid arteries, stenosis or obstruction, 172.721 • Carotid arteries, US, 172.12983, 172.12984
INTRODUCTION
More than 7 years ago, members of the Asymptomatic Carotid Artery Stenosis trial concluded that patients with diameter reduction of more than 60% would benefit from carotid endarterectomy (1). Although ultrasonography (US) would appear to be an appropriate screening tool on the basis of its noninvasiveness, variability among machines and among operators (2–6) and cost concerns (7–9) leave carotid screening of asymptomatic individuals controversial. To address these issues, we developed and evaluated a US technique that rapidly depicts the time-varying three-dimensional (3D) morphology of the extracranial carotid artery and automatically quantifies cross-sectional area (CSA) versus position. Our goal was to be able to separate patients into two groups: those who are healthy and those who require more definitive tests and, possibly, surgery. Our purpose in this article is to report the methods and results of a preliminary evaluation of this technique.
Materials and Methods
The subjects in this study were separated into two groups: (a) eight healthy volunteers (five men, three women; age range, 28–47 years; mean age, 36 years) and (b) 23 patients (13 men, 10 women; age range, 41–80 years; mean age, 66 years). The patients had recently undergone either conventional angiography (n = 6), magnetic resonance (MR) angiography (n = 7), or duplex US (n = 10) to assess carotid atherosclerosis. All subjects signed an informed consent form for this study as required by our institutional review board.
Time-varying 3D US (B-mode and color Doppler energy) data were obtained in all subjects (Sequoia 512; Acuson, Mountain View, Calif). The phased-array transducer used for acquiring the 3D data set was modified so two-dimensional image frame position and orientation in 3D space could be determined relative to the first recorded frame (10). Figure 1 illustrates the modified transducer, which comprises a central image array and two tracking arrays arranged perpendicular to it. In addition, electrocardiography was performed simultaneously with image acquisition, and each image was automatically annotated with the associated cardiac phase. After locating the carotid bifurcation, a US technologist oriented the probe with the image array perpendicular to the vessel and, starting from above the bifurcation, translated the probe inferiorly in the elevation direction (ie, perpendicular to the image array, parallel to the tracking arrays) for 45–90 seconds to scan the bifurcation and as much of the common carotid artery as possible. Total acquisition time—which was defined as the time to find, plan, and scan both carotid bifurcations—was recorded for the volunteers.
fig.ommitted
Figure 1. Schematic depicts the modified transducer for tracking probe displacement from frame to frame. The conventional imaging array (ia) forms the imaging plane; perpendicular tracking arrays (ta1 and ta2) form the tracking planes.
After the examination, we transferred the images acquired with all three arrays, as well as the associated electrocardiographic information, to an off-line workstation for postprocessing to correct for nonparallel and nonuniform spacing of the acquired image planes. Our computer program analyzed speckle data between successive images in all three planes to locate the successive two-dimensional image planes accurately in 3D space relative to the first image. Displacement between successively acquired images was estimated by finding motion vectors for which image features were best matched between successive frames. This determination was initially made with a local search and resulted in motion vectors with 1-pixel resolution. Next, displacement was refined to subpixel resolution by fitting high-order curves to the degree of match (implemented as the minimum sum of absolute differences) for neighboring pixel locations. Once the locations of the set of acquired two-dimensional image frames were known in 3D space, the image data were interpolated onto a regular 3D voxel grid.
Next, we used the electrocardiographic phase to parcel the images into 10 separate 3D volumes, each with nearly constant phase, that covered the cardiac cycle. In the most systolic-phase volume, our computer program segmented the carotid artery from its surroundings and automatically determined the median centerline (11) of the common and internal carotid arteries (Fig 2). The program then extracted a sequence of images perpendicular to the centerline every 1 mm along the path and plotted the CSAs of the color Doppler energy data that intersected the centerline versus the distance along the path (12). A graphic user interface allowed rapid cross referencing of points on these plots to the 3D segmented carotid artery, median centerline, and associated perpendicular cross section. Finally, one author (S.N.), who was not blinded to the results of the comparative examination, determined the maximum percentage stenosis from these plots. The maximum percentage stenosis was defined as 100 times the difference between the normal CSA (either distal or proximal to the stenosis) and the minimal CSA, normalized by the normal CSA. The investigator then classified these results into the following four grades: 1, CSA stenosis of 30% or less; 2, CSA stenosis of 31%–60%; 3, CSA stenosis of 61%–99%; 4, occluded. For truth, we obtained the percentage stenosis from the dictated reports of the comparative studies for the patient group.
fig.ommitted
Figure 2. Three-dimensional reconstruction of US data acquired in a volunteer depicts the segmented carotid bifurcation (a), automatically determined centerline path (b), and a plane perpendicular to the path (c).
Results
In the volunteers, total acquisition times for both vessels averaged 12 minutes. All vessels were assigned grade 1 (stenosis "30%).
Our technique failed to produce analyzable results in six (26%) of the 23 patients. Reasons for failure were inadequate coverage of the bifurcation (three [13%] of 23), poor carotid segmentation due to contact with other vascular structures (two [9%] of 23), and failure to discern internal from external carotid branches (one [4%] of 23).
Results for 17 of the 23 patients with data that were successfully processed are summarized in the Table. Both patients with stenosis of 30% or less determined in the comparative examination were correctly identified with our 3D systolic-phase US technique, as were the three patients with complete occlusions. In addition, stenosis in eight of the nine patients with grade 3 vessels was correctly identified with US; stenosis in the remaining patient was underestimated as grade 2. Stenosis in two of the three patients with grade 3 vessels was correctly classified with 3D US; stenosis in the remaining patient was overestimated as grade 3. Figure 3 shows one representative case in a patient with moderate-to-severe stenosis, which was classified as grade 3 with both MR angiography and our technique.
fig.ommitted
Correspondence between Grade of Stenosis with 3D US and That with Comparative Examinations
fig.ommitted
Figure 3. A-C, Images depict results in a patient with a moderate to severe (grade 3) stenosis (arrow in A and B). A, Maximum intensity projection image obtained from a gadolinium-enhanced MR angiogram (fast 3D gradient-echo sequence: repetition time, 4.6 msec; echo time, 1.1 msec; matrix, 484 x 192; one signal acquired; field of view, 30 cm; 20 coronal 2.6-mm-thick sections acquired). B, Three-dimensional reconstruction from a 3D US scan obtained in the most systolic phase (40% of the R-R interval after the R wave). C, Alternative 3D reconstruction of the image in B depicts the median path, the plane perpendicular to the median path at the most stenotic portion of the internal carotid artery, and the plane of the CSA of the internal carotid artery. D, Plot of CSA versus distance from the most inferior aspect of the common carotid through the most superior aspect of the internal carotid artery shows CSA stenosis of approximately 70% at the arrow.
The time required to perform the 3D reconstruction, segmentation, and computation of CSA versus distance was between 30 and 60 minutes per case with use of a workstation with a 300-MHz R12000 processor (O2; Silicon Graphics, Mountain View, Calif).
Discussion
Our preliminary results suggest that it is possible to categorize the percentage stenosis with 3D systolic-phase US and to achieve reasonable concordance with other techniques. Although we over- or underestimated the percentage stenosis by one grade in two of 17 cases, the percentage stenosis in the two patients with stenosis of 30% or less and in all volunteers (who were presumed to be healthy) was correctly graded. Also, no patients with stenoses greater than 30% were categorized as healthy.
Of the three techniques available to assess extracranial carotid disease, US is the least expensive and least invasive. At our institution, however, a bilateral duplex US examination of the carotid when performed by a highly trained US technologist requires approximately 45 minutes, which sets the lower limit on its cost. Should it become advisable to screen asymptomatic individuals in particular risk categories or, possibly, to screen the entire population over a certain age, this cost may not be acceptable. In comparison, the technologist time required with our technique averaged 12 minutes, or roughly one-fourth the time required for a bilateral duplex US examination. Use of our preliminary implementation required up to 1 additional hour of computer processing per case. We do not consider this a limitation for the following reasons: Processing does not require a technologist and, therefore, does not limit the number of patients that can be examined per day; in any event, operator input during this time is negligible. Further, the potential optimization of our implementation with the expected improvements in computer performance over the next few years will likely result in systems that provide these results in nearly real time.
In addition to cost, results at bilateral duplex US performed by different operators and at different times are variable (2–6). Nonetheless, bilateral duplex US is frequently used in the preliminary assessment of carotid disease, and the results are sometimes used to indicate whether an intervention or more definitive testing is needed. Our technique, which would be used only to separate healthy subjects from those requiring further tests, relies on one sweep of the carotid artery and does not require detailed manipulation of the dials and knobs on the scanner to sample the blood velocity at its peak. Thus, this technique is simpler to perform and results are more likely to be reproducible. We have not attempted to verify this in patients, however, and such a study is needed in the future.
We see the potential usefulness of 3D systolic-phase US with automated stenosis determination as a rapid and inexpensive technique to separate healthy subjects from those requiring a more definitive examination and, possibly, surgery. While we have used color Doppler energy data to quantify CSA and, therefore, percentage stenosis, we are aware of the potential limitations of related types of data on the accuracy of such measurements (13). First, although we applied this technique for automatic computation of percentage stenosis to color Doppler energy data, the method could be applied to other potentially more accurate types of volumetric US (eg, B mode) (14,15). Second, this technique is not intended to provide a definitive determination of percentage stenosis. Therefore, the accuracy with which subjects can be separated into four categories is not as important as the ability to separate healthy subjects from those with carotid disease. Our results, though preliminary, suggest that our technique can perform accurate separation of healthy subjects (eight volunteers and two patients with stenosis of 30% or less) from those with mild to severe disease (15 patients with stenosis of more than 30%).
In six of the 23 patients, the percentage stenosis could not be determined with our technique. Three of the six cases involved patients with high carotid bifurcations; this finding indicates a similar limitation for our technique and for noninvasive US in general. The inability in the remaining three cases was due to current limitations of the software in the segmentation and identification of carotid branches. We are confident that these limitations will be overcome with continued development. In addition, we suspect that acoustic shadowing from calcifications may also make segmentation of the color Doppler energy signal difficult in some patients. In a potential screening scenario, however, patients in whom the percentage stenosis could not be determined would be followed up with a more definitive examination. Although not yet proven, we expect that a large percentage of these patients will have carotid disease and, therefore, should be followed up.
Our technique requires further development in the following areas. First, contact of the common and internal carotid with other vascular structures can result in inadequate segmentation and, therefore, erroneous measurements of CSA. In addition to the two cases for which quantitative result could not be obtained, it also occurred in two of the 17 patient cases for which results are presented. In these two cases, we had to ignore the CSA in a small portion of the common carotid. We are currently working on algorithms to aid segmentation under these circumstances.
Second, CSA measurements increase within the bifurcation region because the plane perpendicular to the centerline path intersects parts of both branches (Fig 3, D, see the peak before the stenotic region). Thus, it is currently hard to detect stenoses located centrally within the bifurcation. We are developing approaches for sculpting out the external branch in this region to increase the sensitivity of our technique in the bifurcation region.
Third, it is often difficult to determine which of the two distal branches is the internal carotid artery. Currently, we determine this by manually inspecting the acquired images for clues, such as branches beyond the bifurcation that occur in only the external branch. For our technique to be widely applicable, this step should be automated. However, disease detected in the external branch is still disease; thus, one solution is to evaluate both branches and refer patients for more definitive examinations whenever disease is detected in either branch.
Fourth, we currently detect complete occlusions manually by means of visual inspection of the 3D volumes. This step must also be automated. Fifth, plots of CSA versus distance were manually evaluated for percentage stenosis by observing a minimum and normalizing on the basis of areas measured proximally or distally to the minimum. This step also requires automation. Sixth, although we collect a time series of volumes, we analyze only the systolic-phase volumes. Perhaps greater accuracy and more information could be derived from a dynamic analysis of these data.
Finally, our study design was intended to help development of this method by supplying some preliminary feedback; therefore, the design has limitations. First, an unblinded observer analyzed the graphs. Thus, it is possible that results at 3D US were influenced by knowledge of the results of the comparative study. We will remove this limitation after further development of the technique and before additional trials are conducted. Second, 11 of 17 of the comparative examinations were either MR angiography or duplex US; neither of these modalities is considered a definitive standard of reference to determine the percentage stenosis. Third, we chose to conduct this investigation as a set of two studies: one study to assess the time required to scan both carotid bifurcations with our technique, and one study to assess the accuracy of its results in single vessels. While the eight volunteers who underwent scanning to help us assess the time involved were presumed to be healthy, no reference standard proof is available to corroborate this presumption. Therefore, we did not include their percentage stenosis data in the Table. Also, we chose not to obtain timing information during examination of the patients; hence, it is possible that examination time could differ between the patients and volunteers. In addition, the number of patients included in our study was too small to allow pertinent statistical assessment. The frequency of disease in our population was much higher than what would be expected in a screening population.
In summary, acquisition and processing of 3D systolic-phase US data in the extracranial carotid artery to compute CSA stenoses is feasible. Stenosis categories can be computed automatically on the basis of orthonormal CSAs. Examination time in volunteers was approximately one-fourth the time required for a conventional bilateral duplex US examination. In this preliminary study, all healthy subjects and all patients with complete occlusions were correctly identified. All patients with moderate to severe disease were also identified, although stenosis in two was misclassified by one category. Results in 17 of the 23 patients show good correlation with comparative diagnostic studies. This simple and rapid technique shows promise for identifying patients with carotid disease, although further study in asymptomatic populations is required.
ACKNOWLEDGMENTS
We are grateful to Diane Orluck, US technologist, who supervised US scanning and personally obtained the large majority of the data used in this study.
REFERENCES
Clinical advisory: carotid endarterectomy for patients with asymptomatic internal carotid artery stenosis. Stroke 1994; 25:2523-2524.
Howard G, Baker WH, Chambless LE, Howard VJ, Jones AM, Toole JF. An approach for the use of Doppler ultrasound as a screening tool for hemodynamically significant stenosis (despite heterogeneity of Doppler performance): a multicenter experience. Asymptomatic Carotid Atherosclerosis Study Investigators. Stroke 1996; 27:1951-1957.
Lee VS, Hertzberg BS, Kliewer MA, Carroll BA. Assessment of stenosis: implications of variability of Doppler measurements in normal-appearing carotid arteries. Radiology 1999; 212:493-498.
Lee VS, Hertzberg BS, Workman MJ, et al. Variability of Doppler US measurements along the common carotid artery: effects on estimates of internal carotid arterial stenosis in patients with angiographically proved disease. Radiology 2000; 214:387-392.
Mead GE, Lewis SC, Wardlaw JM. Variability in Doppler ultrasound influences referral of patients for carotid surgery. Eur J Ultrasound 2000; 12:137-143.
Mathiesen EB, Joakimsen O, Bonaa KH. Intersonographer reproducibility and intermethod variability of ultrasound measurements of carotid artery stenosis: the Tromso study. Cerebrovasc Dis 2000; 10:207-213.
Hill AB. Should patients be screened for asymptomatic carotid artery stenosis?. Can J Surg 1998; 41:208-213.
Lee TT, Solomon NA, Heidenreich PA, Oehlert J, Garber AM. Cost-effectiveness of screening for carotid stenosis in asymptomatic persons. Ann Intern Med 1997; 126:337-346.
Yin D, Carpenter JP. Cost-effectiveness of screening for asymptomatic carotid stenosis. J Vasc Surg 1998; 27:245-255.
Hossack JA, Sliwa JW, Jr, Maslak SH, Gardner EA, Holley GL, Napolitano DJ. U.S. Patent 6,132,376. Multiple ultrasonic image registration system, method and transducer. October 17 2000.
Paik DS, Beaulieu CF, Jeffrey RB, Jr, Rubin GD, Napel S. Automated flight path planning for virtual endoscopy. Med Phys 1998; 25:629-637.
Rubin GD, Paik DS, Johnston PC, Napel S. Measurement of the aorta and its branches with helical CT. Radiology 1998; 206:823-829.
Claudon M, Winninger D, Briancon S, Pesque P. Power Doppler US: evaluation of the morphology of stenoses with a flow phantom. Radiology 2001; 218:109-117.
Mao F, Gill J, Downey D, Fenster A. Segmentation of carotid artery in ultrasound images: method development and evaluation technique. Med Phys 2000; 27:1961-1970.
Gill JD, Ladak HM, Steinman DA, Fenster A. Accuracy and variability assessment of a semiautomatic technique for segmentation of the carotid arteries from three-dimensional ultrasound images. Med Phys 2000; 27:1333-1342., http://www.100md.com(Sandy Napel PhD Haobo Xu MS David S. Paik MS Barbra A. Ross MD Thilaka S. Sumanaweera PhD John A. Ho)
ABSTRACT
Automatic analysis was performed of four-dimensional ultrasonographic (US) data in the carotid artery. The data, which were acquired in 31 subjects (eight healthy volunteers and 23 patients) by using a US scanner fitted with a special probe, were successfully processed. Acquisition time averaged 12 minutes. Data for all healthy volunteers (n = 8) and patients with complete occlusions (n = 3) were correctly classified. Data for two of the 12 patients with mild to severe (but not occlusive) disease were misclassified by one category.
Index terms: Carotid arteries, flow dynamics, 172.12984 • Carotid arteries, stenosis or obstruction, 172.721 • Carotid arteries, US, 172.12983, 172.12984
INTRODUCTION
More than 7 years ago, members of the Asymptomatic Carotid Artery Stenosis trial concluded that patients with diameter reduction of more than 60% would benefit from carotid endarterectomy (1). Although ultrasonography (US) would appear to be an appropriate screening tool on the basis of its noninvasiveness, variability among machines and among operators (2–6) and cost concerns (7–9) leave carotid screening of asymptomatic individuals controversial. To address these issues, we developed and evaluated a US technique that rapidly depicts the time-varying three-dimensional (3D) morphology of the extracranial carotid artery and automatically quantifies cross-sectional area (CSA) versus position. Our goal was to be able to separate patients into two groups: those who are healthy and those who require more definitive tests and, possibly, surgery. Our purpose in this article is to report the methods and results of a preliminary evaluation of this technique.
Materials and Methods
The subjects in this study were separated into two groups: (a) eight healthy volunteers (five men, three women; age range, 28–47 years; mean age, 36 years) and (b) 23 patients (13 men, 10 women; age range, 41–80 years; mean age, 66 years). The patients had recently undergone either conventional angiography (n = 6), magnetic resonance (MR) angiography (n = 7), or duplex US (n = 10) to assess carotid atherosclerosis. All subjects signed an informed consent form for this study as required by our institutional review board.
Time-varying 3D US (B-mode and color Doppler energy) data were obtained in all subjects (Sequoia 512; Acuson, Mountain View, Calif). The phased-array transducer used for acquiring the 3D data set was modified so two-dimensional image frame position and orientation in 3D space could be determined relative to the first recorded frame (10). Figure 1 illustrates the modified transducer, which comprises a central image array and two tracking arrays arranged perpendicular to it. In addition, electrocardiography was performed simultaneously with image acquisition, and each image was automatically annotated with the associated cardiac phase. After locating the carotid bifurcation, a US technologist oriented the probe with the image array perpendicular to the vessel and, starting from above the bifurcation, translated the probe inferiorly in the elevation direction (ie, perpendicular to the image array, parallel to the tracking arrays) for 45–90 seconds to scan the bifurcation and as much of the common carotid artery as possible. Total acquisition time—which was defined as the time to find, plan, and scan both carotid bifurcations—was recorded for the volunteers.
fig.ommitted
Figure 1. Schematic depicts the modified transducer for tracking probe displacement from frame to frame. The conventional imaging array (ia) forms the imaging plane; perpendicular tracking arrays (ta1 and ta2) form the tracking planes.
After the examination, we transferred the images acquired with all three arrays, as well as the associated electrocardiographic information, to an off-line workstation for postprocessing to correct for nonparallel and nonuniform spacing of the acquired image planes. Our computer program analyzed speckle data between successive images in all three planes to locate the successive two-dimensional image planes accurately in 3D space relative to the first image. Displacement between successively acquired images was estimated by finding motion vectors for which image features were best matched between successive frames. This determination was initially made with a local search and resulted in motion vectors with 1-pixel resolution. Next, displacement was refined to subpixel resolution by fitting high-order curves to the degree of match (implemented as the minimum sum of absolute differences) for neighboring pixel locations. Once the locations of the set of acquired two-dimensional image frames were known in 3D space, the image data were interpolated onto a regular 3D voxel grid.
Next, we used the electrocardiographic phase to parcel the images into 10 separate 3D volumes, each with nearly constant phase, that covered the cardiac cycle. In the most systolic-phase volume, our computer program segmented the carotid artery from its surroundings and automatically determined the median centerline (11) of the common and internal carotid arteries (Fig 2). The program then extracted a sequence of images perpendicular to the centerline every 1 mm along the path and plotted the CSAs of the color Doppler energy data that intersected the centerline versus the distance along the path (12). A graphic user interface allowed rapid cross referencing of points on these plots to the 3D segmented carotid artery, median centerline, and associated perpendicular cross section. Finally, one author (S.N.), who was not blinded to the results of the comparative examination, determined the maximum percentage stenosis from these plots. The maximum percentage stenosis was defined as 100 times the difference between the normal CSA (either distal or proximal to the stenosis) and the minimal CSA, normalized by the normal CSA. The investigator then classified these results into the following four grades: 1, CSA stenosis of 30% or less; 2, CSA stenosis of 31%–60%; 3, CSA stenosis of 61%–99%; 4, occluded. For truth, we obtained the percentage stenosis from the dictated reports of the comparative studies for the patient group.
fig.ommitted
Figure 2. Three-dimensional reconstruction of US data acquired in a volunteer depicts the segmented carotid bifurcation (a), automatically determined centerline path (b), and a plane perpendicular to the path (c).
Results
In the volunteers, total acquisition times for both vessels averaged 12 minutes. All vessels were assigned grade 1 (stenosis "30%).
Our technique failed to produce analyzable results in six (26%) of the 23 patients. Reasons for failure were inadequate coverage of the bifurcation (three [13%] of 23), poor carotid segmentation due to contact with other vascular structures (two [9%] of 23), and failure to discern internal from external carotid branches (one [4%] of 23).
Results for 17 of the 23 patients with data that were successfully processed are summarized in the Table. Both patients with stenosis of 30% or less determined in the comparative examination were correctly identified with our 3D systolic-phase US technique, as were the three patients with complete occlusions. In addition, stenosis in eight of the nine patients with grade 3 vessels was correctly identified with US; stenosis in the remaining patient was underestimated as grade 2. Stenosis in two of the three patients with grade 3 vessels was correctly classified with 3D US; stenosis in the remaining patient was overestimated as grade 3. Figure 3 shows one representative case in a patient with moderate-to-severe stenosis, which was classified as grade 3 with both MR angiography and our technique.
fig.ommitted
Correspondence between Grade of Stenosis with 3D US and That with Comparative Examinations
fig.ommitted
Figure 3. A-C, Images depict results in a patient with a moderate to severe (grade 3) stenosis (arrow in A and B). A, Maximum intensity projection image obtained from a gadolinium-enhanced MR angiogram (fast 3D gradient-echo sequence: repetition time, 4.6 msec; echo time, 1.1 msec; matrix, 484 x 192; one signal acquired; field of view, 30 cm; 20 coronal 2.6-mm-thick sections acquired). B, Three-dimensional reconstruction from a 3D US scan obtained in the most systolic phase (40% of the R-R interval after the R wave). C, Alternative 3D reconstruction of the image in B depicts the median path, the plane perpendicular to the median path at the most stenotic portion of the internal carotid artery, and the plane of the CSA of the internal carotid artery. D, Plot of CSA versus distance from the most inferior aspect of the common carotid through the most superior aspect of the internal carotid artery shows CSA stenosis of approximately 70% at the arrow.
The time required to perform the 3D reconstruction, segmentation, and computation of CSA versus distance was between 30 and 60 minutes per case with use of a workstation with a 300-MHz R12000 processor (O2; Silicon Graphics, Mountain View, Calif).
Discussion
Our preliminary results suggest that it is possible to categorize the percentage stenosis with 3D systolic-phase US and to achieve reasonable concordance with other techniques. Although we over- or underestimated the percentage stenosis by one grade in two of 17 cases, the percentage stenosis in the two patients with stenosis of 30% or less and in all volunteers (who were presumed to be healthy) was correctly graded. Also, no patients with stenoses greater than 30% were categorized as healthy.
Of the three techniques available to assess extracranial carotid disease, US is the least expensive and least invasive. At our institution, however, a bilateral duplex US examination of the carotid when performed by a highly trained US technologist requires approximately 45 minutes, which sets the lower limit on its cost. Should it become advisable to screen asymptomatic individuals in particular risk categories or, possibly, to screen the entire population over a certain age, this cost may not be acceptable. In comparison, the technologist time required with our technique averaged 12 minutes, or roughly one-fourth the time required for a bilateral duplex US examination. Use of our preliminary implementation required up to 1 additional hour of computer processing per case. We do not consider this a limitation for the following reasons: Processing does not require a technologist and, therefore, does not limit the number of patients that can be examined per day; in any event, operator input during this time is negligible. Further, the potential optimization of our implementation with the expected improvements in computer performance over the next few years will likely result in systems that provide these results in nearly real time.
In addition to cost, results at bilateral duplex US performed by different operators and at different times are variable (2–6). Nonetheless, bilateral duplex US is frequently used in the preliminary assessment of carotid disease, and the results are sometimes used to indicate whether an intervention or more definitive testing is needed. Our technique, which would be used only to separate healthy subjects from those requiring further tests, relies on one sweep of the carotid artery and does not require detailed manipulation of the dials and knobs on the scanner to sample the blood velocity at its peak. Thus, this technique is simpler to perform and results are more likely to be reproducible. We have not attempted to verify this in patients, however, and such a study is needed in the future.
We see the potential usefulness of 3D systolic-phase US with automated stenosis determination as a rapid and inexpensive technique to separate healthy subjects from those requiring a more definitive examination and, possibly, surgery. While we have used color Doppler energy data to quantify CSA and, therefore, percentage stenosis, we are aware of the potential limitations of related types of data on the accuracy of such measurements (13). First, although we applied this technique for automatic computation of percentage stenosis to color Doppler energy data, the method could be applied to other potentially more accurate types of volumetric US (eg, B mode) (14,15). Second, this technique is not intended to provide a definitive determination of percentage stenosis. Therefore, the accuracy with which subjects can be separated into four categories is not as important as the ability to separate healthy subjects from those with carotid disease. Our results, though preliminary, suggest that our technique can perform accurate separation of healthy subjects (eight volunteers and two patients with stenosis of 30% or less) from those with mild to severe disease (15 patients with stenosis of more than 30%).
In six of the 23 patients, the percentage stenosis could not be determined with our technique. Three of the six cases involved patients with high carotid bifurcations; this finding indicates a similar limitation for our technique and for noninvasive US in general. The inability in the remaining three cases was due to current limitations of the software in the segmentation and identification of carotid branches. We are confident that these limitations will be overcome with continued development. In addition, we suspect that acoustic shadowing from calcifications may also make segmentation of the color Doppler energy signal difficult in some patients. In a potential screening scenario, however, patients in whom the percentage stenosis could not be determined would be followed up with a more definitive examination. Although not yet proven, we expect that a large percentage of these patients will have carotid disease and, therefore, should be followed up.
Our technique requires further development in the following areas. First, contact of the common and internal carotid with other vascular structures can result in inadequate segmentation and, therefore, erroneous measurements of CSA. In addition to the two cases for which quantitative result could not be obtained, it also occurred in two of the 17 patient cases for which results are presented. In these two cases, we had to ignore the CSA in a small portion of the common carotid. We are currently working on algorithms to aid segmentation under these circumstances.
Second, CSA measurements increase within the bifurcation region because the plane perpendicular to the centerline path intersects parts of both branches (Fig 3, D, see the peak before the stenotic region). Thus, it is currently hard to detect stenoses located centrally within the bifurcation. We are developing approaches for sculpting out the external branch in this region to increase the sensitivity of our technique in the bifurcation region.
Third, it is often difficult to determine which of the two distal branches is the internal carotid artery. Currently, we determine this by manually inspecting the acquired images for clues, such as branches beyond the bifurcation that occur in only the external branch. For our technique to be widely applicable, this step should be automated. However, disease detected in the external branch is still disease; thus, one solution is to evaluate both branches and refer patients for more definitive examinations whenever disease is detected in either branch.
Fourth, we currently detect complete occlusions manually by means of visual inspection of the 3D volumes. This step must also be automated. Fifth, plots of CSA versus distance were manually evaluated for percentage stenosis by observing a minimum and normalizing on the basis of areas measured proximally or distally to the minimum. This step also requires automation. Sixth, although we collect a time series of volumes, we analyze only the systolic-phase volumes. Perhaps greater accuracy and more information could be derived from a dynamic analysis of these data.
Finally, our study design was intended to help development of this method by supplying some preliminary feedback; therefore, the design has limitations. First, an unblinded observer analyzed the graphs. Thus, it is possible that results at 3D US were influenced by knowledge of the results of the comparative study. We will remove this limitation after further development of the technique and before additional trials are conducted. Second, 11 of 17 of the comparative examinations were either MR angiography or duplex US; neither of these modalities is considered a definitive standard of reference to determine the percentage stenosis. Third, we chose to conduct this investigation as a set of two studies: one study to assess the time required to scan both carotid bifurcations with our technique, and one study to assess the accuracy of its results in single vessels. While the eight volunteers who underwent scanning to help us assess the time involved were presumed to be healthy, no reference standard proof is available to corroborate this presumption. Therefore, we did not include their percentage stenosis data in the Table. Also, we chose not to obtain timing information during examination of the patients; hence, it is possible that examination time could differ between the patients and volunteers. In addition, the number of patients included in our study was too small to allow pertinent statistical assessment. The frequency of disease in our population was much higher than what would be expected in a screening population.
In summary, acquisition and processing of 3D systolic-phase US data in the extracranial carotid artery to compute CSA stenoses is feasible. Stenosis categories can be computed automatically on the basis of orthonormal CSAs. Examination time in volunteers was approximately one-fourth the time required for a conventional bilateral duplex US examination. In this preliminary study, all healthy subjects and all patients with complete occlusions were correctly identified. All patients with moderate to severe disease were also identified, although stenosis in two was misclassified by one category. Results in 17 of the 23 patients show good correlation with comparative diagnostic studies. This simple and rapid technique shows promise for identifying patients with carotid disease, although further study in asymptomatic populations is required.
ACKNOWLEDGMENTS
We are grateful to Diane Orluck, US technologist, who supervised US scanning and personally obtained the large majority of the data used in this study.
REFERENCES
Clinical advisory: carotid endarterectomy for patients with asymptomatic internal carotid artery stenosis. Stroke 1994; 25:2523-2524.
Howard G, Baker WH, Chambless LE, Howard VJ, Jones AM, Toole JF. An approach for the use of Doppler ultrasound as a screening tool for hemodynamically significant stenosis (despite heterogeneity of Doppler performance): a multicenter experience. Asymptomatic Carotid Atherosclerosis Study Investigators. Stroke 1996; 27:1951-1957.
Lee VS, Hertzberg BS, Kliewer MA, Carroll BA. Assessment of stenosis: implications of variability of Doppler measurements in normal-appearing carotid arteries. Radiology 1999; 212:493-498.
Lee VS, Hertzberg BS, Workman MJ, et al. Variability of Doppler US measurements along the common carotid artery: effects on estimates of internal carotid arterial stenosis in patients with angiographically proved disease. Radiology 2000; 214:387-392.
Mead GE, Lewis SC, Wardlaw JM. Variability in Doppler ultrasound influences referral of patients for carotid surgery. Eur J Ultrasound 2000; 12:137-143.
Mathiesen EB, Joakimsen O, Bonaa KH. Intersonographer reproducibility and intermethod variability of ultrasound measurements of carotid artery stenosis: the Tromso study. Cerebrovasc Dis 2000; 10:207-213.
Hill AB. Should patients be screened for asymptomatic carotid artery stenosis?. Can J Surg 1998; 41:208-213.
Lee TT, Solomon NA, Heidenreich PA, Oehlert J, Garber AM. Cost-effectiveness of screening for carotid stenosis in asymptomatic persons. Ann Intern Med 1997; 126:337-346.
Yin D, Carpenter JP. Cost-effectiveness of screening for asymptomatic carotid stenosis. J Vasc Surg 1998; 27:245-255.
Hossack JA, Sliwa JW, Jr, Maslak SH, Gardner EA, Holley GL, Napolitano DJ. U.S. Patent 6,132,376. Multiple ultrasonic image registration system, method and transducer. October 17 2000.
Paik DS, Beaulieu CF, Jeffrey RB, Jr, Rubin GD, Napel S. Automated flight path planning for virtual endoscopy. Med Phys 1998; 25:629-637.
Rubin GD, Paik DS, Johnston PC, Napel S. Measurement of the aorta and its branches with helical CT. Radiology 1998; 206:823-829.
Claudon M, Winninger D, Briancon S, Pesque P. Power Doppler US: evaluation of the morphology of stenoses with a flow phantom. Radiology 2001; 218:109-117.
Mao F, Gill J, Downey D, Fenster A. Segmentation of carotid artery in ultrasound images: method development and evaluation technique. Med Phys 2000; 27:1961-1970.
Gill JD, Ladak HM, Steinman DA, Fenster A. Accuracy and variability assessment of a semiautomatic technique for segmentation of the carotid arteries from three-dimensional ultrasound images. Med Phys 2000; 27:1333-1342., http://www.100md.com(Sandy Napel PhD Haobo Xu MS David S. Paik MS Barbra A. Ross MD Thilaka S. Sumanaweera PhD John A. Ho)