Quantitative assessment of carotid plaque morphology (geometry and tissue composition) using computed tomography angiography

Quantitative assessment of carotid plaque morphology (geometry and tissue composition) using computed tomography angiography Matthew T. Chrencik, BS,a,b Amir A. Khan, PhD,c Lauren Luther, BS,a,b Laila Anthony, BS,a,b John Yokemick, RVT,a,b Jigar Patel, MD,d John D. Sorkin, MD, PhD,e,f Siddhartha Sikdar, PhD,c and Brajesh K. Lal, MD,a,b Baltimore, Md; and Fairfax, Va

ABSTRACT Objective: Quantification of carotid plaque morphology (geometry and tissue composition) may help stratify risk for future stroke and assess plaque progression or regression in response to medical risk factor modification. We assessed the feasibility and reliability of morphologic measurements of carotid plaques using computed tomography angiography (CTA) and determined the minimum detectable change in plaque features by this approach. Methods: CTA images of both carotid arteries in 50 patients were analyzed by two observers using a semiautomatic image analysis program, yielding 93 observations per user (seven arteries were excluded because of prior stenting). One observer repeated the analyses 4 weeks later. Measurements included total plaque volume; percentage stenosis (by diameter and area); and tissue composition for calcium, lipid-rich necrotic core (LRNC), and intraplaque hemorrhage (IPH). Reliability of measurements was assessed by intraclass and interclass correlation and Bland-Altman plots. Dice similarity coefficient (DSC) and modified Hausdorff distance (MHD) assessed reliability of geometric shape measurements. We additionally computed the minimum amount of change in these features detectable by our approach. Results: The cohort was 51% male (mean age, 70.1 years), and 56% had a prior stroke. The mean (6 standard deviation) plaque volume was 837.3 6 431.3 mm3, stenosis diameter was 44.5% 6 25.6%, and stenosis area was 58.1% 6 29.0%. These measurements showed high reliability. Intraclass correlation coefficients for plaque volume, percentage stenosis by diameter, and percentage stenosis by area were 0.96, 0.87, and 0.83, respectively; interclass correlation coefficients were 0.88, 0.84, and 0.78. Intraclass correlations for tissue composition were 0.99, 0.96, and 0.86 (calcium, LRNC, and IPH, respectively), and interclass correlations were 0.99, 0.92, and 0.92. Shape measurements showed high intraobserver (DSC, 0.95 6 0.04; MHD, 0.16 6 0.10 mm) and interobserver (DSC, 0.94 6 0.05; MHD, 0.19 6 0.12 mm) luminal agreement. This approach can detect a change of at least 3.9% in total plaque volume, 1.2 mm3 in calcium, 4.3 mm3 in LRNC, and 8.6 mm3 in IPH with the same observer repeating measurements and 9.9% in plaque volume, 1.9 mm3 in calcium, 7.9 mm3 in LRNC, and 6.8 mm3 in IPH for two different observers. Conclusions: Carotid plaque geometry (total volume, diameter stenosis, and area stenosis) and tissue composition (calcium, LRNC, and IPH) are measured reliably from clinical CTA images using a semiautomatic image analysis program. The minimum change in plaque volume detectable is w4% if the same observer makes both measurements and w10% for different observers. Small changes in plaque composition can also be detected reliably. This approach can facilitate longitudinal studies for identifying high-risk plaque features and for quantifying plaque progression or regression after treatment. (J Vasc Surg 2019;-:1-11.) Keywords: Carotid; Atherosclerosis; Stenosis; Morphology; CT angiography

The quantification of carotid plaque geometry (plaque volume, luminal area reduction) and composition (calcification, lipid rich necrotic core [LRNC], intraplaque hemorrhage [IPH]) may assist in better identification of patients with a likelihood for future stroke.1-4 Plaques progress along the vessel 2.4 times faster than they

thicken.5 Therefore, methods that capture both longitudinal and circumferential growth (ie, area and volume) are inherently more sensitive than methods limited to thickness measurements (ie, diameter-reducing stenosis). Large LRNC, increased IPH, and reduced calcification may indicate an elevated stroke risk.2,3,6-8 Noninvasive

From the Department of Vascular Surgery, University of Maryland School of

Author conflict of interest: none.

Medicine,a and the Vascular Service, Veterans Affairs Medical Center,b Balti-

Correspondence: Brajesh K. Lal, MD, Division of Vascular Surgery, University of

more; the Department of Bioengineering, George Mason University, Fairfaxc;

Maryland Medical Center, 22 S Greene St, S10B00, Baltimore, MD 21201

and the Imaging Service, VA Maryland Health Care System,d the Baltimore

(e-mail: [email protected]).

VA Medical Center Geriatric Research, Education, and Clinical Center, Balti-

The editors and reviewers of this article have no relevant financial relationships to

more Veterans Affairs Medical Center,e and the Claude D. Pepper Older

disclose per the JVS policy that requires reviewers to decline review of any

Americans Independence Center, University of Maryland School of Medicine,f Baltimore. Funding was provided by VA Merit awards RX000995 and CX001621 and NIH awards NS097876, U01NS080168, and AG000513 to B.K.L. and NIH awards

manuscript for which they may have a conflict of interest. 0741-5214 Copyright Ó 2019 by the Society for Vascular Surgery. Published by Elsevier Inc. https://doi.org/10.1016/j.jvs.2018.11.050

AG028747, DK072488, and VA GRECC to J.D.S.

1

2

Journal of Vascular Surgery

Chrencik et al

---

identification of these features may allow risk stratification, early revascularization, or implementation of aggressive pharmacologic management. Furthermore, quantifying a reduction or increase in total plaque geometric or tissue-specific measures will help determine the efficacy of pharmacologic treatment. We have described a three-dimensional (3D) duplex ultrasound (DUS)-based semiautomatic technique that quantifies plaque morphology (geometry and tissue composition).9 However, DUS can be limited by operator variability, low image resolution, and acoustic shadowing.10 Computed tomography angiography (CTA) or magnetic resonance imaging (MRI) is often used to supplement DUS during clinical evaluation.11 MRI provides high-resolution images, allowing assessment of 3D geometry and plaque tissue composition, but it is sensitive to the patient’s movement and requires longer scan times, specialized equipment, and operator skills, all leading to increased costs.12 CTA provides high spatial resolution, allowing delineation of arterial outlines; imaging protocols are well established in hospitals, shorter scan times ensure reduced motion artifact, and operator variability is limited during image acquisition.13 Whereas CTAderived diameter stenosis measurements correlate well with catheter angiographic measurements, the ability of CTA to measure more complex carotid plaque geometry (area stenosis and plaque volume) and tissue composition (calcium, LRNC, and IPH) has received limited attention.14 Before using CTA-derived measures of plaque morphology in large-scale studies, it is critical to develop and to validate standardized analytic techniques. These should ideally be reliable, repeatable, and able to detect small changes in morphologic appearance. In this study, we analyze the capability of a semiautomatic image processing protocol to measure carotid plaque morphology (geometry and tissue composition) using clinical CTA images. Our three objectives were to assess intraobserver and interobserver repeatability for measures of plaque geometry and tissue composition, to quantify the least amount of change in total plaque volume and in the volume of individual tissue constituents that the protocol can reliably detect, and to determine the time required to complete the analysis. To our knowledge, this is the first study to make these assessments using clinical CTA imaging.

METHODS Patients and study design. A cross-sectional convenience sample of CTA images from 50 patients with varying severities of carotid atherosclerosis was selected from the University of Maryland Vascular Imaging Core Laboratory. Demographics and comorbidities were recorded. Patients underwent contrast-enhanced CTA as part of standard-of-care diagnostic evaluation using a routine clinical imaging protocol. The following inclusion criteria were used for the CTA images: contrast material

2019

ARTICLE HIGHLIGHTS d

d

d

Type of Research: Single-center cross-sectional study Key Findings: A semiautomatic image analysis protocol reliably measured carotid plaque morphology from computed tomography angiography images of 50 patients. The minimum change in plaque volume detectable was approximately 4% if the same observer made both measurements and roughly 10% for different observers. Take Home Message: Quantification of carotid plaque morphology (geometry and tissue composition) may help stratify risk for future stroke and assess plaque progression or regression in response to medical risk factor modification.

must be visualized within the entire carotid arterial tree; axial slice thickness must be #2.0 mm; axial slices must traverse at least 2 cm upstream and downstream of the carotid bifurcation; axial slices must traverse at least 1 cm upstream and downstream of the plaque; and target arteries must not have undergone a revascularization procedure (surgery or stenting). Analysis of all CTA images was performed by two observers, each blinded to the other’s measurements. One observer blindly repeated the analysis after an interval of at least 4 weeks (mean repeat time of 5 weeks) to minimize recall bias. All arteries in the study were measured for plaque geometry (total plaque volume, diameter, and area stenosis). Arteries with $20% stenosis by diameter were included in the plaque tissue composition analysis. The study was approved by the University of Maryland School of Medicine Institutional Review Board. Semiautomatic 3D image analysis. We used an image processing software (vascuCAP, version A.1.1; Elucid Bioimaging, Wenham, Mass)15 to outline (segment) the luminal and outer wall surfaces of the common, internal, and external carotid arteries. Fig 1 illustrates the overall workflow. Step 1: Loading of the CTA images, surveying of images for quality, and assigning of reference points on axial or coronal images. The points serve as the seeds to generate a centerline used for the automatic segmentation of lumen boundaries. The automatic algorithm is driven by thresholding image brightness and texture-based features that define the lumen-tissue interface. Threshold levels can be adjusted manually for cases with low contrast from poor bolus timing. Step 2: The automatically generated lumen boundaries are evaluated and manually refined, particularly in arteries with very tight stenoses, excessive calcium, or low luminal contrast levels. Step 3: The automatically generated outer wall boundaries are evaluated and manually refined as needed.

Journal of Vascular Surgery Volume

-,

Number

Chrencik et al

3

-

Step 4: Plaque tissue composition analysis. Quantification based solely on Hounsfield intensities may introduce errors compared with histologic evaluation.16,17 vascuCAP reduces this error by compensating for the blurring caused by the imaging system point spread function using a Gaussian-regularized multistage level-set approach.15 The resulting plaque tissue composition measures have been histologically validated for calcium, LRNC, and matrix (composed of all stable wall elements including fibrous tissue and all other unlabeled pixels).15 These algorithms mitigate artifacts, such as partial volume effects due to calcium, whereby calcified plaques appear larger than their actual size. Four plaque tissue constituents are labeled at the end of this step: calcium, LRNC, IPH, and matrix. Step 5: Identification of the atherosclerotic plaque. A marker is manually assigned at the origin of the plaque proximal to the stenosis where the lumen is normal in diameter and at the termination of the plaque distal to the stenosis where the lumen returns to a normal diameter. This marking allows subsequent automatic computation of the plaque’s morphologic features (geometry and tissue composition). Fig 2 illustrates the luminal and outer wall boundaries of a stenotic artery, displaying both the 3D reconstruction of the geometry and the contours in a selected axial slice. The left panels (A and C) correspond to automatic segmentation resulting from the software. The right panels (B and D) show the final segmentation after manual refinement of the outlines. Fig 2, A shows segmentation of the lumen in the internal and external carotid arteries in an axial slice at the carotid bifurcation. The luminal boundary is not ideal and incorrectly includes small regions of calcification as the lumen. The 3D reconstruction of the luminal boundary has rough edges, and the distal internal carotid artery lumen mistakenly bleeds into other structures. Therefore, manual refinement was applied to correct the boundaries and to smooth the reconstructed surface, resulting in Fig 2, B. Fig 2, C demonstrates the automatically generated outer wall boundary superimposed on the refined lumen from Fig 2, B. Some extraneous tissue is included in the outer wall between the internal and external carotid arteries, and the outer portion of the vessel is not fully encapsulated (left sector of the axial slice). The 3D reconstruction, Fig 2, C, also indicates an incorrect inclusion of part of

= Fig 1. The overall workflow of the segmentation process. The initial setup involves file upload, surveying of images, and assigning of reference points along the vessel. The program automatically determines a luminal and then an outer wall outline, both of which can be refined. The final steps involve an automatic analysis for tissue composition while the proximal and distal limits of the lesion are manually defined.

4

Chrencik et al

Journal of Vascular Surgery ---

2019

Fig 2. Segmentation and three-dimensional (3D) reconstruction of lumen (A and B) and outer wall boundaries (C and D) in a stenosed artery. A and B, Lumen segmentation in an axial slice and 3D lumen reconstruction: (A) automatically generated before manual refinement and (B) after manual refinement. C and D, Outer wall segmentation in an axial slice and 3D outer wall reconstruction: (C) automatically generated before manual refinement and (D) after manual refinement. Vessels are labeled as common (pink), internal (beige), and external (purple) carotid arteries. The outer wall is labeled in green (internal carotid artery) and blue (external carotid artery).

the mandible, mistaken as vascular calcium, in the distal internal carotid artery. The outer wall boundary was adjusted manually to generate the final outer wall 3D reconstruction for this artery as displayed in Fig 2, D. After segmentation and after the vascular reconstruction has been finalized, plaque geometry and tissue composition are automatically computed for the entire artery. Fig 3, A displays a two-dimensional axial slice of the internal carotid artery (with plaque) and the external carotid artery. The plaque is composed of calcium (dark green), LRNC (yellow), IPH (red), and matrix (blue). When the algorithm is applied to the entire volume, it provides the 3D reconstructed and color-coded artery as seen in Fig 3, B. After the plaque lesion is defined, total plaque volume and North American Symptomatic Carotid Endarterectomy Trial (NASCET)-based percentage stenosis by diameter and area are determined for each artery.18 Statistical analysis. Statistical analysis was performed using MATLAB (version R 2016b; MathWorks Inc, Natick, Mass). Total plaque volume (volume between the luminal and adventitial boundaries) and individual tissue component volumes (calcium, LRNC, IPH, and matrix) were measured in cubic millimeters. Tissue components

were also expressed as a proportion of total plaque volume. Intraobserver and interobserver agreement for the measurements was assessed using Bland-Altman plots.19 Differences between measurements were plotted against the means of the same pairs of measurements. The 95% confidence intervals of the difference were used for the plots to assign the degree of agreement. Bias was computed as the average difference of the means for intraobserver and interobserver measurements. Variability in measurements was quantified as a function of the mean using the coefficient of variation, CV, such that (CV ¼ s/m), where s and m represent the standard deviation (SD) and mean of the difference, respectively. Intraclass and interclass correlation coefficients (ICCs) were used as repeatability measures for plaque volume and tissue composition measurements. The minimum detectable change (MDC) was computed for plaque volume and tissue composition (both absolute and proportional), assuming that the same observer and a different observer repeated the measurements. The MDC was computed as MDC ¼ 1.96 O2 SEM, where SEM is the standard error of measurement, SEM ¼ SDO1  ICC, and SD represents the mean SD across all the subjects for intraobserver and

Journal of Vascular Surgery Volume

-,

Number

Chrencik et al

5

-

Fig 3. Automatic color-coded plaque tissue compositional illustration in a stenosed artery. A, A typical axial slice with the external carotid artery (lumen; left, purple) and internal carotid artery (lumen; right, beige). B, Threedimensional (3D) reconstruction of the plaque tissue volume. Tissue components are defined as calcium (dark green), lipid-rich necrotic core (LRNC; yellow), intraplaque hemorrhage (IPH; red), and matrix (transparent region, blue in the axial slice).

interobserver measurements.20,21 Alignment of the actual geometric contours drawn for the lumen and the outer wall boundaries (intraobserver and interobserver) was evaluated by the Dice similarity coefficient (DSC) and modified Hausdorff distance (MHD).22,23 The DSC measures the amount of overlap between a pair of contours and ranges from 0 (no overlap) to 1 (complete overlap).22 The MHD quantifies average distance (in millimeters) between corresponding points of compared contours.23 Smaller MHD values signify greater similarity between analyzed contours. The DSC and MHD were determined for the lumen and the outer wall boundaries for each axial slice along the entire plaque in each patient. The results are presented as box-whisker plots expressing the distribution of these measures across all the images. Unless otherwise noted, values are reported as mean 6 SD.

RESULTS Patients’ characteristics. Table I presents the demographic and vascular risk factor information for the population of patients. A total of 23,548 axial images of 93 carotid arteries (left and right sides) from 50 patients were analyzed in this study; seven arteries were excluded because of the presence of a stent. For plaque tissue composition analysis, all arteries with $20% diameter stenosis were included (n ¼ 71).

Table I. Demographic and vascular risk factor information Variables

% or mean 6 SD

Age, years

70.1 6 8.2

Male sex

51

White

87.5

Diabetes mellitus

31.6

Coronary artery disease Hypertension

67.6 100

Smoking (past or present)

73.7

Hyperlipidemia

86.1

Asymptomatic

44

Symptomatic

56

SD, Standard deviation.

Plaque geometric characteristics. Plaque volume and maximum stenosis by area and by diameter were computed over the entire population. There was a broad range of atherosclerotic lesions represented in the cohort. The plaque volume was 837.3 6 431.3 mm3 (mean 6 SD), ranging from 118.34 mm3 to 2260.30 mm3. The mean 6 SD stenosis by diameter ([normal luminal diameter  least luminal diameter]/ normal luminal diameter) was 44.50% 6 25.62%, ranging from 1.01% to 87.53%. The mean 6 SD stenosis by area ([cross sectional area of normal lumen  cross sectional

6

Journal of Vascular Surgery

Chrencik et al

---

2019

Table II. Reliability and minimum change in total plaque volume, percentage stenosis, and tissue composition measurements detectable by computed tomography angiography (CTA) imaging (P < .01) Plaque Statistical measures

Volume

Stenosis By diameter

Composition By area

Calcium

Lipid core

Hemorrhage

Intraobserver ICC

0.96

0.87

0.83

0.99

0.96

0.86

CV

0.14

0.30

0.30

0.11

0.33

0.65

5.91%

9.51%

1.16

4.28

8.64

0.88

0.84

0.78

0.99

0.92

0.92

0.26

0.34

0.31

0.14

0.51

0.49

9.75%

12.72%

1.92

7.85

6.75

MDCa

33.17

Interobserver ICC CV MDCa

82.40

CV, Coefficient of variability; ICC, interclass and intraclass correlation coefficient; MDC, minimum detectable change. a MDC is expressed in cubic millimeters, except for measurements of percentage stenosis.

area at the thickest portion of the plaque]/area of normal lumen) was 58.12% 6 29.02%, ranging from 0.02% to 95.88%. Intraobserver and interobserver agreement (ICC), relative variability (CV), and ability to measure change (MDC) in total plaque volume, arterial stenosis by diameter, and stenosis by area are summarized in Table II. Total plaque volume measurements by the same observer were highly repeatable, relative variation was small, and ability to measure change was good (ICC, 0.96; CV, 0.14; and MDC, 33.17 mm3). Although these values were not as good when assessed by two different observers, the degree of agreement remained high (ICC, 0.88; CV, 0.26; and MDC, 82.40 mm3). Based on these results, if one observer were to repeat the measurements on serial CTA imaging, the observer would be able to detect a 3.9% change in total plaque volume with 95% confidence (MDC). If a different observer performed the second evaluation, the observer would be able to detect a 9.9% change in plaque volume with 95% confidence. Similar values, albeit with less precision, were found for diameter and stenosis by area (Table II). The DSC and MHD were computed for the shapes of the luminal and outer wall contours of plaque outlined in each axial image and the results are presented in Fig 4, A and B, respectively. There was excellent agreement between luminal and outer wall boundaries for both intraobserver and interobserver analyses. For the same observer, DSC (mean 6 SD) was 0.95 6 0.04 and 0.96 6 0.03 for the lumen and wall, respectively. For two different observers, it was 0.94 6 0.05 and 0.95 6 0.05, respectively. The red þ markers in Fig 4, A represent outliers based on DSC of data from a small percentage of the analyzed images (range, 4.96%-8.40%). Overall, MHD values of the lumen and outer wall were small. For the same observer, MHD was 0.16 6 0.10 mm and 0.19 6 0.15 mm for the lumen and wall, respectively. For two different observers, it was 0.19 6 0.12 mm and

0.23 6 0.25 mm, respectively. Outliers, red þ markers in Fig 4, B, are a small percentage of data (range, 3.40%-7.90%). Plaque tissue composition. The tissue composition of lesions causing $20% stenosis by diameter (n ¼ 71) was as follows: calcium, 114.26 6 121.87 mm3; LRNC, 49.78 6 59.21 mm3; and IPH, 38.05 6 49.73 mm3. Composition as a proportion of total plaque volume was as follows: calcium, 11.17% 6 9.52%; LRNC, 5.39% 6 5.60%; and IPH, 3.54% 6 3.55%. Intraobserver and interobserver variability for measurements of tissue composition is shown in Table II. Measurements for individual tissue constituents were highly repeatable by one observer, with ICC values of 0.99, 0.96, and 0.86 for calcium, LRNC, and IPH, respectively. Interobserver ICC values were also high at 0.99, 0.92, and 0.92, respectively. The minimum change in tissue constituents that the protocol can detect when measurements are repeated by the same observer is 1.16 mm3 for calcium, 4.28 mm3 for LRNC, and 8.64 mm3 for IPH. When measurements are repeated by a different observer, values are 1.92 mm3 for calcium, 7.85 mm3 for LRNC, and 6.75 mm3 for IPH. The Bland-Altman plots (Figs 5 and 6) confirm that intraobserver and interobserver differences in repeated measurements for total plaque volume and for plaque tissue constituents were low. Bias (mean difference) was 22.5 mm3 in intraobserver plaque volume measurements (Fig 5, A) and 11.25 mm3 in interobserver measurements (Fig 5, B). Bias in tissue component measurements by the same observer was 1.96 mm3, 1.15 mm3, and 4.29 mm3 for calcium (Fig 6, A), LRNC (Fig 6, C), and IPH (Fig 6, E). Interobserver bias was similarly low, displaying mean differences of 3.87 mm3, 2.32 mm3, and 2.96 mm3 for calcium (Fig 6, B), LRNC (Fig 6, D), and IPH (Fig 6, F). The study included 22 asymptomatic patients (44 arteries) and 28 patients with a previous stroke or transient

Journal of Vascular Surgery Volume

-,

Number

Chrencik et al

7

-

Fig 4. Reproducibility of geometric measures (shape) of carotid plaques generated on computed tomography angiography (CTA) imaging. The intraobserver and interobserver variabilities in the Dice similarity coefficient (DSC; A) and in the modified Hausdorff distance (MHD; B) for shapes of the carotid arterial lumen and outer wall boundary outlined in 23,548 individual CTA axial images. The red line indicates the 50th percentile; the lower and upper boundaries of the box (blue lines) represent the 25th and 75th percentiles. The red þ markers indicate outliers that formed a small percentage of the total analyzed images (only 4.96%-8.4% of all images by DSC and 3.4%-7.9% of all images by MHD). Intra-Lumen and Inter-Lumen, Intraobserver and interobserver variability of the outlined shape of the carotid arterial lumen; Intra-Wall and Inter-Wall, intraobserver and interobserver variability of the outlined shape of the carotid arterial outer wall boundary.

Fig 5. Bland-Altman plots for the intraobserver (A) and interobserver (B) variabilities in measurements of total plaque volume made by computed tomography angiography (CTA). The black lines (61.96 s) represent the 95% confidence interval levels, and the blue line (m) represents the mean difference between two measurements (bias).

ischemic attack (28 symptomatic arteries) before exclusion of arteries because of revascularization procedures. We found a trend toward increasing total plaque volume in asymptomatic plaques, increasing calcium in asymptomatic plaques, and increasing intraplaque hemorrhage in symptomatic plaques. Processing time requirements. Manual assignment of reference points to define the artery took 3 6 1 minutes per artery. Semiautomatic segmentation and editing of the lumen and outer wall were completed in 17 6 10 minutes per artery. Manual definition of the lesion and automatic composition analysis took 5 6 3 minutes. Therefore, the overall time for the entire analysis was 25 6 12 minutes per artery.

DISCUSSION Quantification of carotid plaque geometric features (volume, stenosis by area, stenosis by diameter) and tissue composition (calcium, LRNC, and IPH) is feasible using clinically acquired CTA images and yields highly reliable and repeatable results (ICC ranging from 0.78 to 0.99). The protocol used in our study detects small changes in total plaque volume, diameter stenosis, area stenosis, and individual tissue constituents. Processing time is acceptable in a clinical research setting and could potentially be used clinically. We find that one observer can detect a 3.9% change in total plaque volume, whereas repeated measurements by a different observer can detect volume changes of 9.9%. One observer can detect a change of 1.16 mm3, 4.28 mm3, and 8.64 mm3

8

Chrencik et al

Journal of Vascular Surgery ---

2019

Fig 6. Bland-Altman plots for the intraobserver and interobserver variabilities in measurements of individual tissue composition. A, C, and E, Intraobserver plots for measurements of calcium, lipid-rich necrotic core (LRNC), and intraplaque hemorrhage (IPH), respectively. B, D, and F, Interobserver plots of calcium, LRNC, and IPH, respectively. The black lines (61.96 s) represent the 95% confidence interval levels, and the blue line (m) represents the mean difference between two measurements (bias).

in plaque calcium, LRNC, and IPH, respectively, whereas a different observer can detect a change of 1.92 mm3, 7.85 mm3, and 6.75 mm3 for calcium, LRNC, and IPH, respectively.

Plaque disruption is the primary contributor in the pathogenesis of cerebral infarction in carotid stenosis and is likely to be influenced by morphologic changes occurring in the plaque.24 Lesions originate from fatty

Journal of Vascular Surgery Volume

-,

Number

Chrencik et al

9

-

streaks that coalesce into lipid cores.25,26 Some plaques continue enlarging their LRNC and develop IPH and fibrous cap thinning, ultimately leading to disruption.4,27-33 A review of 31 studies with 4447 explanted carotid plaques indicated a strong association between IPH and stroke.34 Whereas the association between plaque morphology and disruption has long been suspected, confirmation of this hypothesis requires reliable plaque assessment at baseline followed by a longitudinal assessment of clinical outcomes. Our findings suggest that CTA, when processed by our protocol, can characterize plaque morphology (geometry and composition) with high reliability and detect small proportional changes in these features. The results describe the limits of minimum changes detectable by CTA. This may facilitate noninvasive identification of plaques at high risk for embolization. Smooth homogeneous fibrous plaques are clinically stable and less prone to atheroembolization.27 Carotid plaques may also stabilize, leading to a phenotype reversal from high- to normal-risk plaques, depending on their surrounding milieu.35,36 For instance, their histologic appearance may transition from unstable to stable morphology after a stroke event.37 Lipid-lowering therapy may also change the composition to a more stable morphology or reduce overall plaque volume.38 The Global Assessment of Plaque Regression with a PCSK9 Antibody as Measured by Intravascular Ultrasound (GLAGOV) trial used invasive coronary intravascular ultrasound to measure plaque volume at baseline and after treatment, an approach that cannot be replicated in the carotid artery because of unacceptable atheroembolic stroke risk.39 Our analysis suggests that CTA-based image analysis is a reliable, noninvasive imaging alternative for such pharmacologic trials. Although not available for comparison to CTA in our cohort, several studies have reported that DUS and MRI can define carotid plaque morphology.4,6 Possible reasons for the inability to translate these reports into clinical practice may be the specialized equipment needed for MRI and 3D DUS, prolonged imaging times, and operator variability for DUS. Despite being commonly performed clinically, CTA has not been systematically evaluated for making detailed assessments of morphology. It provides excellent lumen-tissue interface delineation, which can be expected to provide reliable measures of plaque geometry (volume, area, and diameter). Multidetector CTA equipment is readily available in clinical centers, protocols are well standardized, and the study can be performed quickly (in 3-5 minutes). Although our study systematically examines CTA as a technique to determine plaque morphology, earlier reports have hinted at this ability of CTA. Carotid plaque brightness, measured in Hounsfield units, was lower in 15 symptomatic patients compared with 23 asymptomatic patients, suggesting that more of the plaque was

composed of high-risk LRNC and IPH.40 Furthermore, 13 plaques with a mixture of high-risk histologic tissue (LRNC, IPH) showed larger amounts of low-Hounsfield unit regions in CTA images compared with 18 plaques with more stable histology (fibrous tissue).41 These qualitative studies have subsequently been substantiated by more direct comparisons of CTA with specific tissue components. Using a combination of in vivo CTA and ex vivo micro-computed tomography, the mean Hounsfield unit brightness intensity corresponding to LRNC, IPH, and calcification was identified on the basis of corresponding histologic specimens of eight explanted carotid plaques.42 Manual outlining of plaque boundaries is time-consuming, and additional processing time is added by manual segmentation algorithms to segment tissue types on CTA images, resulting in analysis times of 1 to 2 hours. Semiautomatic algorithms have recently been developed that correlate Hounsfield unit brightness thresholds with tissue types and account for blooming artifacts associated with calcification.15,16,41-43 Our results demonstrate that with use of these algorithms, carotid morphometry can be performed consistently from clinically acquired CTA images in a shorter time. The approach is not without limitations. The interobserver and intraobserver correlation coefficients for IPH were lower compared with the other tissues measured. Although this can be explained by the fact that IPH was the least of the tissue constituents in a plaque (mean volume, 3.54% of total plaque volume) and therefore presented the smallest target to outline, it does indicate that IPH may not be quantified as reliably as other tissues,44,45 and this is the subject of ongoing research in our laboratory. Even though the algorithm mitigates the effects of artifacts, such as streak artifacts from implanted hardware, in extreme cases, this can still compromise the quality of segmentation and analysis. Although the most intensive steps in the analysis are automated, it still requires manual input in marking reference points and editing the segmented outlines, driving processing times to approximately 25 minutes. This compares favorably with currently available processing times for 3D DUS and MRI-based morphology assessment, but there is room for improvement, and algorithms are being continually refined and enhanced. Although image processing time for CTA may be extensive, imaging time for CTA (<10 minutes) compares favorably with that for 3D DUS (30 minutes) and MRI (20-45 minutes). Therefore, the total time needed to obtain and to process CTA data may not be much more that that required for other modalities.

CONCLUSIONS This study demonstrates the first comprehensive analysis of a semiautomatic approach to quantifying the repeatability of and MDC achievable with clinical

10

Chrencik et al

Journal of Vascular Surgery ---

CTA for the measurement of carotid plaque volume, stenosis measures, and tissue volumes. Our analyses included 23,548 images from 93 arteries in 50 patients and involved >94,000 comparisons, and this approach has been demonstrated to be repeatable, reliable, and sensitive to small changes in carotid plaque morphology. These findings will facilitate future longitudinal studies to track plaque progression or regression in large, randomized trials, such as the Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis Trial (CREST-2), and aid in risk stratification for future stroke.46

AUTHOR CONTRIBUTIONS Conception and design: BL Analysis and interpretation: MC, AK, LL, JP, JS, SS Data collection: MC, AK, LL, LA, JY Writing the article: MC, AK, BL Critical revision of the article: MC, AK, LL, LA, JY, JP, JS, SS, BL Final approval of the article: MC, AK, LL, LA, JY, JP, JS, SS, BL Statistical analysis: AK, JS Obtained funding: BL Overall responsibility: BL

REFERENCES 1. Gupta A, Baradaran H, Schweitzer AD, Kamel H, Pandya A, Delgado D, et al. Carotid plaque MRI and stroke risk: a systematic review and meta-analysis. Stroke 2013;44:3071-7. 2. Wendorff C, Wendorff H, Pelisek J, Tsantilas P, Zimmermann A, Zernecke A, et al. Carotid plaque morphology is significantly associated with sex, age, and history of neurological symptoms. Stroke 2015;46:3213-9. 3. Howard DP, Van Lammeren GW, Rothwell PM, Redgrave JN, Moll FL, de Vries JP, et al. Symptomatic carotid atherosclerotic disease: correlations between plaque composition and ipsilateral stroke risk. Stroke 2015;46:182-9. 4. Cires-Drouet RS, Mozafarian M, Ali A, Sikdar S, Lal BK. Imaging of high-risk carotid plaques: ultrasound. Semin Vasc Surg 2017;30:44-53. 5. Spence JD, Eliasziw M, DiCicco M, Hackam DG, Galil R, Lohmann T. Carotid plaque area: a tool for targeting and evaluating vascular preventive therapy. Stroke 2002;33: 2916-22. 6. Sun J, Zhao XQ, Balu N, Neradilek MB, Isquith DA, Yamada K, et al. Carotid plaque lipid content and fibrous cap status predict systemic CV outcomes: the MRI substudy in AIMHIGH. JACC Cardiovasc Imaging 2017;10:241-9. 7. McNally JS, McLaughlin MS, Hinckley PJ, Treiman SM, Stoddard GJ, Parker DL, et al. Intraluminal thrombus, intraplaque hemorrhage, plaque thickness, and current smoking optimally predict carotid stroke. Stroke 2015;46:84-90. 8. Nandalur KR, Hardie AD, Raghavan P, Schipper MJ, Baskurt E, Kramer CM. Composition of the stable carotid plaque: insights from a multidetector computed tomography study of plaque volume. Stroke 2007;38:935-40. 9. Khan AA, Koudelka C, Goldstein C, Zhao L, Yokemick J, Dux M, et al. Semiautomatic quantification of carotid plaque volume with three-dimensional ultrasound imaging. J Vasc Surg 2017;65:1407-17. 10. Sutton-Tyrrell K, Wolfson SK, Thompson T, Kelsey SF. Measurement variability in duplex scan assessment of carotid atherosclerosis. Stroke 1992;23:215-20.

2019

11. Gaitini D, Soudack M. Diagnosing carotid stenosis by Doppler sonography. J Ultrasound Med 2005;24:1127-36. 12. Buskens E, Nederkoorn PJ, Buijs-Van Der Woude T, Mali WP, Kappelle LJ, Eikelboom BC, et al. Imaging of carotid arteries in symptomatic patients: cost-effectiveness of diagnostic strategies. Radiology 2004;233:101-12. 13. Silvennoinen HM, Ikonen S, Soinne L, Railo M, Valanne L. CT angiographic analysis of carotid artery stenosis: comparison of manual assessment, semiautomatic vessel analysis, and digital subtraction angiography. Am J Neuroradiol 2007;28: 97-103. 14. Chen CJ, Lee TH, Hsu HL, Tseng YC, Lin SK, Wang LJ, et al. Multi-slice CT angiography in diagnosing total versus near occlusions of the internal carotid artery: comparison with catheter angiography. Stroke 2004;35:83-5. 15. Sheahan M, Ma X, Paik D, Obuchowski NA, St. Pierre S, Newman WP III, et al. Atherosclerotic plaque tissue: noninvasive quantitative assessment of characteristics with software-aided measurements from conventional CT angiography. Radiology 2018;286:622-31. 16. de Weert TT, Ouhlous M, Meijering E, Zondervan PE, Hendriks JM, van Sambeek MR, et al. In vivo characterization and quantification of atherosclerotic carotid plaque components with multidetector computed tomography and histopathological correlation. Arterioscler Thromb Vasc Biol 2006;26:2366-72. 17. van Engelen A, Niessen WJ, Klein S, Groen HC, Verhagen HJ, Wentzel JJ, et al. Atherosclerotic plaque component segmentation in combined carotid MRI and CTA data incorporating class label uncertainty. PloS One 2014;9:e94840. 18. Moneta GL, Edwards JM, Chitwood RW, Taylor LM, Lee RW, Cummings CA, et al. Correlation of North American Symptomatic Carotid Endarterectomy Trial (NASCET) angiographic definition of 70% to 99% internal carotid artery stenosis with duplex scanning. J Vasc Surg 1993;17:152-9. 19. Bland JM, Altman D. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;327:307-10. 20. Beckerman H, Roebroeck ME, Lankhorst GJ, Becher JG, Bezemer PD, Verbeek AL. Smallest real difference, a link between reproducibility and responsiveness. Qual Life Res 2001;10:571-8. 21. Stratford PW. Estimating the standard error of measurement from reliability studies. Physiother Can 2004;56:27-30. 22. Dice LR. Measures of the amount of ecologic association between species. Ecology 1945;26:297-302. 23. Dubuisson MP, Jain AK. A modified Hausdorff distance for object matching. Proceedings of the 12th International Conference on Pattern Recognition. Available at: http://ieeexplore. ieee.org/xpls/abs_all.jsp?arnumber¼576361. Accessed July 28, 2016. 24. Golledge J, Greenhalgh RM, Davies AH. The symptomatic carotid plaque. Stroke 2000;31:774-81. 25. Stary HC, Chandler AB, Glagov S, Guyton JR, Insull W Jr, Rosenfeld ME, et al. A definition of initial, fatty streak, and intermediate lesions of atherosclerosis. A report from the Committee on Vascular Lesions of the Council on Arteriosclerosis, American Heart Association. Circulation 1994;89: 2462-78. 26. Stary HC, Chandler AB, Dinsmore RE, Fuster V, Glagov S, Insull W Jr, et al. A definition of advanced types of atherosclerotic lesions and a histological classification of atherosclerosis. A report from the Committee on Vascular Lesions of the Council on Arteriosclerosis, American Heart Association. Arterioscler Thromb Vasc Biol 1995;15:1512-31.

Journal of Vascular Surgery Volume

-,

Number

Chrencik et al

11

-

27. Stary HC. Natural history and histological classification of atherosclerotic lesions: an update. Arterioscler Thromb Vasc Biol 2000;20:1177-8. 28. Bassiouny HS, Sakaguchi Y, Mikucki SA, McKinsey JF, Piano G, Gewertz BL, et al. Juxtalumenal location of plaque necrosis and neoformation in symptomatic carotid stenosis. J Vasc Surg 1997;26:585-94. 29. Carr S, Farb A, Pearce WH, Virmani R, Yao JS. Atherosclerotic plaque rupture in symptomatic carotid artery stenosis. J Vasc Surg 1996;23:755-6. 30. Avril G, Batt M, Guidoin R, Marois M, Hassen-Khodja R, Daune B, et al. Carotid endarterectomy plaques: correlations of clinical and anatomic findings. Ann Vasc Surg 1991;5:50-4. 31. Schaar JA, Muller JE, Falk E, Virmani R, Fuster V, Serruys PW, et al. Terminology for high-risk and vulnerable coronary artery plaques. Eur Heart J 2004;25:1077-82. 32. Park AE, McCarthy WJ, Pearce WH, Matsumura JS, Yao JS. Carotid plaque morphology correlates with presenting symptomatology. J Vasc Surg 1998;27:872-9. 33. Fisher M, Paganini-Hill A, Martin A, Cosgrove M, Toole JF, Barnett HJ, et al. Carotid plaque pathology: thrombosis, ulceration, and stroke pathogenesis. Stroke 2005;36:253-7. 34. Gao P, Chen ZQ, Bao YH, Jiao LQ, Ling F. Correlation between carotid intraplaque hemorrhage and clinical symptoms: systematic review of observational studies. Stroke 2007;38: 2382-90. 35. Li ZY, Tang TY, Jiang F, Zhang Y, Gillard JH. Reduction in arterial wall strain with aggressive lipid-lowering therapy in patients with carotid artery disease. Circ J 2011;75:1486-92. 36. Chatzizisis YS, Coskun AU, Jonas M, Edelman ER, Feldman CL, Stone PH. Role of endothelial shear stress in the natural history of coronary atherosclerosis and vascular remodeling. Molecular, cellular, and vascular behavior. J Am Coll Cardiol 2007;49:2379-93. 37. Redgrave JN, Lovett JK, Rothwell PM. Histological features of symptomatic carotid plaques in relation to age and smoking: the Oxford plaque study. Stroke 2010;41:2288-94. 38. Belcaro G, Nicolaides AN, Ramaswami G, Cesarone MR, De Sanctis M, Incandela L, et al. Carotid and femoral ultrasound

39.

40.

41.

42.

43.

44.

45.

46.

morphology screening and cardiovascular events in low risk subjects: a 10-year follow-up study (the CAFES-CAVE study 1). Atherosclerosis 2001;156:379-87. Nicholls SJ, Puri R, Anderson T, Ballantyne CM, Cho L, Kastelein JJ, et al. Effect of evolocumab on progression of coronary disease in statin-treated patients: the GLAGOV randomized clinical trial. JAMA 2016;316:2373-84. Haraguchi K, Houkin K, Koyanagi I, Nonaka T, Baba T. Evaluation of carotid plaque composition by computed tomography angiography and black blood magnetic resonance image. Interv Neuroradiol 2008;14:39-43. Veraldi GF, Nocini PF, Eccher A, Fenzi A, Sboarina A, Mezzetto L. Correlation between MDCTA and carotid plaque histological heterogeneity: a pilot study. Eur J Vasc Endovasc Surg 2018;56:7-14. Wintermark M, Jawadi SS, Rapp JH, Tihan T, Tong E, Glidden DV, et al. High-resolution CT imaging of carotid artery atherosclerotic plaques. Am J Neuroradiol 2008;29: 875-82. Pierre SS, Siegelman J, Obuchowski NA, Ma X, Paik D, Buckler AJ. Measurement accuracy of atherosclerotic plaque structure on CT using phantoms to establish ground truth. Acad Radiol 2017;24:1203-15. Virmani R, Kolodgie FD, Burke AP, Finn AV, Gold HK, Tulenko TN, et al. Atherosclerotic plaque progression and vulnerability to rupture: angiogenesis as a source of intraplaque hemorrhage. Arterioscler Thromb Vasc Biol 2005;25: 2054-61. Lal BK, Cerveira JJ, Seidman C, Haser PB, Kubicka R, Jamil Z, et al. Observer variability of iliac artery measurements in endovascular repair of abdominal aortic aneurysms. Ann Vasc Surg 2004;18:644-52. Lal BK, Meschia JF, Brott TG. Clinical need, design, and goals for the Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis trial. Semin Vasc Surg 2017;30:2-7.

Submitted Sep 28, 2018; accepted Nov 26, 2018.