Original Article
A Novel Ultrasound-Based Carotid Plaque Risk Index Associated with the Presence of Cerebrovascular Symptoms Ein neuer ultraschallbasierter Karotis-Plaque-Risikoindex ist assoziiert mit dem Auftreten zerebrovaskulärer Symptome Authors
B. Kanber1, T. C. Hartshorne2, M. A. Horsfield1, A. R. Naylor2, T. G. Robinson1, 3, K. V. Ramnarine4
Affiliations
1
3 4
Department of Cardiovascular Sciences, University of Leicester, Leicester, UK Department of Vascular and Endovascular Surgery, University Hospitals of Leicester NHS Trust, Leicester, UK NIHR Biomedical Research Unit for Cardiovascular Sciences, University of Leicester, Leicester, UK Department of Medical Physics, University Hospitals of Leicester NHS Trust, Leicester, UK
Key words
Abstract
Zusammenfassung
"
!
!
Purpose: The purpose of this study was to determine the efficacy of a novel ultrasound-based carotid plaque risk index (CPRI) in predicting the presence of cerebrovascular symptoms in patients with carotid artery stenosis. Materials and Methods: This was a cross-sectional, observational study involving 56 patients (mean age 76.6 years, 62.5 % male). Plaque grayscale median (GSM) and surface irregularity indices (SII) were measured in 82 stenosed carotid arteries (range 10 – 95 %) and combined with the degree of stenosis (DOS) in the form of (DOS*SII)/(1 + GSM). A reduced index DOS/(1 + GSM) not incorporating plaque surface irregularities was also investigated. Receiver operating characteristic curves (ROC) were used to study the diagnostic efficacy of CPRI, comparing against DOS and an equivalent risk index constructed using a conventional logistic regression based method with model parameters optimized to the dataset (CPRIlogistic). Results: There were 42 stenosed carotid arteries with cerebrovascular symptoms, and 40 without symptoms. The presence of symptoms significantly correlated with DOS, GSM and SII (p < 0.01). The median CPRI of the symptomatic (asymptomatic) groups were 23.2 (9.2) compared with 0.71 (0.30) for CPRIlogistic (p < 0.01). The diagnostic performance of CPRI exceeded that of CPRIlogistic and DOS, and demonstrated a better separation of the symptomatic and asymptomatic groups. Conclusion: Our novel risk index combines quantitative measures of carotid plaque echogenicity and surface irregularities with the degree of stenosis. It is a better predictor of cerebrovascular symptoms than the degree of stenosis and could be valuable in studies and clinical trials aimed at identifying vulnerable carotid artery stenoses.
Ziel: Ziel der Studie war die Bestimmung der Leistungsfähigkeit des neuen ultraschallbasierten Karotis-Plaque-Risikoindex (CPRI) zur Vorhersage von zerebrovaskulären Symptomen bei Patienten mit Karotisstenose. Material und Methoden: Eine Querschnitts-Beobachtungsstudie in 56 Patienten (∅ 76,6 Jahre, 62,5 % ♂). Medianer Grauwert (GSM) der Plaque und Oberflächenunregelmäßigkeits-Indizes (SII) wurden in 82 stenosierten Karotisarterien (Bereich 10 – 95 %) bestimmt und mit dem Stenosegrad (DOS) mittels der Formel (DOS*SII)/(1 + GSM) verbunden. Ein reduzierter Index DOS/(1 + GSM), ohne Berücksichtigung von SII wurde untersucht. „Receiver Operating Characteristic“ (ROC)-Kurven wurden angewandt, um die diagnostische Effizienz des CPRI mit DOS und einem gleichwertigen Risikoindex zu vergleichen, letzterer erstellt mittels einer auf konventionelle logistische Regression basierenden Methode mit für den Datensatz optimierten Modellparametern (CPRIlogistic). Ergebnisse: 42 Stenosen mit zerebrovaskulären Symptomen und 40 symptomlose Stenosen lagen vor. Das Auftreten von Symptomen korrelierte signifikante mit DOS, GSM und SII (p < 0,01). Der Median des CPRI in den symtomatischen (asymtomatischen) Gruppen betrug 23,2 % (9,2) im Vergleich zu 0,71 (0,30) des CPRIlogistic der symptomatischen (asymtomatischen) Gruppen (p < 0,01). In diagnostischer Leistung war CPRI der CPRIlogistic und dem DOS überlegen und zeigte eine bessere Trennung zwischen symptomatischen und asymptomatischen Gruppen. Schlussfolgerung: Unser neuer Risikoindex vereinigt die quantitative Messung der KarotisplaqueEchogenität und der Unregelmäßigkeiten der Oberfläche mit dem Grad der Stenose. Es ist ein besserer Prädiktor für zerebrovaskuläre Symptome als der DOS und könnte sich in Untersuchungen und klinischen Studien, die eine Identifizierung
● carotid plaque ● risk index ● CPRI " "
received accepted
15.3.2014 14.9.2014
Bibliography DOI http://dx.doi.org/ 10.1055/s-0034-1385462 Published online: 2014 Ultraschall in Med © Georg Thieme Verlag KG Stuttgart · New York · ISSN 0172-4614 Correspondence Dr. Baris Kanber Department of Cardiovascular Sciences, University of Leicester, Leicester, United Kingdom Tel.: ++ 44/1 16/2 58 51 40 Fax: ++ 44/1 16/2 58 60 70 [email protected]
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
2
Introduction
Analysis
!
!
The degree of stenosis (DOS) of the carotid artery is an established parameter that is routinely used in clinical practice [1, 2]. However, it is recognized that some plaques may be more vulnerable than others, and their identification would aid clinical decision making [3]. Previous investigations have shown that ultrasound plaque characteristics such as the plaque grayscale median (GSM) and plaque surface irregularities may be useful for identifying such vulnerable plaques [4 – 21]. Neovascularization and intra-plaque hemorrhage are other potentially important markers of plaque vulnerability, but they are more readily assessed using contrastenhanced ultrasound (CEUS) [22]. Plaque compression, deformation, intra-plaque/fibrous cap strains, and shear forces are all additional, potentially important markers of plaque vulnerability. Several studies have attempted to combine ultrasound-plaque measures with the degree of stenosis in order to derive a plaque risk index but either qualitative measures have been employed and/or the relevant risk index was based on cut-off values or weighting factors optimized to a particular dataset [23 – 28]. The incorporation of dataset-optimized factors limits applicability and validity, since the assessment of ultrasound plaque characteristics, including that of the degree of stenosis, can vary depending on the method of assessment and can be subject to variation between centers [4, 29]. We have developed an ultrasound-based carotid plaque risk index (CPRI) that is based on quantitative measures of plaque echogenicity and surface irregularities, combining these with the degree of stenosis without incorporating any parameters optimized for our dataset. Our study compares the performance of this risk index with that of DOS and a conventional logistic regressionbased model with dataset-optimized weighting factors.
Quantitative analyses were carried out using MATLAB version 7.14 (MathWorks, Natick, Massachusetts, USA) and SPSS version 20 (IBM Corporation, Armonk, New York, USA). The degree of carotid artery stenosis was measured by experienced vascular scientists using criteria consistent with the NASCET method, utilizing blood flow velocities in conjunction with B-mode and color flow imaging findings [2, 30, 31]. The measurements were carried out not using a single longitudinal imaging plane but multiple imaging planes both in the longitudinal and transverse cross-section. However, as Doppler velocity measurements are not able to reliably discriminate degrees of stenosis below 50 %, B-mode diameter measurements and color flow imaging were used to grade the degree of stenosis into deciles for minor stenoses. The readers are directed to the appropriate references for a detailed consensus description of the measurement of the degree of carotid artery stenosis using Duplex ultrasound methods [2, 30, 31]. In relation to grayscale plaque characteristics, previously described methods were employed to evaluate the normalized plaque GSM [4] and the plaque surface irregularity index (SII) [5]. SII was calculated by computationally summing the angular deviations from a straight line, of the luminal plaque surface, and dividing this by the physical length of the plaque surface, and is described in detail elsewhere [5]. The luminal plaque surface is the boundary separating the plaque body from the bloodstream and " Fig. 1. is where the green and purple dashed lines overlap in ● Since SII is an angle divided by a length, it is measured in units of radians (or degrees) per meter (or centimeter). SII has been shown to be a reproducible measurement with a mean intra-observer coefficient of vari7ation (CV) of 4.4 %, and a mean intra-observer, inter-frame CV of 10.6 % [5]. GSM, on the other hand, is simply the median grayscale (echo) intensity within the plaque body [4]. Plaque bodies are the regions bounded by the green da" Fig. 1. For the normalized plaque GSM, a mean shed lines in ● intra-observer CV of 2.8 % was previously reported [4]. We averaged GSM and SII values over all ultrasound image frames acquired for each artery. On average, 180 consecutive frames were analyzed for each artery. The non-parametric Wilcoxon-MannWhitney test was used to assess whether quantitative measures differed significantly between symptomatic and asymptomatic groups, and ROC curves were used to assess classification performance. Two-tailed tests of significance were used and p-values less than 0.05 were considered statistically significant. A carotid plaque risk index based on logistic regression methods (CPRIlogistic) was constructed, similar to Momjian-Mayor et al. [24] in the form of 1/(1 + exp(-x)) where x is equal to β0 + β1*DOS + β2*GSM + β3*SII and β0, β1, β2, β3 are factors that are optimized for the data. Our CPRI, on the other hand, was defined as (DOS*SII)/(GSM + 1) with no dataset-optimized parameters. DOS values in this equation were between 0.0 and 1.0 (e. g. 0.5 for 50 % stenosis), while the SII were input in radians per meter. The value 1 was added to the GSM to avoid having a mathematical singularity for plaques with a GSM of 0. Box and whisker plots were used to study the distribution of CPRI and CPRIlogistic among the plaque groups with and without associated cerebrovascular symptoms.
Methods !
This was a cross-sectional study in which 56 patients (35 male, 21 female) who attended the University Hospitals of Leicester NHS Trust's Rapid Access Transient Ischemic Attack (TIA) clinic were recruited. Ultrasound image sequences of the carotid plaque in the longitudinal cross-section were acquired by experienced sonographers using a Philips iU22 ultrasound scanner (Philips Healthcare, Eindhoven, The Netherlands) and an L9 – 3 probe. B-mode (grayscale) and color Doppler image sequences were recorded as DICOM files over an average duration of 5.6 seconds (mean frame rate was 32 frames per second for the B-mode acquisitions) using the vascular carotid preset on the scanner (Vasc Car™ preset, persistence low, XRES™ and SONOCT™ on). Color Doppler image sequences were used as a qualitative aid to identifying the location and extent of the plaques, while the grayscale data were used for the quantitative analyses. The study was approved by the National Research Ethics Service (NRES) Committee East Midlands – Northampton (reference 11/EM/ 0249) and followed institutional guidelines. Each patient gave informed consent before participating in the study. 82 stenosed carotid arteries (stenosis range 10 – 95 %) were investigated. Plaques were classified as either having caused cerebrovascular symptoms relating to the ipsilateral brain hemisphere within the past six-month period (i. e. symptomatic) or as asymptomatic following the review of patient symptoms and clinical/radiological (CT/MRI) findings.
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Original Article
Fig. 1 Examples of luminal plaque surfaces (where the purple and green dashed lines overlap) and plaque bodies (bounded by the green dashed lines). Abb. 1 Beispiele für luminale Plaque-Oberflächen (wo sich die violetten und grünen gestrichelten Linien überlappen) und Plaquekörper (umgrenzt von den grünen gestrichelten Linien).
Results !
The mean age of the patients was 76.6 (range 58 to 95) years. Patient age was not significantly different between the asymptomatic (76.9 years) and symptomatic groups (76.4 years, p > 0.05). Patients who had previous TIA/stroke episodes (44.6 %) were more likely to be symptomatic (p = 0.01), while other patient characteristics did not have a significant association to the presence of " Table 1). symptoms (● There were 82 carotid arteries with stenoses ranging from 10 % to 95 %. The distribution of the stenoses was as follows: 49 with DOS less than 50 %, 24 with DOS between 50 %-70 %, and 9 with DOS greater than 70 %. 42 of these arteries were associated with symptoms relating to the ipsilateral brain hemisphere, while 40 were asymptomatic. The degree of stenosis and the SII were significantly higher in the arteries with symptoms, while the normalized plaque GSM was lower (p < 0.01 for all). The mean SII of symptomatic plaques was 1970 radians/m compared to 1780 radians/m for the asymptomatic ones. The corresponding normalized GSM values were 45.1 and 59.5, respectively. Patients for which image quality was not sufficiently good (for example complete shadowing) were excluded from the study, but there were only a small number of such cases (n = 1). 7 patients had atrial fibrillation present on pre-scan ECG. 3 of these patients were asymptomatic since they were not diagnosed to have suffered a cerebrovascular event. The remaining 4 patients had no brain infarcts observed on CT/MRI, but their body symptoms were contralateral to their stenosed carotid arteries. Examples of a plaque with ipsilateral hemispheric symptoms and " Fig. 2. The risk index a plaque without symptoms are shown in ● based on optimized logistic regression (CPRIlogistic) and our CPRI were significantly higher for stenoses that were associated with symptoms (p < 0.01 for both), but CPRI showed a better separa-
Fig. 2 Example of a symptomatic plaque (top) and an asymptomatic plaque (bottom) with respective risk indices (CPRI) of 27.5 and 3.25. The degree of stenosis (DOS) caused by the symptomatic plaque was 70 %, while it was 20 % for the asymptomatic plaque. The value contrast between the symptomatic and the asymptomatic case was better with CPRI than with DOS. The ratio between the two degrees of stenosis was 70/20 (3.5x), while the ratio between the two risk indices was 27.5/3.25 (8.5x). Abb. 2 Beispiel einer symptomatischen Plaque (oben) und einer asymptomatischen Plaque (unten) mit den entsprechenden Risiko-Indizes (CPRI) von 27,5 (symptomatisch) und 3,25 (asymptomatisch). Der durch die symptomatische Plaque verursachte Stenosegrad (DOS) betrug 70 %, der durch die asymptomatischen Plaque verursachte hingegen 20 %. Der Kontrastwert zwischen dem symptomatischen und asymptomatischen Fall war mit CPRI besser als mit DOS; die Ratio zwischen den beiden Stenosegraden betrug 70/20 (3,5x), während die Ratio zwischen den beiden Indizes 27,5/ 3,25 (8,5x) betrug.
" Fig. 3). ROC curve analysis showed that tion of the two groups (● CPRIlogistic and CPRI had similar classification performance, better than that of DOS, with CPRI providing a better overall accuracy " Fig. 4, " Table 2). The area under the ROC curve (AUC) was (● ● " Table 2). 0.849 for CPRI, 0.845 for CPRIlogistic and 0.771 for DOS (● The median CPRI of the symptomatic and asymptomatic groups were 23.2 and 9.2 units compared with 0.71 and 0.30 for CPRIlogistic. GSM and SII appeared to have similar ROC curves, which were " Fig. 4). A reduced risk index, not utilizing below that of DOS (● plaque surface irregularities, of the form DOS/(GSM + 1) also out" Fig. 5). performed DOS on its own (●
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Original Article
Original Article
patient characteristic
mean age (years) sex (males)
all patients
asymptomatic
symptomatic
significance
group
group
(p-value)
76.6
76.9
76.4
0.80
62.5 %
54.5 %
67.6 %
0.33
hypertension
64.3 %
63.6 %
64.7 %
0.94
hypercholesterolemia
53.6 %
54.5 %
52.9 %
0.91
ischemic heart disease
23.2 %
13.6 %
29.4 %
0.18
diabetes
35.7 %
22.7 %
44.1 %
0.11
previous TIA/stroke
44.6 %
22.7 %
58.8 %
0.01*
peripheral vascular disease
14.3 %
9.1 %
17.6 %
0.38
smoking
67.9 %
68.2 %
67.6 %
0.97
alcohol consumption
37.5 %
27.3 %
44.1 %
0.21
family history of stroke
33.9 %
22.7 %
41.2 %
0.16
atrial fibrillation
12.5 %
13.6 %
11.8 %
stroke (n = 8)
14.3 %
not applicable
23.5 %
not applicable
TIA (n = 16)
28.6 %
not applicable
47.1 %
not applicable
transient vision loss (n = 10)
17.9 %
not applicable
29.4 %
not applicable
Table 1 Patient characteristics and the significance of association with cerebrovascular symptoms.
0.84
Fig. 3 Box and whisker plots showing the distribution of CPRIlogistic (top left) and CPRI (bottom left) in carotid artery stenoses with and without cerebrovascular symptoms. The corresponding plots on the right further show the distribution of CPRIlogistic (top right) and CPRI (bottom right) within the two groups. Abb. 3 Die Box-Whisker-Plots zeigen die Verteilung von CPRIlogistic (oben links) und CPRI (unten links) in Karotisstenosen mit und ohne zerebrovaskuläre Symptome. Die entsprechenden Darstellungen rechts zeigen die Verteilung von CPRIlogistic (oben rechts) und CPRI (unten rechts) innerhalb der beiden Gruppen.
Discussion !
Our study described a novel risk index which intuitively combines quantitative measures of plaque echogenicity and surface irregularities with the degree of carotid artery stenosis. This risk index, despite not having any parameters optimizing it for our dataset, was found to have increased diagnostic performance compared to DOS, and a risk index based on logistic regression with such optimized parameters. Several studies have previously derived risk indices that combine ultrasound plaque characteristics with DOS in order to help identify vulnerable carotid plaques [23 – 28]. Prati et al. defined a risk index incorporating DOS, and qualitative measures of surface irregularities, plaque echogenicity and texture [23]. Momjian-
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Mayor et al. defined a risk index combining DOS with a quantitative measure of the plaque surface echogenicity and used a logistic regression-based statistical model with weighting factors optimized for their dataset [24]. Pedro et al. similarly developed an activity index, comprising several measures including that of plaque surface disruption (qualitative assessment), GSM, percentage of plaque area with gray levels less than 40, plaque heterogeneity (qualitative assessment), and the presence of juxta-luminal echolucent areas (qualitative assessment), in addition to DOS. Weighting scores for various criteria based on these parameters were evaluated and used to assign an activity index to each plaque [25, 26]. This was later extended by Seabra et al. to an enhanced activity index which incorporated additional parameters, including those based on the radiofrequency (RF) ultrasound signal. Ni-
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
The statistical methods used to test the associations were the non-parametric Wilcoxon-Mann-Whitney test for patient age and the χ2 test for the rest of the patient characteristics. Significant associations are marked with an asterisk (*).
Original Article
Fig. 4 ROC curves showing the classification performance of the degree of stenosis (DOS), plaque surface irregularity index (SII), the normalized plaque grayscale median (GSM), CPRIlogistic and CPRI. Abb. 4 Die ROC-Kurven zeigen die Klassifizierungsleistung für den Grad der Stenose (DOS), den Oberflächen-Unregelmäßigkeitsindex (SII), den normalisierten medianen Grauwert der Plaque (GSM), CPRIlogistic und den CPRI.
Fig. 5 ROC curves showing the classification performance of the degree of stenosis (DOS), compared with the reduced version of the carotid plaque risk index DOS/(GSM + 1). Areas under ROC curves are 0.771 for DOS vs. 0.844 for the reduced index. Abb. 5 Die ROC-Kurven zeigen die Klassifizierungsleistung für den Grad der Stenose (DOS) im Vergleich zur reduzierten Version des Karotis-PlaqueRisikoindex DOS/(GSM + 1). Die Flächen unter der ROC-Kurve betragen 0,771 für DOS vs. 0,844 für den reduzierten Index.
colaides et al. constructed three risk models based on DOS, clinical and plaque features including GSM, plaque area, plaque type (qualitative assessment), and the presence of discrete white areas [27]. Multivariable, fractional polynomials, which were optimized for the dataset, were used to derive the risk indices. Kyriacou et al. followed a similar procedure but incorporated several plaque texture features to build a stroke risk model, with weighting factors optimized using their own dataset [28].
parameter
DOS
CPRIlogistic
CPRI
area under ROC curve (AUC)
0.771
0.845
0.849
95 % confidence interval for AUC
0.670 – 0.873
0.757 – 0.932
0.765 – 0.934
sensitivity (%)
73.8
85.7
90.5
specificity (%)
65.0
77.5
75.0
accuracy (%)
69.5
81.7
82.9
positive predictive value (%)
68.9
80.0
79.2
negative predictive value (%)
70.3
83.8
88.2
A common limitation of the existing methods has been the incorporation of weighting factors and other parameters that were optimized using the particular datasets studied. Ultrasound plaque characteristics such as the GSM, and even DOS, can be subject to variations when measured by different centers and methods [4]. Therefore, regardless of how large the sample size used might be, these variations compromise the general validity and applicability of these risk indices. Obtaining good classification accuracy need not mean the model is generally valid, if the model includes parameters that have been optimized for the dataset method of assessment used. The incorporation of qualitative measures into some risk models can also be considered a separate limitation since intra- and inter-observer agreement can be low for such subjective assessments [24, 29]. Another group of studies have attempted to classify plaques using methods such as pattern recognition, neural networks, support vector machines, and other machine learning algorithms but these do not provide risk indices and suffer from the same limitations [32 – 39]. Our study has found that a new, intuitive model that does not have any parameters which are optimized for our data has better diagnostic performance than DOS and a dataset-optimized risk index. The area under the ROC curve (0.849 for CPRI vs. 0.771 for DOS and 0.845 for CPRIlogistic) compared favorably with that of the more complicated, dataset-optimized risk index defined by Momjian-Mayor et al. [24]. This model has the additional benefit that it may be employed by different centers without re-optimization of the weighting factors. Our intuitive method of risk index construction was based on the observation that the presence of symptoms appears to be directly related to DOS and SII, while being inversely related to GSM [4, 5]. In our study, the carotid plaque risk index was defined as (DOS*SII)/(GSM + 1), but our investigation also suggested that a reduced risk index of the form of DOS/(GSM + 1) also performs better than DOS (AUC 0.771 for DOS vs. 0.844 for the reduced index). The reduced index can be particularly useful at centers where quantitative assessment of plaque surface irregularities is not available. Normalized plaque GSM can be easily measured using widely available software packages such as Photoshop (Adobe Systems, San Jose, California, USA) [16] and the reduced index can be used to obtain an enhanced predictor of symptoms compared to DOS at a busy clinical service with relative ease. The reduced index has the added convenience that for a stenosis severity of 50 % (in percentage format) and a moderate plaque GSM of 50, an index of approximately 1.0 is obtained, which increases with increasing stenosis severity and decreasing plaque GSM, and vice versa.
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Table 2 Comparison of diagnostic performance between the degree of stenosis (DOS), the logistic regression-based, dataset-optimized risk index (CPRI logistic) and our risk index (CPRI).
It should be noted that the plaque grayscale median (GSM), the surface irregularity index (SII), and the risk indices described in this paper are not related to the particular imaging platform we have used (i. e. Philips iu22) or the use of compound imaging (i. e. SonoCT). In other words, they can be used on any imaging platform as they are not related to or based on any company products or patents. In relation to the plaque grayscale median and the surface irregularity index, we calculated these using previously described methods which give GSM [4] and SII [5] averaged over multiple ultrasound images, while the suggestion of the Photoshop (Adobe Systems, San Jose, California, USA) [16] software gives those readers, who do not have access to specialized software, means to measure the GSM on a single ultrasound image. SII, which measures the degree of irregularity of the plaque surface, is a surrogate marker of plaque ulceration and other plaque surface defects which are discussed in length in the appropriate publication [5]. Further work is required to demonstrate the clinical benefit of these novel risk indices. This can be in the form of longitudinal studies looking prospectively at the development of cerebrovascular symptoms, or in a cross-sectional comparison against plaque histology or other measures of plaque vulnerability.
Conclusion !
Our study defined a novel carotid plaque risk index, an intuitive method of combining quantitative measures of plaque echogenicity, surface irregularities and DOS without the use of weighting factors optimized for the dataset. It was found that this risk index performs better than both DOS and a risk index which does include such optimized parameters. The clinical value of this risk index should be investigated in further studies.
Acknowledgments !
This research was funded by and took place at the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care based at the University Hospitals of Leicester NHS Trust. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
References 01 European Carotid Surgery Trialists' Collaborative Group. MRC European Carotid Surgery Trial: interim results for symptomatic patients with severe (70-99%) or with mild (0-29%) carotid stenosis. Lancet 1991; 337: 1235 – 1243 02 North American Symptomatic Carotid Endarterectomy Trial Collaborators. Beneficial effect of carotid endarterectomy in symptomatic patients with high-grade carotid stenosis. The New England Journal of Medicine 1991; 325: 445 – 453 03 Naylor AR. Time to rethink management strategies in asymptomatic carotid artery disease. Nature Reviews Cardiology 2012; 9: 116 – 124 04 Kanber B, Hartshorne TC, Horsfield MA et al. Dynamic variations in the ultrasound greyscale median of carotid artery plaques. Cardiovascular Ultrasound 2013; 11: 21 05 Kanber B, Hartshorne TC, Horsfield MA et al. Quantitative assessment of carotid plaque surface irregularities and correlation to cerebrovascular symptoms. Cardiovascular Ultrasound 2013; 11: 38 06 Aburahma AF, Kyer PD, Robinson PA et al. The correlation of ultrasonic carotid plaque morphology and carotid plaque hemorrhage: clinical implications. Surgery 1998; 124: n721 – n 726
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
07 Aburahma AF, Covelli MA, Robinson PA et al. The role of carotid duplex ultrasound in evaluating plaque morphology: potential use in selecting patients for carotid stenting. Journal of Endovascular Surgery 1999; 6: 59 – 65 08 Kessler C, Von Maravic M, Brückmann H et al. Ultrasound for the assessment of the embolic risk of carotid plaques. Acta Neurologica Scandinavica 1995; 92: 231 – 234 09 Pedro LM, Pedro MM, Gonçalves I et al. Computer-assisted carotid plaque analysis: characteristics of plaques associated with cerebrovascular symptoms and cerebral infarction. European Journal of Vascular and Endovascular Surgery 2000; 19: 118 – 123 10 Carra G, Visonà A, Bonanome A et al. Carotid plaque morphology and cerebrovascular events. International Angiology 2003; 22: 284 – 289 11 Ding S, Zhang M, Zhao Y et al. The role of carotid plaque vulnerability and inflammation in the pathogenesis of acute ischemic stroke. American Journal of the Medical Sciences 2008; 336: 27 – 31 12 Golledge J, Cuming R, Ellis M et al. Carotid plaque characteristics and presenting symptom. British Journal of Surgery 1998; 84: 1697 – 1701 13 Steinke W, Hennerici M, Rautenberg W et al. Symptomatic and asymptomatic high-grade carotid stenoses in Doppler color-flow imaging. Neurology 1992; 42: 131 – 138 14 Prabhakaran S, Rundek T, Ramas R et al. Carotid plaque surface irregularity predicts ischemic stroke: the northern Manhattan study. Stroke 2006; 37: 2696 – 2701 15 Biasi GM, Sampaolo A, Mingazzini P et al. Computer analysis of ultrasonic plaque echolucency in identifying high risk carotid bifurcation lesions. European Journal of Vascular and Endovascular Surgery 1999; 17: 476 – 479 16 Elatrozy T, Nicolaides A, Tegos T et al. The objective characterisation of ultrasonic carotid plaque features. European Journal of Vascular and Endovascular Surgery 1998; 16: 223 – 230 17 Grønholdt ML, Nordestgaard BG, Schroeder TV et al. Ultrasonic echolucent carotid plaques predict future strokes. Circulation 2001; 104: 68 – 73 18 Polak JF, Shemanski L, Leary DH et al. Hypoechoic plaque at US of the carotid artery: an independent risk factor for incident stroke in adults aged 65 years or older. Cardiovascular Health Study. Radiology 1998; 208: 649 – 654 19 Tegos TJ, Stavropoulos P, Sabetai MM et al. Determinants of carotid plaque instability: echoicity versus heterogeneity. European Journal of Vascular and Endovascular Surgery 2001; 22: 22 – 30 20 Salem MK, Sayers RD, Bown MJ et al. Patients with recurrent ischaemic events from carotid artery disease have a large lipid core and low GSM. European Journal of Vascular and Endovascular Surgery 2012; 43: 147 – 153 21 Ruiz-ares G, Fuentes B, Martínez-sánchez P et al. Utility of the assessment of echogenicity in the identification of symptomatic carotid artery atheroma plaques in ischemic stroke patients. Cerebrovascular Diseases 2012; 32: 535 – 541 22 Hoogi A, Akkus Z, Van Oord D et al. Quantitative analysis of ultrasound contrast flow behavior in carotid plaque neovasculature. Ultrasound in Medicine & Biology 2013; 38: 2072 – 2083 23 Prati P, Tosetto A, Casaroli M et al. Carotid plaque morphology improves stroke risk prediction: usefulness of a new ultrasonographic score. Cerebrovascular Diseases 2011; 31: 300 – 304 24 Momjian-mayor I, Kuzmanovic I, Momjian S et al. Accuracy of a novel risk index combining degree of stenosis of the carotid artery and plaque surface echogenicity. Stroke 2012; 43: 1260 – 1265 25 Pedro LM, Fernandes J, Pedro MM et al. Ultrasonographic risk score of carotid plaques. European Journal of Vascular and Endovascular Surgery 2002; 24: 492 – 498 26 Afonso D, Seabra J, Suri JS et al. A CAD system for atherosclerotic plaque assessment. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. 2013; 2012: 1008 – 1011 27 Nicolaides AN, Kakkos SK, Kyriacou E et al. Asymptomatic internal carotid artery stenosis and cerebrovascular risk stratification. Journal of Vascular Surgery 2010; 52: 1486 – 1496 28 Kyriacou E, Nicolaides A, Pattichis CS et al. First and second order statistical texture features in carotid plaque image analysis: preliminary results from ongoing research. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. 2012; 2011: 6655 – 6658
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Original Article
Original Article
35
36
37
38
39
automated classification: a paradigm for stroke risk assessment. Medical & Biological Engineering & Computing 2013; 51: 513 – 523 Acharya UR, Sree SV, Krishnan MMR et al. Atherosclerotic risk stratification strategy for carotid arteries using texture-based features. Ultrasound in Medicine & Biology 2012; 38: 899 – 915 Tsiaparas NN, Golemati S, Andreadis I et al. Comparison of multiresolution features for texture classification of carotid atherosclerosis from B-mode ultrasound. IEEE Transactions on Information Technology in Biomedicine 2011; 15: 130 – 137 Mougiakakou SGR, Golemati S, Gousias I et al. Computer-aided diagnosis of carotid atherosclerosis based on ultrasound image statistics, laws' texture and neural networks. Ultrasound in Medicine & Biology 2006; 33: 26 – 36 Christodoulou CI, Pattichis CS, Pantziaris M et al. Texture-based classification of atherosclerotic carotid plaques. IEEE Transactions on Medical Imaging 2003; 22: 902 – 912 Stoitsis J, Golemati S, Nikita KS et al. Characterization of carotid atherosclerosis based on motion and texture features and clustering using fuzzy c-means. Proceedings of the IEEE Engineering in Medicine and Biology Society 2007; 2: 1407 – 1410
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
29 Mayor I, Momjian S, Lalive P et al. Carotid plaque: comparison between visual and grey-scale median analysis. Ultrasound in Medicine & Biology 2003; 29: 961 – 966 30 Grant EG, Benson CB, Moneta GL et al. Carotid artery stenosis: grayscale and Doppler US diagnosis–Society of Radiologists in Ultrasound Consensus Conference. Radiology 2003; 229: 340 – 346 31 Oates CP, Naylor AR, Hartshorne T et al. Joint recommendations for reporting carotid ultrasound investigations in the United Kingdom. European Journal of Vascular and Endovascular Surgery 2008; 37: 251 – 261 32 Piliouras N, Kalatzis I, Theocharakis P et al. Development of the probabilistic neural networkraphic characterization and identification of symptomatic carotid plaques. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society 2011; 2010: 6110 – 6113 33 Kyriacou EC, Petroudi S, Pattichis CS et al. Prediction of high-risk asymptomatic carotid plaques based on ultrasonic image features. IEEE Transactions on Information Technology in Biomedicine 2013; 16: 966 – 973 34 Acharya UR, Mookiah MRK, Vinitha Sree S et al. Atherosclerotic plaque tissue characterization in 2D ultrasound longitudinal carotid scans for
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
A Novel Ultrasound-Based Carotid Plaque Risk Index Associated with the Presence of Cerebrovascular Symptoms Ein neuer ultraschallbasierter Karotis-Plaque-Risikoindex ist assoziiert mit dem Auftreten zerebrovaskulärer Symptome Authors
B. Kanber1, T. C. Hartshorne2, M. A. Horsfield1, A. R. Naylor2, T. G. Robinson1, 3, K. V. Ramnarine4
Affiliations
1
3 4
Department of Cardiovascular Sciences, University of Leicester, Leicester, UK Department of Vascular and Endovascular Surgery, University Hospitals of Leicester NHS Trust, Leicester, UK NIHR Biomedical Research Unit for Cardiovascular Sciences, University of Leicester, Leicester, UK Department of Medical Physics, University Hospitals of Leicester NHS Trust, Leicester, UK
Key words
Abstract
Zusammenfassung
"
!
!
Purpose: The purpose of this study was to determine the efficacy of a novel ultrasound-based carotid plaque risk index (CPRI) in predicting the presence of cerebrovascular symptoms in patients with carotid artery stenosis. Materials and Methods: This was a cross-sectional, observational study involving 56 patients (mean age 76.6 years, 62.5 % male). Plaque grayscale median (GSM) and surface irregularity indices (SII) were measured in 82 stenosed carotid arteries (range 10 – 95 %) and combined with the degree of stenosis (DOS) in the form of (DOS*SII)/(1 + GSM). A reduced index DOS/(1 + GSM) not incorporating plaque surface irregularities was also investigated. Receiver operating characteristic curves (ROC) were used to study the diagnostic efficacy of CPRI, comparing against DOS and an equivalent risk index constructed using a conventional logistic regression based method with model parameters optimized to the dataset (CPRIlogistic). Results: There were 42 stenosed carotid arteries with cerebrovascular symptoms, and 40 without symptoms. The presence of symptoms significantly correlated with DOS, GSM and SII (p < 0.01). The median CPRI of the symptomatic (asymptomatic) groups were 23.2 (9.2) compared with 0.71 (0.30) for CPRIlogistic (p < 0.01). The diagnostic performance of CPRI exceeded that of CPRIlogistic and DOS, and demonstrated a better separation of the symptomatic and asymptomatic groups. Conclusion: Our novel risk index combines quantitative measures of carotid plaque echogenicity and surface irregularities with the degree of stenosis. It is a better predictor of cerebrovascular symptoms than the degree of stenosis and could be valuable in studies and clinical trials aimed at identifying vulnerable carotid artery stenoses.
Ziel: Ziel der Studie war die Bestimmung der Leistungsfähigkeit des neuen ultraschallbasierten Karotis-Plaque-Risikoindex (CPRI) zur Vorhersage von zerebrovaskulären Symptomen bei Patienten mit Karotisstenose. Material und Methoden: Eine Querschnitts-Beobachtungsstudie in 56 Patienten (∅ 76,6 Jahre, 62,5 % ♂). Medianer Grauwert (GSM) der Plaque und Oberflächenunregelmäßigkeits-Indizes (SII) wurden in 82 stenosierten Karotisarterien (Bereich 10 – 95 %) bestimmt und mit dem Stenosegrad (DOS) mittels der Formel (DOS*SII)/(1 + GSM) verbunden. Ein reduzierter Index DOS/(1 + GSM), ohne Berücksichtigung von SII wurde untersucht. „Receiver Operating Characteristic“ (ROC)-Kurven wurden angewandt, um die diagnostische Effizienz des CPRI mit DOS und einem gleichwertigen Risikoindex zu vergleichen, letzterer erstellt mittels einer auf konventionelle logistische Regression basierenden Methode mit für den Datensatz optimierten Modellparametern (CPRIlogistic). Ergebnisse: 42 Stenosen mit zerebrovaskulären Symptomen und 40 symptomlose Stenosen lagen vor. Das Auftreten von Symptomen korrelierte signifikante mit DOS, GSM und SII (p < 0,01). Der Median des CPRI in den symtomatischen (asymtomatischen) Gruppen betrug 23,2 % (9,2) im Vergleich zu 0,71 (0,30) des CPRIlogistic der symptomatischen (asymtomatischen) Gruppen (p < 0,01). In diagnostischer Leistung war CPRI der CPRIlogistic und dem DOS überlegen und zeigte eine bessere Trennung zwischen symptomatischen und asymptomatischen Gruppen. Schlussfolgerung: Unser neuer Risikoindex vereinigt die quantitative Messung der KarotisplaqueEchogenität und der Unregelmäßigkeiten der Oberfläche mit dem Grad der Stenose. Es ist ein besserer Prädiktor für zerebrovaskuläre Symptome als der DOS und könnte sich in Untersuchungen und klinischen Studien, die eine Identifizierung
● carotid plaque ● risk index ● CPRI " "
received accepted
15.3.2014 14.9.2014
Bibliography DOI http://dx.doi.org/ 10.1055/s-0034-1385462 Published online: 2014 Ultraschall in Med © Georg Thieme Verlag KG Stuttgart · New York · ISSN 0172-4614 Correspondence Dr. Baris Kanber Department of Cardiovascular Sciences, University of Leicester, Leicester, United Kingdom Tel.: ++ 44/1 16/2 58 51 40 Fax: ++ 44/1 16/2 58 60 70 [email protected]
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
2
Introduction
Analysis
!
!
The degree of stenosis (DOS) of the carotid artery is an established parameter that is routinely used in clinical practice [1, 2]. However, it is recognized that some plaques may be more vulnerable than others, and their identification would aid clinical decision making [3]. Previous investigations have shown that ultrasound plaque characteristics such as the plaque grayscale median (GSM) and plaque surface irregularities may be useful for identifying such vulnerable plaques [4 – 21]. Neovascularization and intra-plaque hemorrhage are other potentially important markers of plaque vulnerability, but they are more readily assessed using contrastenhanced ultrasound (CEUS) [22]. Plaque compression, deformation, intra-plaque/fibrous cap strains, and shear forces are all additional, potentially important markers of plaque vulnerability. Several studies have attempted to combine ultrasound-plaque measures with the degree of stenosis in order to derive a plaque risk index but either qualitative measures have been employed and/or the relevant risk index was based on cut-off values or weighting factors optimized to a particular dataset [23 – 28]. The incorporation of dataset-optimized factors limits applicability and validity, since the assessment of ultrasound plaque characteristics, including that of the degree of stenosis, can vary depending on the method of assessment and can be subject to variation between centers [4, 29]. We have developed an ultrasound-based carotid plaque risk index (CPRI) that is based on quantitative measures of plaque echogenicity and surface irregularities, combining these with the degree of stenosis without incorporating any parameters optimized for our dataset. Our study compares the performance of this risk index with that of DOS and a conventional logistic regressionbased model with dataset-optimized weighting factors.
Quantitative analyses were carried out using MATLAB version 7.14 (MathWorks, Natick, Massachusetts, USA) and SPSS version 20 (IBM Corporation, Armonk, New York, USA). The degree of carotid artery stenosis was measured by experienced vascular scientists using criteria consistent with the NASCET method, utilizing blood flow velocities in conjunction with B-mode and color flow imaging findings [2, 30, 31]. The measurements were carried out not using a single longitudinal imaging plane but multiple imaging planes both in the longitudinal and transverse cross-section. However, as Doppler velocity measurements are not able to reliably discriminate degrees of stenosis below 50 %, B-mode diameter measurements and color flow imaging were used to grade the degree of stenosis into deciles for minor stenoses. The readers are directed to the appropriate references for a detailed consensus description of the measurement of the degree of carotid artery stenosis using Duplex ultrasound methods [2, 30, 31]. In relation to grayscale plaque characteristics, previously described methods were employed to evaluate the normalized plaque GSM [4] and the plaque surface irregularity index (SII) [5]. SII was calculated by computationally summing the angular deviations from a straight line, of the luminal plaque surface, and dividing this by the physical length of the plaque surface, and is described in detail elsewhere [5]. The luminal plaque surface is the boundary separating the plaque body from the bloodstream and " Fig. 1. is where the green and purple dashed lines overlap in ● Since SII is an angle divided by a length, it is measured in units of radians (or degrees) per meter (or centimeter). SII has been shown to be a reproducible measurement with a mean intra-observer coefficient of vari7ation (CV) of 4.4 %, and a mean intra-observer, inter-frame CV of 10.6 % [5]. GSM, on the other hand, is simply the median grayscale (echo) intensity within the plaque body [4]. Plaque bodies are the regions bounded by the green da" Fig. 1. For the normalized plaque GSM, a mean shed lines in ● intra-observer CV of 2.8 % was previously reported [4]. We averaged GSM and SII values over all ultrasound image frames acquired for each artery. On average, 180 consecutive frames were analyzed for each artery. The non-parametric Wilcoxon-MannWhitney test was used to assess whether quantitative measures differed significantly between symptomatic and asymptomatic groups, and ROC curves were used to assess classification performance. Two-tailed tests of significance were used and p-values less than 0.05 were considered statistically significant. A carotid plaque risk index based on logistic regression methods (CPRIlogistic) was constructed, similar to Momjian-Mayor et al. [24] in the form of 1/(1 + exp(-x)) where x is equal to β0 + β1*DOS + β2*GSM + β3*SII and β0, β1, β2, β3 are factors that are optimized for the data. Our CPRI, on the other hand, was defined as (DOS*SII)/(GSM + 1) with no dataset-optimized parameters. DOS values in this equation were between 0.0 and 1.0 (e. g. 0.5 for 50 % stenosis), while the SII were input in radians per meter. The value 1 was added to the GSM to avoid having a mathematical singularity for plaques with a GSM of 0. Box and whisker plots were used to study the distribution of CPRI and CPRIlogistic among the plaque groups with and without associated cerebrovascular symptoms.
Methods !
This was a cross-sectional study in which 56 patients (35 male, 21 female) who attended the University Hospitals of Leicester NHS Trust's Rapid Access Transient Ischemic Attack (TIA) clinic were recruited. Ultrasound image sequences of the carotid plaque in the longitudinal cross-section were acquired by experienced sonographers using a Philips iU22 ultrasound scanner (Philips Healthcare, Eindhoven, The Netherlands) and an L9 – 3 probe. B-mode (grayscale) and color Doppler image sequences were recorded as DICOM files over an average duration of 5.6 seconds (mean frame rate was 32 frames per second for the B-mode acquisitions) using the vascular carotid preset on the scanner (Vasc Car™ preset, persistence low, XRES™ and SONOCT™ on). Color Doppler image sequences were used as a qualitative aid to identifying the location and extent of the plaques, while the grayscale data were used for the quantitative analyses. The study was approved by the National Research Ethics Service (NRES) Committee East Midlands – Northampton (reference 11/EM/ 0249) and followed institutional guidelines. Each patient gave informed consent before participating in the study. 82 stenosed carotid arteries (stenosis range 10 – 95 %) were investigated. Plaques were classified as either having caused cerebrovascular symptoms relating to the ipsilateral brain hemisphere within the past six-month period (i. e. symptomatic) or as asymptomatic following the review of patient symptoms and clinical/radiological (CT/MRI) findings.
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Original Article
Fig. 1 Examples of luminal plaque surfaces (where the purple and green dashed lines overlap) and plaque bodies (bounded by the green dashed lines). Abb. 1 Beispiele für luminale Plaque-Oberflächen (wo sich die violetten und grünen gestrichelten Linien überlappen) und Plaquekörper (umgrenzt von den grünen gestrichelten Linien).
Results !
The mean age of the patients was 76.6 (range 58 to 95) years. Patient age was not significantly different between the asymptomatic (76.9 years) and symptomatic groups (76.4 years, p > 0.05). Patients who had previous TIA/stroke episodes (44.6 %) were more likely to be symptomatic (p = 0.01), while other patient characteristics did not have a significant association to the presence of " Table 1). symptoms (● There were 82 carotid arteries with stenoses ranging from 10 % to 95 %. The distribution of the stenoses was as follows: 49 with DOS less than 50 %, 24 with DOS between 50 %-70 %, and 9 with DOS greater than 70 %. 42 of these arteries were associated with symptoms relating to the ipsilateral brain hemisphere, while 40 were asymptomatic. The degree of stenosis and the SII were significantly higher in the arteries with symptoms, while the normalized plaque GSM was lower (p < 0.01 for all). The mean SII of symptomatic plaques was 1970 radians/m compared to 1780 radians/m for the asymptomatic ones. The corresponding normalized GSM values were 45.1 and 59.5, respectively. Patients for which image quality was not sufficiently good (for example complete shadowing) were excluded from the study, but there were only a small number of such cases (n = 1). 7 patients had atrial fibrillation present on pre-scan ECG. 3 of these patients were asymptomatic since they were not diagnosed to have suffered a cerebrovascular event. The remaining 4 patients had no brain infarcts observed on CT/MRI, but their body symptoms were contralateral to their stenosed carotid arteries. Examples of a plaque with ipsilateral hemispheric symptoms and " Fig. 2. The risk index a plaque without symptoms are shown in ● based on optimized logistic regression (CPRIlogistic) and our CPRI were significantly higher for stenoses that were associated with symptoms (p < 0.01 for both), but CPRI showed a better separa-
Fig. 2 Example of a symptomatic plaque (top) and an asymptomatic plaque (bottom) with respective risk indices (CPRI) of 27.5 and 3.25. The degree of stenosis (DOS) caused by the symptomatic plaque was 70 %, while it was 20 % for the asymptomatic plaque. The value contrast between the symptomatic and the asymptomatic case was better with CPRI than with DOS. The ratio between the two degrees of stenosis was 70/20 (3.5x), while the ratio between the two risk indices was 27.5/3.25 (8.5x). Abb. 2 Beispiel einer symptomatischen Plaque (oben) und einer asymptomatischen Plaque (unten) mit den entsprechenden Risiko-Indizes (CPRI) von 27,5 (symptomatisch) und 3,25 (asymptomatisch). Der durch die symptomatische Plaque verursachte Stenosegrad (DOS) betrug 70 %, der durch die asymptomatischen Plaque verursachte hingegen 20 %. Der Kontrastwert zwischen dem symptomatischen und asymptomatischen Fall war mit CPRI besser als mit DOS; die Ratio zwischen den beiden Stenosegraden betrug 70/20 (3,5x), während die Ratio zwischen den beiden Indizes 27,5/ 3,25 (8,5x) betrug.
" Fig. 3). ROC curve analysis showed that tion of the two groups (● CPRIlogistic and CPRI had similar classification performance, better than that of DOS, with CPRI providing a better overall accuracy " Fig. 4, " Table 2). The area under the ROC curve (AUC) was (● ● " Table 2). 0.849 for CPRI, 0.845 for CPRIlogistic and 0.771 for DOS (● The median CPRI of the symptomatic and asymptomatic groups were 23.2 and 9.2 units compared with 0.71 and 0.30 for CPRIlogistic. GSM and SII appeared to have similar ROC curves, which were " Fig. 4). A reduced risk index, not utilizing below that of DOS (● plaque surface irregularities, of the form DOS/(GSM + 1) also out" Fig. 5). performed DOS on its own (●
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Original Article
Original Article
patient characteristic
mean age (years) sex (males)
all patients
asymptomatic
symptomatic
significance
group
group
(p-value)
76.6
76.9
76.4
0.80
62.5 %
54.5 %
67.6 %
0.33
hypertension
64.3 %
63.6 %
64.7 %
0.94
hypercholesterolemia
53.6 %
54.5 %
52.9 %
0.91
ischemic heart disease
23.2 %
13.6 %
29.4 %
0.18
diabetes
35.7 %
22.7 %
44.1 %
0.11
previous TIA/stroke
44.6 %
22.7 %
58.8 %
0.01*
peripheral vascular disease
14.3 %
9.1 %
17.6 %
0.38
smoking
67.9 %
68.2 %
67.6 %
0.97
alcohol consumption
37.5 %
27.3 %
44.1 %
0.21
family history of stroke
33.9 %
22.7 %
41.2 %
0.16
atrial fibrillation
12.5 %
13.6 %
11.8 %
stroke (n = 8)
14.3 %
not applicable
23.5 %
not applicable
TIA (n = 16)
28.6 %
not applicable
47.1 %
not applicable
transient vision loss (n = 10)
17.9 %
not applicable
29.4 %
not applicable
Table 1 Patient characteristics and the significance of association with cerebrovascular symptoms.
0.84
Fig. 3 Box and whisker plots showing the distribution of CPRIlogistic (top left) and CPRI (bottom left) in carotid artery stenoses with and without cerebrovascular symptoms. The corresponding plots on the right further show the distribution of CPRIlogistic (top right) and CPRI (bottom right) within the two groups. Abb. 3 Die Box-Whisker-Plots zeigen die Verteilung von CPRIlogistic (oben links) und CPRI (unten links) in Karotisstenosen mit und ohne zerebrovaskuläre Symptome. Die entsprechenden Darstellungen rechts zeigen die Verteilung von CPRIlogistic (oben rechts) und CPRI (unten rechts) innerhalb der beiden Gruppen.
Discussion !
Our study described a novel risk index which intuitively combines quantitative measures of plaque echogenicity and surface irregularities with the degree of carotid artery stenosis. This risk index, despite not having any parameters optimizing it for our dataset, was found to have increased diagnostic performance compared to DOS, and a risk index based on logistic regression with such optimized parameters. Several studies have previously derived risk indices that combine ultrasound plaque characteristics with DOS in order to help identify vulnerable carotid plaques [23 – 28]. Prati et al. defined a risk index incorporating DOS, and qualitative measures of surface irregularities, plaque echogenicity and texture [23]. Momjian-
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Mayor et al. defined a risk index combining DOS with a quantitative measure of the plaque surface echogenicity and used a logistic regression-based statistical model with weighting factors optimized for their dataset [24]. Pedro et al. similarly developed an activity index, comprising several measures including that of plaque surface disruption (qualitative assessment), GSM, percentage of plaque area with gray levels less than 40, plaque heterogeneity (qualitative assessment), and the presence of juxta-luminal echolucent areas (qualitative assessment), in addition to DOS. Weighting scores for various criteria based on these parameters were evaluated and used to assign an activity index to each plaque [25, 26]. This was later extended by Seabra et al. to an enhanced activity index which incorporated additional parameters, including those based on the radiofrequency (RF) ultrasound signal. Ni-
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
The statistical methods used to test the associations were the non-parametric Wilcoxon-Mann-Whitney test for patient age and the χ2 test for the rest of the patient characteristics. Significant associations are marked with an asterisk (*).
Original Article
Fig. 4 ROC curves showing the classification performance of the degree of stenosis (DOS), plaque surface irregularity index (SII), the normalized plaque grayscale median (GSM), CPRIlogistic and CPRI. Abb. 4 Die ROC-Kurven zeigen die Klassifizierungsleistung für den Grad der Stenose (DOS), den Oberflächen-Unregelmäßigkeitsindex (SII), den normalisierten medianen Grauwert der Plaque (GSM), CPRIlogistic und den CPRI.
Fig. 5 ROC curves showing the classification performance of the degree of stenosis (DOS), compared with the reduced version of the carotid plaque risk index DOS/(GSM + 1). Areas under ROC curves are 0.771 for DOS vs. 0.844 for the reduced index. Abb. 5 Die ROC-Kurven zeigen die Klassifizierungsleistung für den Grad der Stenose (DOS) im Vergleich zur reduzierten Version des Karotis-PlaqueRisikoindex DOS/(GSM + 1). Die Flächen unter der ROC-Kurve betragen 0,771 für DOS vs. 0,844 für den reduzierten Index.
colaides et al. constructed three risk models based on DOS, clinical and plaque features including GSM, plaque area, plaque type (qualitative assessment), and the presence of discrete white areas [27]. Multivariable, fractional polynomials, which were optimized for the dataset, were used to derive the risk indices. Kyriacou et al. followed a similar procedure but incorporated several plaque texture features to build a stroke risk model, with weighting factors optimized using their own dataset [28].
parameter
DOS
CPRIlogistic
CPRI
area under ROC curve (AUC)
0.771
0.845
0.849
95 % confidence interval for AUC
0.670 – 0.873
0.757 – 0.932
0.765 – 0.934
sensitivity (%)
73.8
85.7
90.5
specificity (%)
65.0
77.5
75.0
accuracy (%)
69.5
81.7
82.9
positive predictive value (%)
68.9
80.0
79.2
negative predictive value (%)
70.3
83.8
88.2
A common limitation of the existing methods has been the incorporation of weighting factors and other parameters that were optimized using the particular datasets studied. Ultrasound plaque characteristics such as the GSM, and even DOS, can be subject to variations when measured by different centers and methods [4]. Therefore, regardless of how large the sample size used might be, these variations compromise the general validity and applicability of these risk indices. Obtaining good classification accuracy need not mean the model is generally valid, if the model includes parameters that have been optimized for the dataset method of assessment used. The incorporation of qualitative measures into some risk models can also be considered a separate limitation since intra- and inter-observer agreement can be low for such subjective assessments [24, 29]. Another group of studies have attempted to classify plaques using methods such as pattern recognition, neural networks, support vector machines, and other machine learning algorithms but these do not provide risk indices and suffer from the same limitations [32 – 39]. Our study has found that a new, intuitive model that does not have any parameters which are optimized for our data has better diagnostic performance than DOS and a dataset-optimized risk index. The area under the ROC curve (0.849 for CPRI vs. 0.771 for DOS and 0.845 for CPRIlogistic) compared favorably with that of the more complicated, dataset-optimized risk index defined by Momjian-Mayor et al. [24]. This model has the additional benefit that it may be employed by different centers without re-optimization of the weighting factors. Our intuitive method of risk index construction was based on the observation that the presence of symptoms appears to be directly related to DOS and SII, while being inversely related to GSM [4, 5]. In our study, the carotid plaque risk index was defined as (DOS*SII)/(GSM + 1), but our investigation also suggested that a reduced risk index of the form of DOS/(GSM + 1) also performs better than DOS (AUC 0.771 for DOS vs. 0.844 for the reduced index). The reduced index can be particularly useful at centers where quantitative assessment of plaque surface irregularities is not available. Normalized plaque GSM can be easily measured using widely available software packages such as Photoshop (Adobe Systems, San Jose, California, USA) [16] and the reduced index can be used to obtain an enhanced predictor of symptoms compared to DOS at a busy clinical service with relative ease. The reduced index has the added convenience that for a stenosis severity of 50 % (in percentage format) and a moderate plaque GSM of 50, an index of approximately 1.0 is obtained, which increases with increasing stenosis severity and decreasing plaque GSM, and vice versa.
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Table 2 Comparison of diagnostic performance between the degree of stenosis (DOS), the logistic regression-based, dataset-optimized risk index (CPRI logistic) and our risk index (CPRI).
It should be noted that the plaque grayscale median (GSM), the surface irregularity index (SII), and the risk indices described in this paper are not related to the particular imaging platform we have used (i. e. Philips iu22) or the use of compound imaging (i. e. SonoCT). In other words, they can be used on any imaging platform as they are not related to or based on any company products or patents. In relation to the plaque grayscale median and the surface irregularity index, we calculated these using previously described methods which give GSM [4] and SII [5] averaged over multiple ultrasound images, while the suggestion of the Photoshop (Adobe Systems, San Jose, California, USA) [16] software gives those readers, who do not have access to specialized software, means to measure the GSM on a single ultrasound image. SII, which measures the degree of irregularity of the plaque surface, is a surrogate marker of plaque ulceration and other plaque surface defects which are discussed in length in the appropriate publication [5]. Further work is required to demonstrate the clinical benefit of these novel risk indices. This can be in the form of longitudinal studies looking prospectively at the development of cerebrovascular symptoms, or in a cross-sectional comparison against plaque histology or other measures of plaque vulnerability.
Conclusion !
Our study defined a novel carotid plaque risk index, an intuitive method of combining quantitative measures of plaque echogenicity, surface irregularities and DOS without the use of weighting factors optimized for the dataset. It was found that this risk index performs better than both DOS and a risk index which does include such optimized parameters. The clinical value of this risk index should be investigated in further studies.
Acknowledgments !
This research was funded by and took place at the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care based at the University Hospitals of Leicester NHS Trust. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
References 01 European Carotid Surgery Trialists' Collaborative Group. MRC European Carotid Surgery Trial: interim results for symptomatic patients with severe (70-99%) or with mild (0-29%) carotid stenosis. Lancet 1991; 337: 1235 – 1243 02 North American Symptomatic Carotid Endarterectomy Trial Collaborators. Beneficial effect of carotid endarterectomy in symptomatic patients with high-grade carotid stenosis. The New England Journal of Medicine 1991; 325: 445 – 453 03 Naylor AR. Time to rethink management strategies in asymptomatic carotid artery disease. Nature Reviews Cardiology 2012; 9: 116 – 124 04 Kanber B, Hartshorne TC, Horsfield MA et al. Dynamic variations in the ultrasound greyscale median of carotid artery plaques. Cardiovascular Ultrasound 2013; 11: 21 05 Kanber B, Hartshorne TC, Horsfield MA et al. Quantitative assessment of carotid plaque surface irregularities and correlation to cerebrovascular symptoms. Cardiovascular Ultrasound 2013; 11: 38 06 Aburahma AF, Kyer PD, Robinson PA et al. The correlation of ultrasonic carotid plaque morphology and carotid plaque hemorrhage: clinical implications. Surgery 1998; 124: n721 – n 726
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
07 Aburahma AF, Covelli MA, Robinson PA et al. The role of carotid duplex ultrasound in evaluating plaque morphology: potential use in selecting patients for carotid stenting. Journal of Endovascular Surgery 1999; 6: 59 – 65 08 Kessler C, Von Maravic M, Brückmann H et al. Ultrasound for the assessment of the embolic risk of carotid plaques. Acta Neurologica Scandinavica 1995; 92: 231 – 234 09 Pedro LM, Pedro MM, Gonçalves I et al. Computer-assisted carotid plaque analysis: characteristics of plaques associated with cerebrovascular symptoms and cerebral infarction. European Journal of Vascular and Endovascular Surgery 2000; 19: 118 – 123 10 Carra G, Visonà A, Bonanome A et al. Carotid plaque morphology and cerebrovascular events. International Angiology 2003; 22: 284 – 289 11 Ding S, Zhang M, Zhao Y et al. The role of carotid plaque vulnerability and inflammation in the pathogenesis of acute ischemic stroke. American Journal of the Medical Sciences 2008; 336: 27 – 31 12 Golledge J, Cuming R, Ellis M et al. Carotid plaque characteristics and presenting symptom. British Journal of Surgery 1998; 84: 1697 – 1701 13 Steinke W, Hennerici M, Rautenberg W et al. Symptomatic and asymptomatic high-grade carotid stenoses in Doppler color-flow imaging. Neurology 1992; 42: 131 – 138 14 Prabhakaran S, Rundek T, Ramas R et al. Carotid plaque surface irregularity predicts ischemic stroke: the northern Manhattan study. Stroke 2006; 37: 2696 – 2701 15 Biasi GM, Sampaolo A, Mingazzini P et al. Computer analysis of ultrasonic plaque echolucency in identifying high risk carotid bifurcation lesions. European Journal of Vascular and Endovascular Surgery 1999; 17: 476 – 479 16 Elatrozy T, Nicolaides A, Tegos T et al. The objective characterisation of ultrasonic carotid plaque features. European Journal of Vascular and Endovascular Surgery 1998; 16: 223 – 230 17 Grønholdt ML, Nordestgaard BG, Schroeder TV et al. Ultrasonic echolucent carotid plaques predict future strokes. Circulation 2001; 104: 68 – 73 18 Polak JF, Shemanski L, Leary DH et al. Hypoechoic plaque at US of the carotid artery: an independent risk factor for incident stroke in adults aged 65 years or older. Cardiovascular Health Study. Radiology 1998; 208: 649 – 654 19 Tegos TJ, Stavropoulos P, Sabetai MM et al. Determinants of carotid plaque instability: echoicity versus heterogeneity. European Journal of Vascular and Endovascular Surgery 2001; 22: 22 – 30 20 Salem MK, Sayers RD, Bown MJ et al. Patients with recurrent ischaemic events from carotid artery disease have a large lipid core and low GSM. European Journal of Vascular and Endovascular Surgery 2012; 43: 147 – 153 21 Ruiz-ares G, Fuentes B, Martínez-sánchez P et al. Utility of the assessment of echogenicity in the identification of symptomatic carotid artery atheroma plaques in ischemic stroke patients. Cerebrovascular Diseases 2012; 32: 535 – 541 22 Hoogi A, Akkus Z, Van Oord D et al. Quantitative analysis of ultrasound contrast flow behavior in carotid plaque neovasculature. Ultrasound in Medicine & Biology 2013; 38: 2072 – 2083 23 Prati P, Tosetto A, Casaroli M et al. Carotid plaque morphology improves stroke risk prediction: usefulness of a new ultrasonographic score. Cerebrovascular Diseases 2011; 31: 300 – 304 24 Momjian-mayor I, Kuzmanovic I, Momjian S et al. Accuracy of a novel risk index combining degree of stenosis of the carotid artery and plaque surface echogenicity. Stroke 2012; 43: 1260 – 1265 25 Pedro LM, Fernandes J, Pedro MM et al. Ultrasonographic risk score of carotid plaques. European Journal of Vascular and Endovascular Surgery 2002; 24: 492 – 498 26 Afonso D, Seabra J, Suri JS et al. A CAD system for atherosclerotic plaque assessment. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. 2013; 2012: 1008 – 1011 27 Nicolaides AN, Kakkos SK, Kyriacou E et al. Asymptomatic internal carotid artery stenosis and cerebrovascular risk stratification. Journal of Vascular Surgery 2010; 52: 1486 – 1496 28 Kyriacou E, Nicolaides A, Pattichis CS et al. First and second order statistical texture features in carotid plaque image analysis: preliminary results from ongoing research. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. 2012; 2011: 6655 – 6658
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
Original Article
Original Article
35
36
37
38
39
automated classification: a paradigm for stroke risk assessment. Medical & Biological Engineering & Computing 2013; 51: 513 – 523 Acharya UR, Sree SV, Krishnan MMR et al. Atherosclerotic risk stratification strategy for carotid arteries using texture-based features. Ultrasound in Medicine & Biology 2012; 38: 899 – 915 Tsiaparas NN, Golemati S, Andreadis I et al. Comparison of multiresolution features for texture classification of carotid atherosclerosis from B-mode ultrasound. IEEE Transactions on Information Technology in Biomedicine 2011; 15: 130 – 137 Mougiakakou SGR, Golemati S, Gousias I et al. Computer-aided diagnosis of carotid atherosclerosis based on ultrasound image statistics, laws' texture and neural networks. Ultrasound in Medicine & Biology 2006; 33: 26 – 36 Christodoulou CI, Pattichis CS, Pantziaris M et al. Texture-based classification of atherosclerotic carotid plaques. IEEE Transactions on Medical Imaging 2003; 22: 902 – 912 Stoitsis J, Golemati S, Nikita KS et al. Characterization of carotid atherosclerosis based on motion and texture features and clustering using fuzzy c-means. Proceedings of the IEEE Engineering in Medicine and Biology Society 2007; 2: 1407 – 1410
Downloaded by: WEST VIRGINIA UNIVERSITY. Copyrighted material.
29 Mayor I, Momjian S, Lalive P et al. Carotid plaque: comparison between visual and grey-scale median analysis. Ultrasound in Medicine & Biology 2003; 29: 961 – 966 30 Grant EG, Benson CB, Moneta GL et al. Carotid artery stenosis: grayscale and Doppler US diagnosis–Society of Radiologists in Ultrasound Consensus Conference. Radiology 2003; 229: 340 – 346 31 Oates CP, Naylor AR, Hartshorne T et al. Joint recommendations for reporting carotid ultrasound investigations in the United Kingdom. European Journal of Vascular and Endovascular Surgery 2008; 37: 251 – 261 32 Piliouras N, Kalatzis I, Theocharakis P et al. Development of the probabilistic neural networkraphic characterization and identification of symptomatic carotid plaques. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society 2011; 2010: 6110 – 6113 33 Kyriacou EC, Petroudi S, Pattichis CS et al. Prediction of high-risk asymptomatic carotid plaques based on ultrasonic image features. IEEE Transactions on Information Technology in Biomedicine 2013; 16: 966 – 973 34 Acharya UR, Mookiah MRK, Vinitha Sree S et al. Atherosclerotic plaque tissue characterization in 2D ultrasound longitudinal carotid scans for
Kanber B et al. A Novel Ultrasound-Based … Ultraschall in Med
