A Noninvasive Imaging Approach to Assess Plaque Severity: The Carotid Atherosclerosis Score

Study Sample

Multicontrast carotid MR images from 435 individuals with 16%–99% stenosis by duplex sonography in at least 1 carotid artery were pooled from 4 institutions: 1) AH, Capital Medical University, Beijing, China (n = 78; stenosis, >50%); 2) MSU, East Lansing, Michigan (n = 90; stenosis, >50%); 3) PLA, Beijing, China (n = 67; stenosis, >50%); and 4) UW, Seattle, Washington (n = 80; stenosis,16%–49%; n= 120; stenosis, 50%–79%). The study procedures and consent forms were reviewed and approved by the institutional review board at each institution before study initiation. The artery with the greatest stenosis by duplex sonography, termed the index artery, was selected for image review because image acquisition was centered on that artery. Arteries were excluded from review if there was the following: 1) prior carotid endarterectomy on the index carotid artery, 2) prior radiation therapy to the neck, 3) insufficient coverage (<10 mm including the bifurcation) by MR imaging, or 4) poor image quality by MR imaging. Clinical information was obtained through chart review.

MR Imaging Acquisition

Participants underwent carotid MR imaging on either a 1.5T or 3T scanner by using bilateral phased-array carotid surface coils. Previously published multicontrast carotid imaging protocols^14,15 were adapted for each study center. Sequences and imaging parameters are detailed in On-line Table 1. Notably, at 2 sites (AH and MSU) participants also underwent CE-MRA (On-line Table 1).

Image Analysis

All images of the index carotid artery from each subject were interpreted at a core laboratory by teams of 2 reviewers, all with >1.5 years' experience in vascular MR imaging, via consensus opinion and blinded to clinical information. At each axial location, image quality was assessed with a 4-point scale (1 = poor, 2 = adequate, 3 = good, 4 = excellent). For arteries with image quality ≥2, image analysis software¹⁶ was used to draw the lumen and outer wall boundaries at each axial location. Lumen area, wall area, total vessel area, maximum wall thickness, and NWI (wall area/total vessel area) were recorded. In addition, the presence or absence of calcification, LRNC, IPH (Fig 1, top row), and fibrous cap status was determined by using multicontrast imaging criteria that have been previously validated with histology^6,17–19 and collectively depicted in a recently published atlas of carotid MR imaging and histology.²⁰

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

In vivo identification of IPH, FCR, and ulceration. Each row contains multicontrast axial images from a single location in the carotid artery. The outer wall boundary (white arrowheads) of either the internal or common carotid artery, the lumen of the internal or common (asterisk) carotid artery, the external (cross, where applicable) carotid artery, and the JV are identified. The top row is of the right internal carotid artery from a 69-year-old man imaged at MSU. IPH (arrows) is characterized by a hyperintense signal intensity on TOF and T1WI. The second row is of the right internal carotid artery from an 80-year-old man imaged at PLA. An FCR (white arrow) is evident on TOF imaging by the hyperintense signal intensity extending from the lumen into the plaque and the absence of a fibrous cap on T2WI and CE-T1WI. The third row is of the left common carotid artery from a 74-year-old man imaged at PLA. An ulceration (black arrows) is present. Notably, the varying appearance of the ulcer between the different contrast weightings is due to reduced flow suppression caused by turbulent flow in the ulcerated region, particularly after contrast administration.

Signal-intensity criteria for identifying calcification, LRNC, and IPH are summarized in Table 1. Fibrous cap status was identified as 1) intact thick if fibrous tissue was evident on CE-T1WI or T2WI between the LRNC and the lumen, 2) intact thin if fibrous tissue was not evident on CE-T1WI or T2WI between the LRNC and lumen, and 3) ruptured if there was an absence of a fibrous cap on CE-T1WI and T2WI and juxtaluminal IPH on TOF (Fig 1, middle row). Area measurements of the LRNC and calcification, when present, were also collected, and the proportion of each component relative to the wall area (eg, percentage LRNC area = 100% × LRNC area / wall area) was subsequently calculated for each MR imaging location. Proportional measurements of plaque composition normalized the data for arterial size. Percentage stenosis was determined from the CE-MRA by using the established North American Symptomatic Carotid Endarterectomy Trial criterion²¹: 100% × (1 − luminal diameter at the point of maximal narrowing / the diameter of the normal distal internal carotid artery).

Statistical Analysis

All statistical analyses were performed with SPSS for Windows (Version 12.0, SPSS, Chicago, Illinois). Due to differences among institutions in longitudinal coverage of the artery, the maximum arterial value for each continuous metric was used during data analysis (as opposed to mean arterial values or volume data), except for lumen area, in which case the minimum arterial value was used. Individuals with ulceration, defined as a surface disruption with invagination into the plaque, were excluded from analysis (Fig 1, bottom row). Ulceration alters the morphology and composition of the plaque, which obscures the conditions that existed before the development of ulceration. To account for differences among imaging protocols, field-strengths, and patient demographics at different sites, we randomly divided the remaining evaluable study sample into 2 datasets with an intended ratio of 60:40 (training set/test set) by using the random selection of data command in SPSS. This method of creating a training and test set through random subselection creates 2 datasets that are equivalent to 2 independent random draws from the source collection of patients.²² Although measures of reproducibility for carotid MR imaging parameters have been previously reported (intra- and inter-reader^6,19; interscan^15,23,24), partitioning of the dataset in this manner, after image interpretation had been complete for all arteries, enabled the unbiased evaluation of reproducibility of the results obtained from the training set on the test set. To verify the randomness of allocation between the training and test sets, we evaluated differences in the baseline demographic and arterial characteristics with the independent t test for continuous variables and the Fisher exact test for categoric variables. Statistical significance, based on 2-sided tests, was defined as P < .05.

The CAS was constructed via the following steps: Step 1, with the training set, univariate binary logistic regression for continuous variables or the Fisher exact test for categoric variables was used to identify potential predictors of IPH and FCR from each of the clinical and arterial parameters listed in Table 2. IPH and FCR were not considered in the univariate analysis because they were the target, dependent variables used as indicators of a high-risk lesion. Step 2, with the training set, we created a multivariate model separately for IPH and FCR by using backward elimination of predictor variables, starting from all parameters with an association (P < .10) identified during univariate analysis and by using P < .10 (F test) to retain a variable in the model. OR and the 95% CI are reported for results from univariate and multivariate analysis. Step 3, by using the training set, we assessed the accuracy of classification for each statistically significant parameter selected in the multivariate analysis with ROC curves and the AUC. Step 4, parameters identified during multivariate analysis of the training set were used to construct the CAS. Notably, methods used to construct the CAS were dependent on the results and are, thus, described in the “Results” section. Step 5, after formulation of the CAS from the training set was complete, performance of CAS was evaluated with the test set. ROC curves and AUC were used to determine the strength of the classification for identifying lesions with IPH or, separately, FCR in the test set.

Prevalence of calcification, LRNC, IPH, and fibrous cap status (thick, thin, or ruptured) are reported for each level of CAS separately for the training and the test sets. The prevalence of IPH, FCR, and measures used to construct the CAS were computed at each imaging location (relative to the bifurcation) for the entire dataset (training + test). Finally, in all participants with a CE-MRA, strength of classification for determining the presence of IPH and FCR was determined for both MRA stenosis and CAS.