Computed tomography angiography vs 3 T black-blood cardiovascular magnetic resonance for identification of symptomatic carotid plaques

Patients

Data acquisition

CT angiography

As stroke patient assessment is performed by several units in our clinical center, scans were performed on various CT scanners: Bright Speed S (GE Healthcare), Aquilion (Toshiba), Somatom Definition Flash (Siemens), Somatom Definitition AS + (Siemens), Sensation 64 (Siemens). All MDCTA images were obtained with coverage at least from the aortic arch to the cranium using the respective standard protocol parameters, notably a collimation of 0.625 mm or less. Non-ionic iodinated contrast material was applied intravenously adjusted to patient weight (Ultravist 370, Bayer Schering Pharma, Berlin, Germany, i.e. 0.35-0.50 g iodine per kg bodyweight at an injection rate of 4.5-6 ml/s followed by 100 ml saline at identical flow). Axial images were reconstructed to volume rendered (VR) images, which has been reported to improve depiction of carotid plaque ulceration if used additionally to axial scans [23]. Bone and soft tissue impairing the 360° view of the bifurcation were manually removed from the images. In order to prevent biases in the further reviewing process, anonymized images of the right and left carotid arteries were stored separately.

Image review

All images were reviewed by two radiologists with 4 and 11 years of experience blinded to the patient's clinical history. Classification was reached in consensus. Images of all patients' left carotid arteries were reviewed first, followed by all images of the right side, in random order.

Image quality of each examination was rated on a five-point scale (1 = very poor, 2 = poor, 3 = acceptable, 4 = good, 5 = excellent) and cases with image quality <3 were excluded. All following analyses were performed on axial images.

On CMR images area-measurements of lumen, wall, and components were obtained using the image analysis tool CASCADE (University of Washington, Seattle, US). T1w images were used for obtaining values for lumen, vessel wall and total vessel area. The normalized wall index (NWI) was calculated by dividing the wall area by the total vessel area. Atherosclerotic tissue components (lipid-rich necrotic core, hemorrhage, fibrous tissue, calcification) were identified and quantified based on previously published criteria [24]. For definition of a complicated AHA-LT6 plaque according to the modified AHA classification [8], at least one of the following three criteria was required: ruptured fibrous cap, intraplaque hemorrhage or juxtaluminal hemorrhage / thrombus.

For MDCTA window settings were adjusted depending on image properties to optimize the delineation of plaque properties [19]. The relative content of calcification in the stenotic plaque (calcification / tissue <40% vs. >40%) was visually quantified and categorized. As a high grade of calcification can cause beam hardening artifacts and thus spoil measurements in adjacent tissue [25], tissue density was not measured in plaques containing calcifications >40%. MDCTA images were evaluated for presence of plaque, plaque density / plaque type, surface configuration, calcification and thrombus. Plaque type was categorized as non-calcified, mixed, and calcified. Calcification volume in mm³ was determined using the standard plug-in on a Syngo MultiModality Workplace (software version VE36A, Siemens Healthcare, Erlangen, Germany).

Tissue attenuation was measured in representative non-calcified plaque regions using several manually drawn ellipsoid regions of interest (ROI) per axial slice encompassing at least approximately 75% of the plaque area. Depending on plaque size, 4 to 16 ROIs with an area of up to 2 mm² were drawn. During this procedure calcifications and the contrast enhanced lumen were carefully avoided. The average HU of the measured ROIs on each axial slice was recorded. From this, a mean representative HU-value for the whole plaque was calculated. Based on the mean plaque HU, plaque type was categorized as follows: non-calcified plaque associated with lipid-rich necrotic core <60 HU, mixed plaque associated with fibrous tissue 60-130 HU and calcified plaque >130 HU [26]. Plaques with calcification >40% were not measured and rated as calcified plaques. Outpouching of contrast material into or adjacent to the plaque of < 1 was considered a surface irregularity, whereas ≥1 mm was considered ulceration [27].

Data analysis

Area measurements for each artery are given as average, minimum or maximum absolute areas, as appropriate. Plaque components are calculated as percentages of the vessel wall. Categorical variables are presented as absolute frequencies, while continuous variables are presented as mean ± [SD]. Wilcoxon's signed-rank test was used to test differences for continuous variablesthe McNemar test was used to determine differences between categorical variables. Kruskal-Wallis test was used if more than two categories were present, e.g. degree of stenosis and of calcification. Statistical evaluation was performed using SPSS version 16.0 (IBM, Armonk, USA). A p-value of <0.05 was considered statistically significant. Sensitivity, specificity positive and negative predictive values as well as odds ratios were calculated using a 2-by-2 table.

Patients

CMR

Figures1, 2, and 3 show black-blood CMR and MDCTA imaging examples of symptomatic (Figures1 and 2) and asymptomatic (Figure3) carotid plaques.

Computed tomography

Surface

CT showed 8 vessels with ulceration on the symptomatic side (40%) and one (5%) in an asymptomatic internal carotid artery (p =0.005). Multiple ulcerations within one vessel were not observed. In one case we found ulcers in both carotid arteries. The proportion of ulcerations in all arteries was not significantly different between 50-69% stenoses (3 out of 1127%) compared to stenoses of 70-99% (6 out of 1932%p =1).

Tissue density and calcification

Symptomatic plaques had a significantly lower density compared to asymptomatic plaques (48.9 ±15.6 HU vs. 61.8 ±15.6 HUp =0.046). Kruskal-Wallis and McNemar tests revealed no significant differences between symptomatic and asymptomatic plaques regarding presence of calcification. Calcification volume was not significantly different between symptomatic and asymptomatic plaques.

Thrombus

Using MDCTA we could observe the `donut sign' - a filling defect within the lumen completely surrounded by contrast media indicating presence of a thrombus - in only one of the 20 symptomatic carotid arteries (5%) versus none in the asymptomatic arteries (0%).

Best predictors

AHA-LT6 as single predictor for the symptomatic side resulted in a sensitivity, specificity, PPV and a NPV of 80%, 80%, 80% and 80%, respectively, yielding an accuracy of 80% and an odds ratio of 16.0 (95% confidence interval: 2.8-108.9). With CMR 4 out of 20 symptomatic plaques were not predicted correctly. In one of these 4 patients, MDCTA was able to detect an ulcerated plaque.

When a thin or ruptured fibrous cap was used as single predictor, sensitivity, specificity, PPV and NPV were 60%, 95%, 92% and 70%, respectively. The accuracy was calculated at 77.5%the odds ratio was 28.5 (95% confidence interval 3.2-257.5).

Using ulceration in MDCTA as single predictor for the symptomatic side led to a sensitivity, specificity, PPV and a NPV of 40%, 95%, 89% and 61%, accuracy of 67.5% and an odds ratio of 12.7 (95% confidence interval: 1.3-306).

Ulcer as detected in CT combined with AHA-LT6 Plaque in CMR led to a sensitivity of 85%, specificity of 75%, a PPV of 77% and a NPV of 83%, yielding an accuracy of 80% and an odds ratio of 17.0 (95% confidence interval: 2.8-121.2Table 4).

The ROC graph in Figure4 gives an overview over the presented variables and their respective sensitivity and specificity.

AHA-LT6

Complicated AHA-LT6 plaques were more frequently encountered in symptomatic than in asymptomatic arteries (80% vs. 20%, p < 0.001). Consequently type III, IV, V and VII, which reflect clinically more stable plaques, were more common in asymptomatic vessels. These results are in line with findings of Saam et al. who found a statistically significant association between complicated AHA-LT6 carotid plaques and ipsilateral symptoms in a cohort of 23 symptomatic patients who underwent CMR-plaque imaging [13]. This makes AHA-LT6 a useful parameter for risk assessment in atherosclerotic carotid plaques. In our study, as a single criterion, the presence of an AHA-LT6 plaque proved to be the best overall predictor for the symptomatic side.

We noted a greater frequency of AHA-LT6 plaques in high degree stenoses compared to low degree stenoses. Even though not statistically significant, this result is in line with findings of previous studies [28],[29], which report an increasing prevalence of vulnerable plaques with increasing degree of stenosis.

Ulceration

Former studies could show that ulceration of the carotid arteries is associated with cerebral lesions and is encountered more frequently in symptomatic patients [17],[30]-[32]. The present study found ulcerations with a prevalence of 40% in symptomatic carotid arteries and in 5% of the asymptomatic arteries. Overall prevalence of ulceration was 32% in high grade stenoses (70-99%), 27% in low grade stenoses (50-69%) and 0% in non-stenosed arteries. The reported prevalence of plaque ulceration in our study is similar to the NASCET study, which reported ulcerations in 35% of vessels with a stenosis >70% as detected by DSA [33].

In our study, using ulceration in MDCTA as the sole predictor for the symptomatic side led to a relatively low sensitivity but a high specificity. The lower sensitivity may be explained by the fact that other mechanisms that are hard to detect in MDCTA, such as rupture of the fibrous cap or intraplaque hemorrhage potentially followed by thrombus formation and consequent embolisation may also cause cerebrovascular incidents.

Other factors

While other plaque properties analyzed in this study failed to reach the predictive power of AHA-LT6 or plaque ulceration, they may still be useful for plaque characterization and general risk assessment.

Analysis of the plaque density showed that plaques on the symptomatic side had significantly lower HU values than asymptomatic plaques. This may be due to higher lipid content and confirms the findings of other studies by Saba et al. [30] and Serfaty et al. [25] which both found a statistically significant association between fatty plaque or decreased plaque density and neurological symptoms in cohorts of 112 and 141 patients. Trelles et al. similarly found a greater thickness of the non-calcified plaque component in AHA-LT6 plaques in a study on 51 stroke patients [34]. These results are also in line with previous observations in coronary plaques which suggested that plaques with a large lipid-rich/necrotic core strongly correlate with formation of thrombi and plaque disruption [35]. Despite these differences in HU, a reliable cut-off point to distinguish symptomatic from asymptomatic plaques has not yet been determined [19].

Juxtaluminal hemorrhage / thrombus impose a high risk of distal embolization and thus thrombotic occlusion of a distal vessel. In a previous histological study with 241 patients Fisher et al. observed that thrombus is highly associated with ipsilateral symptoms and plaque ulceration (p < 0.005) [33]. Using MDCTA we could identify a fresh thrombus ('donut sign') in only one carotid artery (2.5%). CMR, however, showed juxtaluminal hemorrhage / thrombus in 30% of the symptomatic and 5% of the asymptomatic arteries. Our study thus shows that for detection of juxtaluminal hemorrhage / thrombus CMR is more sensitive than MDCTA, a fact that may be attributed to MDCTA's comparatively poor soft tissue contrast and its inability to identify the fibrous cap and intraplaque hemorrhage. In 6 out of 7 cases thrombi detected by CMR were associated with a rupture of the fibrous cap. This supports the thesis that thrombi evolve from surface defects of plaques presenting thrombogenic components.

In our study a thin or ruptured fibrous cap itself was closely associated with the symptomatic side (OR 28.5). This may be attributed to thrombogenic potency of surface defects of the fibrous cap with subsequent formation of unstable thrombi, potentially causing embolism. Our results indicate that a thin or ruptured fibrous cap may be a more accurate predictor than juxtaluminal hemorrhage/thrombus. Compared to AHA-LT6, a thin or ruptured fibrous cap as single predictor showed superior specificity and positive predictive value, while sensitivity and negative predictive value were inferior. Because of its high specificity and positive predictive value a thin or ruptured fibrous cap may be a factor to be considered separately.

Bb-CMR or MDCTA?

Based on the data of our study bb-CMR is better suited than MDCTA to identify symptomatic carotid plaques. This is mainly owed to the better detection of subtle plaque features on bb-CMR. In order to further assess the potential of bb-CMR, several prospective studies, such as the CAPIAS trial (NCT01284933) [36] or the PARISK study (NCT01208025) are currently under way which examine the ability of carotid bb-CMR to predict recurrence of ischemic stroke in patients with less than 70% stenosis.

While the diagnostic value of bb-CMR regarding the characterization of carotid plaques cannot reasonably be disputed, it is still not routinely applied in many centers. This may be attributed to a generally lower availability of suitable CMR equipment and longer scan times of bb-CMR compared to MDCTA, requiring a higher degree of patient cooperation. Another limiting factor for carotid CMR in the acute setting is the narrow time window for treating acute strokes, so examination times need to be kept reasonably short. It needs to be noted that the use of a dedicated carotid coil, which is not available at every center, enhances MR image quality. However, with careful design of the examination and its sequence parameters, image quality should be sufficient to determine the major plaque features even without such a dedicated coil.

On the other hand, MDCTA requires application of a considerable amount of ionizing radiation and, more importantly, iodinated contrast agent with the known adverse effects, especially in the patient population with cardiovascular risk factors. Logistically, bb-CMR of the carotids can be performed immediately following brain CMR with reasonable extension of the examination time, which will at least in part be compensated by avoiding a transfer of the patient between the CT and CMR device.

Limitations

One limitation of this study is the use of different CT scanners, so that examination parameters are not completely identical. We chose to include examinations from various CT scanners to increase the number of patients and accounted for possible differences in image acquisition parameters by applying a uniform image quality score to all images.

Furthermore, our study followed a cross sectional design by examining patients who were already symptomatic. To assess the predictive value of different plaque characteristics regarding the risk of future cerebrovascular events, longitudinal studies are necessary. It should be noted that our study population contains a comparatively large number of vessels with high degree stenoses. This is owed to the inclusion criteria requiring 50% stenosis in the artery ipsilateral to the ischemic stroke and does not represent the prevalence of high degree stenosis in the general population. It needs to be noted that we performed all tests on subjects of a symptomatic population where by definition the prevalence of symptomatic carotid plaques is 100%. This may influence the absolute validity of sensitivity, specificity, positive and negative predictive values for the reported plaque features at predicting the symptomatic side. It should, however, not significantly impair the comparability of the different predictors with each other.

3D sequences were not part of this study. While these might have increased the arterial coverage or decreased examination times, we chose to not apply them in this study because they are less well validated and may provide less spatial in-plane resolution. Furthermore, coverage of the 2D sequences used was sufficient to depict the entire plaque in each case so that it seemed safe to use the previously established 2D protocol, which has proven to deliver excellent image quality while being robust and reliable.

We did not use contrast enhanced MR angiography (CEMRA) because we performed perfusion imaging of the plaques, which was not evaluated for this study. Evaluation of CEMRA might have increased the sensitivity of CMR to detect surface irregularities in carotid plaques. However, several studies have shown that ulcerations are an unreliable predictor of patient's symptoms and we therefore believe that while CEMRA might have further accentuated the superiority of CMR, this would not have significantly changed our conclusion.