Carotid geometry is an independent predictor of wall thickness – a 3D cardiovascular magnetic resonance study in patients with high cardiovascular risk

From April 2018 to February 2019, all patients from our in- and outpatient clinic undergoing carotid ultrasound were consecutively and prospectively screened for eligibility. Inclusion criteria were: ≥50 years of age, arterial hypertension and at least one additional risk factor (smoking, diabetes mellitus, hyperlipidemia, peripheral arterial disease (PAD), coronary artery disease (CAD), history of ischemic stroke or transient ischemic attack (TIA)) and evidence of at least one 1.5 mm thick plaque of the ICA or the distal common carotid artery (CCA) in ultrasound.

Exclusion criteria were: contraindications to 3 T CMR such as ferromagnetic implants, claustrophobia, poor clinical condition (modified ranking scale (mRS) > 3), atrial fibrillation or other relevant cardiac arrhythmias interfering with the electrocardiographic (ECG)-trigger), ICA-stenosis > 50% or ICA-occlusion according to North American Symptomatic Carotid Endarterectomy Trial (NASCET) criteria [18], expectation of life < 2 years, pregnancy, distance to place of residence > 100 km and refusal of study participation.

Six hundred and thirty-one patients fulfilled the inclusion criteria of which 494 were excluded (Fig. 1) leaving 121 patients for analysis. Patients’ demographics and cardiovascular disease risk factors were obtained from patients’ electronic charts and personal interviews before CMR study.

Written informed consent was obtained from all participants. The local ethics committee approved the study and all ultrasound and CMR procedures were in accordance with institutional guidelines.

All patients underwent 2D Duplex sonography of extracranial and intracranial arteries (IU22, Philips Healthcare, Bothwell, Washington, USA). A linear array probe (L 9–3 MHz, Philips Healthcare) was used for extracranial and a curved array probe (S 5–1 MHz, Philips Healthcare) for intracranial examination. Ultrasound was performed by two medical technical assistants with experience in neurovascular ultrasound > 20 years or by a physician with several years of ultrasound experience. Patients were examined in a supine position and maximum wall thickness and plaque thickness were measured manually by electronic calipers at the far wall of the left and right CCA and ICA. ICA stenosis was graded according to NASCET criteria based on 2D Duplex sonography [18].

Multi-contrast CMR included 3D time of flight (TOF), T1-, T2-, and proton density (PD)-weighted imaging. For T1-, T2- and PD-weighted imaging a variable-flip-angle 3D turbo spin echo (TSE) sequence (Sampling Perfection with Application optimized Contrasts using different flip angle Evolution–SPACE) with fat saturation and dark-blood preparation was used. Imaging was executed on a 3 T CMR scanner (Prisma, Siemens Healthineers, Erlangen, Germany) with an 8-channel surface coil (NORAS MRI products GmbH, Hoechberg, Germany).

An isotropic spatial resolution of 0.6mm³ was used for all sequences except for 3D-TOF (0.5 × 0.5 × 0.6 mm). Other sequence parameters were: 3D-TOF: repetition time (TR)/echo time (TE) = 21/3.3 ms, field of view (FOV) = 180 × 135 mm², matrix 384 × 290, flip angle = 20°, pixel bandwidth (BW) = 250 Hz/Px, total acquisition time (TA) = 4:32 min. T1-weighted SPACE: TR/TE = 900/26 ms, BW = 405 Hz/Px, echo train length (ETL) = 36, TA = 7:32 min. T2-weighted SPACE: TR/TE = 2000/159 ms, BW = 405 Hz/Px, ETL = 61, TA = 7:22 min. PD-weighted SPACE: TR/TE = 1900/26 ms, BW = 405 Hz/Px, ETL = 61, TA = 8:04 min.

4D flow data (spatial/temporal resolution = 0.8 mm³/52.8 ms) were acquired with prospective ECG-triggering using a k-t-accelerated time-resolved 3D phase contrast sequence [19, 20]. Other sequence parameters were: TR/TE = 52.8/3.9 ms, flip angle = 12°, peak GRAPPA acceleration factor = 5, FOV = 140x140mm², slice thickness = 0.8 mm, BW = 460 Hz/Px, VENC (in-plane) = 0.6 m/s, VENC (through plane) = 1.0 m/s).

Blood pressure levels of the upper right arm were recorded before and after CMR examination after resting in a supine position for at least 5 min. Heart rate was documented every 4 min during 4D flow CMR.

For image processing data sets were imported into a custom-made extension (CaroTo) of the MEVISFlow research software (Fraunhofer MEVIS, Bremen, Germany [21]). Preprocessing of the 4D flow data included noise filtering, correction for Eddy currents and velocity aliasing. Then, a carotid artery mask (3D-phase-contrast CMR (PC-CMR)) was calculated from the 4D flow CMR data. In the next step, a centerline was created in each carotid bifurcation and landmarks (flow diverter and ICA) were marked. From these landmarks predefined points on the centerline were automatically generated, indicating the level of each analysis plane, which was initialized automatically in the next step. If necessary, analysis planes were manually oriented perpendicular to the carotid lumen. For quantitative image analysis we used a plane- and segment-based model as described in a previous study [8]. This was performed in 8 cross-section planes (Fig. 2a). The first plane was positioned on the CCA centerline 1 cm below the flow diverter. The second plane was placed within the ICA bulb, planes 3 to 6 along the ICA with the starting point at the flow diverter and oriented perpendicularly to the centerline with a spacing of each 3 mm. Plane 7 was the most distal ICA plane outside the plaque and positioned manually by the user. Plane 8 was placed automatically in the proximal external carotid artery (ECA). As described previously [8] each analysis plane was divided into 12 wall segments. Segment 1 was located at the posterior bulb (Fig. 2b). The remaining segments were numbered clockwise for the left and counterclockwise for the right carotid arteries. Plaques and stenoses are typically located in the distal CCA and proximal ICA at the posterior wall segments [7]. These were represented by segments 1–3; 11 and 12 in planes 2–6 (Fig. 2b), defined as atherosclerosis-prone region and used to evaluate the distribution of WSS parameters and of wall thickness. The required processing time of one 3D CMR dataset, including the analysis of bifurcation geometry, hemodynamics and wall thickness was 45–60 min.

Geometry of all 242 bifurcations was analyzed automatically based on the 3D-TOF CMR data after manually defining CCA, ICA, and ECA as start/endpoints for centerline computation and based on the location of the flow diverter (FD) as illustrated in Fig. 3. Geometric parameters were described and successfully used previously [8, 10]. They included a) ICA/CCA-diameter ratio, which was calculated using maximum ICA diameter in plane 6 and CCA diameter in plane 1, respectively; b) bifurcation angle, which was measured by two tangential lines of the first 1 cm of the outer wall starting at the FD; c) CCA-ICA tortuosity, which was assessed by calculating the ratio of the direct line and the centerline connection of the CCA in plane 1 and plane 6 in the ICA.

For each carotid bifurcation, the vessel-lumen boundary was manually outlined in the magnitude images of the 4D flow CMR data and propagated for each timeframe. Absolute WSS (in N/m² = 1 Pa) was time-averaged over the cardiac cycle and derived for each vessel segment and for each plane. Systolic WSS was derived by averaging the time frames 2–5 in systole, which presumably contained peak systole. OSI (in %) was calculated as the degree of WSS inversion over the entire cardiac cycle as described previously [8].

Post-processing of vessel wall thickness was based on 3D-T1-weighed CMR using the same custom-made software (CaroTo, Fraunhofer MEVIS, Bremen, Germany) comprising centerline computation after manual initialization and marking flow diverter and ICA. Quantitative assessment of vessel lumen and wall thickness was carried out in the same eight analysis planes and the 12 segments, which were positioned in the same fashion as described for WSS calculation. Wall thickness was measured after manual delineation of the inner and outer contours of the vessel wall (illustration of workflow in Fig. 4).

In order to describe the distribution of wall thickness and hemodynamic parameters (absolute/systolic WSS and OSI) mean plots by side and stenosis are presented of the means in anterior and posterior segments. For each of these parameters a corresponding linear mixed model was performed with the respective parameter as dependent variable and a patient’s identity (ID) as random effect. Independent variables were anterior segment (yes/no), right side (yes/no), stenosis (yes/no, side specific), and planes 2–6 (see Fig. 2). In the following, the model corresponding to the primary outcome parameter “wall thickness” is called base model.

In order to investigate the impact of geometric (ICA/CCA-diameter ratio, bifurcation angle, tortuosity) and hemodynamic parameters (absolute/systolic WSS, OSI) on wall thickness, the base model was expanded with the respective parameters as independent variables. The base model was expanded to the final model by adding geometric and haemodynamic parameters while adjusting for risk factors (male gender, age, body mass index (BMI), diabetes mellitus, PAD, smoking, history of stroke or TIA, CAD, hypercholesteremia, hemoglobin A1c (Hb_A1c) and low density lipoprotein (LDL)-cholesterol). Multiple measures in each patient were taken into account with the mixed effect in the model.

Table 1 summarizes the prevalence of cardiovascular disease risk factors and the number of patients with ≥10% ICA stenosis.

All three geometric parameters were higher in the left compared to the right carotid bifurcation. In stenosis ≥10%, bifurcation angle and tortuosity were higher, the diameter ratio, however, was smaller compared to vessels without stenosis (Table 2).

Mean absolute WSS and systolic WSS were significantly lower in the posterior than in the anterior wall of the bulb (planes 2–6, see Fig. 2) and lower in the right than in the left bifurcation (p < 0.001). Lowest values were found in the proximal bulb (=plane 2) with a slight increase downstream (Fig. 5). Values of absolute and systolic WSS were significantly higher in ≥10% ICA stenoses (p < 0.001), but still lower in the posterior wall compared to the anterior wall, especially in the right bifurcation (Additional file 1).

Inversely, OSI was higher in the posterior bulb (p < 0.001) and decreased significantly (p < 0.001) when stenoses were present. Highest values were found on the left side and in the proximal ICA bulb (=plane 2) (Fig. 5 and Additional file 1).

Figure 6 exemplarily illustrates hemodynamic differences in a patient with an ICA plaque and in another with a 40% ICA stenosis, demonstrating distinct differences in 3D blood flow (i.e., blood flow acceleration and elimination of the helical flow pattern) and WSS distribution depending on the presence or absence of the bulb.

Wall thickness in the carotid bulb (planes 2–6) was significantly higher in posterior versus anterior wall segments, left versus right side, and in ≥10% stenoses versus ICA without stenoses (p < 0.001). Highest values were measured in the proximal bulb (=plane 2). Wall thickness decreased downstream of the ICA (Fig. 5 and Additional file 1).

The ICA/CCA-diameter ratio was significantly and independently but inversely associated with vessel wall thickness (regression coefficient = − 0.302; p < 0.001). Accordingly, patients with thinner vessel wall had larger proximal ICA-diameters compared to those with thicker walls. In addition, vessel tortuosity was a significant and independent predictor of wall thickness in the carotid bulb (regression coefficient = 0.764, p < 0.001), even when adjusting for cardiovascular risk factors. However, we found no association of the bifurcation angle with wall thickness (regression coefficient = 0.000; p = 0.931). We neither identified an independent effect of absolute WSS, systolic WSS nor OSI on wall thickness when adjusting for risk factors.

Only male sex, but no other cardiovascular risk factor was independently related to increased wall thickness.

Previous studies [7, 8, 10, 22] demonstrated the concentration of low absolute WSS or high OSI at the posterior wall of the ICA bulb in normal arteries, suggesting a causal connection of WSS and atherosclerosis, as this is the predilection site of carotid atheroma [7, 23, 24]. Consistently, we observed the same distribution in high-risk patients with carotid plaques. However, this pattern disappeared in ≥10% ICA stenosis due to the conversion of the ICA bulb into a straight or even stenotic segment by the plaque. Similarly, in case series of moderate to high-grade ICA stenosis, the typical WSS and OSI distribution disappeared and was restored after carotid surgery, respectively [8, 25]. Thus, our results support the hypothesis that low WSS and high OSI could influence the initiation of atherosclerosis, whereas in manifest stenosis other mechanisms, such as high maximum shear or mechanical stress may trigger plaque progression [16, 24, 26‐28]. This concept is underlined by previous studies [13, 14] showing an association of high maximum shear stress with intra-plaque hemorrhage and conversion into a vulnerable plaque. Normal values or cut-off values for WSS or OSI predicting the individual risk of plaque development, progression or rupture are not yet available. In the past, arbitrary cut-offs representing 10% or 20% of the lowest or highest values in the respective study population were used. They differed according to study population, measurement accuracy (spatial/temporal resolution) and methods (CFD versus 4D flow CMR) [8, 10]. In the present study, we analyzed the distribution and local differences of absolute and mean values, which allow a direct comparison with future similar studies.

A large 2D-ultrasound study revealed that a dorsal/dorsomedial ICA outlet increased the likelihood of increased intima-media thickness of the ICA bulb [4], but was not able to assess geometry or wall thickness on a 3D level. This was later overcome by CTA [15], showing that decreased ICA-CCA diameter ratio and increased ICA angle were independently related to the degree of ICA stenosis. Finally, a 2D-CMR study [16] demonstrated that increased diameter of the bulb region and increased curvature, which is comparable to tortuosity [10], were independent but weak predictors of early wall thickening in the CCA and proximal ICA. Similar to Phan et al. [15] we observed an independent inverse relationship of the ICA/CCA diameter ratio and were able to correlate it directly to local wall thickness. This suggests that the ICA/CCA diameter ratio, thought to be one of the most influential predictors of disturbed flow [8, 10, 15, 16], loses its influence in patients with advanced atherosclerotic disease because of inward remodeling. Comparable to Bijari et al. [16], tortuosity was an independent predictor of local wall thickness in our study. The influence of tortuosity on the development of femoral artery atherosclerosis was also shown in 232 hyperlipidemic patients [29]. Surprisingly, increasing bifurcation angle, which is partly considered in the parameter tortuosity, representing changes of blood flow induced by geometry more adequately, was not an independent predictor for wall thickness, although having been correlated with WSS distribution and degree of ICA stenosis [8, 10, 15].

Healthy subject studies [8, 10] detected a concentration of atherogenic shear parameters in the ICA bulb, but only very limited data is available from studies [17, 30‐32] investigating their influence on carotid wall thickness: in a study with 14 patients, phase-contrast CMR and CFD demonstrated that reduced WSS was inversely related to increased carotid wall thickness. However, influence of bifurcation geometry was not considered [17]. A 2D-ultrasound study of 48 patients showed that low WSS was an independent predictor of plaque progression over a 12-year period [33]. In addition, animal studies [11, 12, 34] impressively demonstrated that WSS parameters correlated with wall thickening, that low WSS led to the development of unstable plaques and high OSI formed stable plaques. However, animal models, as well as CFD studies, are not directly comparable to studies performed in humans, because of artificial induction of atherosclerotic lesions by a special diet in animal models and differences in vessel compliance and blood viscosity in CFD models.

In our study, we evaluated flow conditions and wall thickness under naturalistic conditions in 242 carotid bifurcations and directly compared WSS and OSI with wall thickness along the entire bifurcation. However, there was no independent association of WSS and OSI with carotid wall thickness. Since approximately 35% of patients had carotid stenoses ≥10%, we speculate that these hemodynamic parameters played an important role in early stages of atherosclerosis, whereas at the time of CMR measurement other factors may trigger progression of wall thickness and plaque size. Thus, following the literature and our results, the independent role of WSS or OSI for the formation of local atherosclerosis is still equivocal and requires further investigation in large-scale studies.

Our protocol intra−/inter-observer agreement and reproducibility for WSS and OSI were good. In addition, measurement of carotid wall thickness employing a similar CMR protocol in patients [35, 36] was good to excellent for these items.

The comprehensive acquisition of 3D data of morphology and blood flow in one single CMR scan was realized with a reasonable spatial and temporal resolution. Post-processing software analysis automatically placed centerlines and analysis planes in a standardized manner in relation to anatomically defined markers. In addition, geometrical information was calculated automatically in contrast to previous in vivo studies [8]. This allowed for a minimization of measurement errors, an optimal comparison of geometry and WSS with wall thickness values and thus a comprehensive study of fluid-structure interaction in vivo.

We recruited patients with ≥2 cardiovascular disease risk factors and atherosclerotic plaques, including ≥10–50% ICA stenosis, allowing us to study different stages of atherosclerosis, but also resulting in a somewhat heterogeneous cohort, which may be one reason why WSS was not an independent predictor of local wall thickening. Alternatively, the measurements of WSS and OSI were performed at a too advanced stage of atherosclerosis. Also, a larger cohorts or follow-up measurements may be needed to detect the independent influence of WSS or OSI. Furthermore, based on our cross-sectional data, we are not able to state if individual geometry or WSS are predictors of future increase of local wall thickness, plaque growth, and rupture.

We used carotid wall thickness as dependent variable in our study. Plaque volume or plaque composition are other interesting parameters of carotid atherosclerosis that should be studied in their interaction with geometry and shear stress in the ICA bulb in a cross-sectional analysis and over time. This would be of high clinical interest as it is currently unclear if they provide additional or superior information compared to wall thickness. Unfortunately, a semi-automated quantification of plaque volume was not yet available but is currently under development and will be tested by our group in the near future.

Apart from the above-mentioned patient population-related effects, accuracy of blood flow and wall thickness measurements strongly depend on spatial and temporal resolution of the CMR data and require optimal image quality. Although our CMR protocol allowed consistent and standardized analysis, further improvements in resolution and reduction of wall motion through respiratory-, swallowing-, or ECG-gating may be beneficial and the duration of the examination may be reduced by new acceleration methods such as compressed sensing [37]. Processing of one dataset including the determination of geometry, hemodynamics and wall thickness required 45–60 min. Thus, our methods are well suited for research in smaller patient cohorts to study the underlying pathophysiology of atherosclerosis e.g. at the carotid bifurcation as done in the current study. However, due to the high effort of time our methods are not yet suited for the application in large-scale studies or clinical routine. Thus, further automatization and reduction of processing time to a few minutes are necessary.