Fusion of fibrous cap thickness and wall shear stress to assess plaque vulnerability in coronary arteries: a pilot study

Data were gathered at the Thoraxcenter, Erasmus MC (Rotterdam, The Netherlands). Fourteen patients (mean age \(61.0\,\pm \,10.3\) years old, 7 males), suffering from coronary artery disease and referred for possible percutaneous coronary intervention, were involved in the study. The most representative and largest plaque was selected in each pullback. The only inclusion criteria were the presence of a fibrous plaque (nine patients) or a fibrocalcific plaque (five patients) in the acquired pullbacks. The image selection was reviewed by a clinician expert in OCT imaging. Ten left anterior descending arteries, one left main artery, and three right coronary arteries were simultaneously imaged with angiography and OCT. Informed consent was acquired from the patients for use of their imaging data. All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2008 (5).

Angiography recordings were acquired at a frame rate of 15 frames/s with the Axiom Artis system (Siemens, Forchheim, Germany). The two recording angles were selected to avoid overlap of branches other than the selected coronary artery. Table movement in between the two recordings was avoided as well. The ECG signal was registered during the recordings. The angiography images were acquired at \(1024 \times 1024\) pixels, with a pixel size of \(76\,\upmu \text {m}\).

OCT imaging was realized with three different scanners, to assess the applicability of the framework. Pullbacks were acquired using one of the following apparatus: (1) Ilumien frequency domain imaging system with the Dragonfly Duo intracoronary imaging catheter (Lightlab/St Jude, Minneapolis, MN, USA) (nine patients), (2) Lunawave system with the Fastview imaging catheter (Terumo Corporation, Tokyo, Japan) (two patients), or (3) MGH prototype optical frequency domain imaging system (Massachusetts General Hospital, Boston, MA, USA), and the Terumo Fastview catheter (three patients). Image acquisition was performed with a previously described non-occlusive technique [33]. For the Lightlab system, pullbacks were acquired over a total length of either 54 mm (six patients) or 75 mm (three patients) along the vessel, at either 105 or 180 frames/s, with an inter-frame distance of either 200 or 100 \(\upmu \)m. For the Terumo system, pullbacks were acquired over a total length of 47 mm, at 160 frames/s, with an inter-frame distance of 125 \(\upmu \)m. For the MGH system, pullbacks were acquired over a total length of 61, 81, and 121 mm along the vessel, at 100 frames/s, with an inter-frame distance of 200 \(\upmu \)m. For all scanners, the spatial resolution was 20 and 30 \(\upmu \)m in the axial and lateral directions, respectively. The depth of the scan range was 4.3 mm. Acquired images were sampled at \(968 \times 504\) pixels.

Fibrous cap thickness was quantified using a previously validated framework [46, 47]. This semi-automated approach consists in a robust contour segmentation scheme based on the fast marching methodology [5]. Briefly, a cost function is derived from the intensity gradient of the polar OCT image along the radial direction. A front propagation scheme is then run, favoring low cost points (i.e., data attachment term) while penalizing radial displacements (i.e., smoothness constraint term). The optimal path, corresponding to the segmentation contour, is finally extracted by means of a back-tracking scheme in the propagated values.

The three principal phases of the framework are (i) a manual selection of the region of interest (ROI) containing the fibrous cap to be analyzed; (ii) the automatic extraction of the luminal interface over the entire vessel circumference; and (iii) the automatic extraction of the abluminal interface of the fibrous cap within the ROI. Fibrous cap thickness is finally calculated as the distance between both contours of the cap, along a set of lines perpendicular to the luminal interface.

To compute WSS, the lumen of a coronary artery was first reconstructed based on two angiography images, following a previously adopted method [15]. These two angiography images were recorded with angular difference of at least \(30^{\circ }\) and were selected in the same cardiac phase, as determined from the ECG information. The contrast-filled lumen in the two-dimensional (2D) images was segmented by a trained observer (JS). From the 2D segmented contours, a 3D volume was generated, using validated commercially available software (CAAS v5.11, Pie Medical Imaging, Maastricht, the Netherlands) [14]. Extension at the inlet and the outlets of five times the radius was added to exclude computations artifacts [29]. Next, based on the volume of the lumen, a mesh was generated to compute WSS.

The volume meshes were built using the standard meshing tools in ICEM (ANSYS ICEM-CFD v14.5, Ansys Inc, Canonsburg, USA). Prior to the final simulation, a grid dependency study was performed. The grid size was decreased stepwise until the velocity and wall shear stress did not differ more than \(3\,\%\) for each node. This resulted in a typical cell size of 0.1 mm. At critical points (i.e., stenoses and curvature in bifurcation regions), the cell size was reduced up to \(30\,\%\) of the original size. A five-element layer of prism cells was added to optimally capture boundary layer effects. Finally, this resulted in a typical mesh size of \(2\times 10^6\) cells. The inlet and outlet boundary conditions for these computations were assigned by applying scaling laws derived in a previous study [10]. In that study, the absolute inflow and the outflow ratio over bifurcations were derived from in vivo measurements in mildly diseased coronary bifurcations. The measured flow rate was subsequently related to the local diameter of the mother and daughter branches. The boundary conditions for our study were assigned by applying these diameter-based scaling laws. WSS was calculated with CFD using a finite volume solver and by applying standard numerical techniques to perform steady-state computations (Fluent v14.5, Ansys Inc, Canonsburg, USA) [28].

From the results, the 3D WSS magnitude at the wall was extracted and converted to a 2D coordinate system. To do so, the planes perpendicular to the centerline were defined at 0.2 mm intervals. The planes were subsequently subdivided into intervals of \(10^\circ \). The average WSS values within the resulting bins were mapped to a 2D representation.

We recently demonstrated that normalized WSS is especially robust for potential errors in inflow and outflow boundary conditions [28]. Therefore, in addition to absolute WSS, this parameter was investigated as well. For the normalization procedure, each 2D WSS map was normalized by its 50th percentile value.

Fusion of OCT and angiography was performed manually using a previously proposed approach [38]. The matching procedure involves two main steps, namely axial and rotational registration (Fig. 1). Axial registration consists in aligning the length and position of a given arterial segment of the OCT pullback with the corresponding segment in the 3D reconstructed vessel. Rotational registration is the task to determine the angular transformation so that the two segments have the same orientation and overlap with each other. These two steps are described below.

To visually display the risk index of the interrogated region on a single map, fusion of cap thickness and WSS was finally performed after the co-registration of OCT and angiograms. Recent findings demonstrated that the best cutoff to predict rupture was \(151\,\upmu \text {m}\) for most representative fibrous caps [45]. Therefore, this value was used as an upper threshold to identify at-risk regions based on cap thickness information. Similarly, \(1.7\,\text {Pa}\) is often used to represent the transition between intermediate and high WSS [31], and this value was selected as a lower threshold to characterize at-risk regions based on WSS information. The different risk levels, determined from fusion of cap thickness and WSS, are presented in Table 1.

The sensitivity of both thresholds parameters was assessed by iteratively building nine different risk maps, corresponding to the combination of three different cap thickness thresholds and three different WSS thresholds. The assessed parameter values were equal to the initially determined central value, as well as the central value \({\pm }7\,\%\), namely 141, 151, and 161 \(\upmu \text {m}\) for the cap thickness, and 1.6, 1.7, and 1.8 Pa for the WSS. Then, the area of each of the four regions corresponding to a specific risk level (Table 1) was quantified, and the total variability of each region under the nine different parameters settings was calculated.

For each included patient, at least one site corresponding to an atherosclerotic plaque covered by a fibrous cap could be identified in the OCT pullback. When more than one site was found, the one corresponding to the largest plaque, both axially and circumferentially, was selected. The lesion was subsequently localized in the two corresponding angiograms. Volumetric reconstruction from angiography could not be performed for two cases, due to insufficient image quality of the side branches that were used as landmarks (i.e., presence of a trifurcation with vessels overlapping with each others in the image plane for the first case, and contours poorly visible with high level of noise for the second case). These two patients were removed from the study, and the framework was applied on the 12 remaining cases.

Volumetric 3D geometries were reconstructed with two angiography views corresponding to the same cardiac cycle, based on the recorded ECG. Ten cases were reconstructed in diastole. The two remaining cases were reconstructed in systole, since over-projection of the side branches did not allow the diastolic phase. Nevertheless, only rigid transformation without deformation of the ROI was observed by the trained reader (JS) between the diastolic and systolic phases.

Cap thickness and WSS were computed using the methods detailed in “Quantification of fibrous cap thickness from OCT images” and “Computation of wall shear stress from angiography images” sections. Co-registration between OCT and angiography was then successfully performed for all cases. Results displaying the co-registration of WSS and cap thickness are presented in Fig. 4, and identification of at-risk regions was finally realized via the fusion of these two parameters in a single map (“Parameters fusion” section).

To quantify the accuracy of the co-registration procedure, the length between the proximal and distal side branches that were used as landmarks (Fig. 1) was measured along the vessel centerline, in both the angiography-derived 3D volume and the OCT pullback. The average length (\({\pm }\)SD) between the two side branches was \(19.0\,\pm \,11.1\,\text {mm}\) in OCT, and \(18.6\,\pm \,10.8\,\text {mm}\) in angiography. Regression analysis demonstrated a strong linear relationship between the length derived from the two modalities with \(R=0.99\), as depicted in Fig. 2. The mean absolute difference between the distance measured in OCT and angiography was \(1.4\,\pm \,0.9\,\text {mm}\), corresponding to \(8.6\,\pm \,6.7\,\%\) of the measured length, with a bias of \(0.4\,\text {mm}\) and a \(95\,\%\) CI of \([-2.8, 3.6]\,\text {mm}\). The average absolute rotation angle between the two side branches was \(77^\circ \,\pm \, 53^\circ \), with a rotation angle per frame corresponding to \(0.6^\circ \,\pm \,0.4^\circ \).

Additionally, for the 12 analyzed patients, the lumen area of the arterial segment was measured in each OCT frame between the two landmark side branches and compared with the lumen area measured at the corresponding location in the 3D volume derived from angiography. A good correlation (\(R=0.89\)) was found between the two set of measurements, as displayed in Fig. 3.

For the 12 analyzed patients, the average (±SD) cap thickness over the entire plaque was \(278\,\pm \,119\,\upmu \text {m}\), and the thinnest portion of each fibrous cap was \(129\,\pm \,69\,\upmu \text {m}\). The most representative cap thickness value was also quantified, following a previously established approach [45]. Briefly, this value was defined as the average of a collection of cap thickness measurements, performed in the cross-sectional image at five randomly selected sites, and every 0.5 mm along pullback direction throughout the plaque. The average value of the most representative cap thickness for the 12 cases was \(280\,\pm \,126\,\upmu \text {m}\). No patients had a representative cap thickness inferior to the \(151\,\upmu \text {m}\) threshold [range \((179{-}518\,\upmu \text {m})\)]. Along the z-axis of the OCT pullback, the average length of the region encompassing the analyzed fibrous cap was \(4.5\,\pm \,2.4\,\text {mm}\), corresponding to an average of \(33\,\pm \,19\,\text {frames}\). Along the circumference of the vessel, the average angle of the ROI encompassing the plaque was \(140^\circ \,\pm \,58^\circ \).

The average WSS value at the location of the fibrous cap was \(1.46\,\pm \,1.16\,\text {Pa}\). Such magnitude is typically considered as a medium WSS value. However, the normalized WSS value at the location of the fibrous cap was \(1.58\,\pm \,1.12\). This operation was performed by normalizing the values with respect to the median WSS, within the entire region located between the two side branches used for the co-registration. The normalized value is greater than 1, which implies that the WSS at the location of the fibrous cap was relatively higher than the WSS values of the surrounding regions. Only one case had a relatively lower WSS at the ROI, with a normalized value of \(0.59\,\pm \,0.13\). This demonstrates that, in all cases but one, the highest stress values also corresponded to the location of the fibrous cap.

No direct association between WSS and cap thickness was observed. For the 12 analyzed patients, the average composition of the risk map (Fig. 1) was the following: \(35\,\%\) of the total region corresponded to a high WSS (i.e., \({\ge }1.7\,\text {Pa}\), medium risk), \(5\,\%\) to a thin cap (i.e., \({\le }151\,\upmu \text {m}\), medium risk), \(2\,\%\) to both of these two factors combined (i.e., high risk), and \(58\,\%\) to none of them (i.e., low risk). For three patients, more than \(80\,\%\) of the plaque was under high WSS levels. For two other patients, a thin cap covered more than \(20\,\%\) of the total plaque region. For another patient, both of these two factors were present in an area equal to \(15\,\%\) of the plaque. The other participants did not meet these criteria and had a plaque with relatively smaller regions of elevated WSS and/or thin fibrous cap.

For the 12 assessed cases, the distance between the center of the analyzed plaque and the two side branches encompassing the ROI was assessed. No caps were present in the bifurcation regions. The average distance was \(11\,\pm \,9\,\text {mm}\) from the proximal side branch and \(9\,\pm \,6\,\text {mm}\) from the distal side branch. By subdividing the arterial segment between the proximal and distal side branches into four sections of identical length, the plaque was located in the first section for one case, in the second section for three cases, in the third section for four cases, and in the fourth section for four cases.

The sensitivity of both the cap thickness threshold and the WSS threshold was assessed by generating a set of different risk maps with parameter values that ranged between \({\pm }7\,\%\) of the central value. For all cases, the mean total variability of the area of the regions in the risk map was the following: from \({-}7\) to \({+}6\,\%\) for the gray region (low risk); from \({-}9\) to \({+}9\,\%\) for the blue region (intermediate risk); from \({-}31\) to \({+}55\,\%\) for the yellow region (intermediate risk); and from \({-}67\) to \({+}70\,\%\) for the red region (high risk).

The computations were performed on a desktop computer equipped with an Intel Xeon 2.4 GHz processor with 4 cores and 12 GB of memory. For each patient, the average computation time was the following: 7 s to compute the fibrous cap thickness, 10 s to generate the 3D geometry reconstruction, and 2 h to compute the WSS.

Proof-of-concept results were obtained by processing 12 coronary segments imaged from 12 in vivo patients. An atherosclerotic plaque covered by a fibrous cap was identified in all OCT pullbacks. In all cases, the computed WSS was within the expected physiological range, with elevated WSS at the plaque location. Co-registration between OCT and angiography was performed using two side branches as anatomical landmarks. The accuracy of the co-registration method was validated by the high agreement between the corresponding lengths of the co-registered segments.

The 12 patients analyzed in this study had a representative fibrous cap of intermediate thickness (i.e., \(280\,\pm \,126\,\upmu \text {m}\)). Nevertheless, the average value at the thinnest point of the cap was \(129\,\pm \,69\,\upmu \text {m}\), which is inferior to the lower threshold of \(151\,\upmu \text {m}\) used to characterize at-risk regions. It is expected that a broader range of cap thickness, including caps thinner than \(65\,\upmu \text {m}\), would be observed by including additional patients in the study. The method to quantify cap thickness was developed in-house and previously validated [46]. The clinical applicability of the proposed method is supported by a relatively accurate quantification of cap thickness, with a mean absolute error (±SD) of \(30\,\pm \,37\,\upmu \text {m}\).

The association between coronary WSS levels and various atherosclerosis markers was investigated in several studies [31, 34, 37]. Nevertheless, these studies do not include the side branches in their 3D reconstruction framework when computing the WSS. Although such approaches simplify the modeling procedure, the presence—or absence—of side branches critically influences the resulting flow phenomena near bifurcations, as demonstrated in a recent study [18]. The computed flow field in a straight segment directly in between two side branches is not likely to result in adequate WSS patterns, especially at the proximal and distal ends. Moreover, bifurcated areas also correspond to predilection sites for atherosclerosis development, thus making accurate computation in these areas particularly pressing. In the present study, special care has been taken to incorporate both proximal and distal side branches in the CFD computation in order to provide reliable WSS levels.

A unique and patient-specific risk index map (Fig. 4e) could successfully be generated for all cases. This tool translates all the measured parameters into a single map for interpretation of the results. The proposed risk index map enables fast and simple identification of at-risk lesions. Such approach has potential to provide a useful aid for interventional planning and decision making. In this pilot study, all the (semi-) automatic steps of the framework (Fig. 1) are currently processed off-line. The implementation of the method can be optimized by running the code on a graphics processing unit, thus resulting in faster computations and enabling online utilization of the tool directly in the catheterization laboratory.

Both angiography and OCT imaging provide valuable information to evaluate the patient, namely WSS and cap thickness. No clear relationship was found between extreme WSS values and cap thickness; therefore, the two parameters are likely not to be redundant but instead complementary. Thin fibrous cap reflects a short-term risk of plaque rupture, while elevated WSS better indicates a midterm risk. The underlying concept of atherosclerosis is complex, and new approaches are required to better assess the risk of the patient.

In this study, threshold values are used to define the risk index from cap thickness measurements and WSS computations. Regarding fibrous cap thickness, the threshold of \(65\,\upmu \)m is widely adopted to identify high-risk lesions [2]. However, this empirical value is likely to be underevaluated, since ex vivo tissues usually undergo variable shrinkage rate during histological preparation [33, 39]. Moreover, in such pathology studies [39], vulnerable plaques from patients who died from cardiovascular disease were included. Assuming that these patients harbor a more vulnerable plaque phenotype than our population implies that the cap thickness in our study should be larger as well. A different study established that ruptured plaques in acute coronary syndrome are often associated with a fibrous cap thickness of up to \(100\,\upmu \)m [35]. Another group [45] reported a value of \(80\,\upmu \)m for the thinnest fibrous cap, and \(188\,\upmu \)m micrometer for the representative fibrous caps. Therefore, values reported in this study fall within the range presented in that study, especially given the low number of fibrous caps we investigated. In the present work, the \(151\,\upmu \)m threshold value was adopted, as it has been demonstrated to be the best cutoff to predict plaque rupture for most representative fibrous caps [45].

As for WSS, a consensus in the literature uniformly associates values lower than 1.0 Pa to low WSS levels [26, 41]. Nevertheless, there is still an ongoing debate about the absolute WSS levels used to categorize the high WSS profiles. A first study [3] showed that pathobiological processes started destabilizing the cap above 1.5 Pa. A different team [26] used a cutoff value of 2.5 Pa and reported that this level is associated with a transition into a phenotype of vulnerable plaque. Another study [42] analyzed regions where the WSS levels were above 4.0 Pa, and observed that these regions corresponded to areas with larger necrotic core and higher plaque burden. In the present work, we adopt the threshold value of 1.7 Pa. The choice of this value of 1.7 Pa was motivated by the fact that this value was proposed as the optimal threshold between moderate and high WSS in the largest cohort study to date that investigated atherosclerotic disease progression and localized WSS values [31]. Nevertheless, critical WSS levels need to be further evaluated to reach a consensus and determine the actual clinically relevant values. The novel risk index introduced in the present study can potentially contribute to re-establish relevant WSS values.

The resulting risk map (Fig. 4) depends on the value of both cap thickness and WSS threshold parameters. A certain variability could be observed when changing one or both of the parameter values. The largest variation was observed in the at-risk region, with changes ranging between \({-}67\) and \({+}70\,\%\) of the area of the initial region. The large amplitude of these variations can be explained by the fact that the actual size of this region was also very small (namely \(2\,\%\) of the total plaque region); therefore, even a small change in the area would result in a large relative variation.

The clinical significance of this study is also supported by previous work [37] that combined OCT and angiography imaging, and reported that coronary regions exposed to low WSS were associated with larger lipid burden, thinner fibrous cap, and higher prevalence of thin-cap fibroatheroma. Nevertheless, no association was found between peak WSS and low cap thickness in the present study. However, the number of patients involved in this study is relatively limited. Furthermore, results only represent a snapshot in time from which the current values of cap thickness and WSS can be quantified; therefore, the clinical interpretation is limited. Atherosclerosis is a process that develops over time and follows a complex evolution [23]. The different manifestations of the syndrome on the arterial wall, such as plaque composition, morphological alteration, flow patterns, and WSS, also undergo a progressive variation of which the rate may greatly vary from one case to another. A follow-up study would be required to assess the state of the plaque over time, better characterize the evolution of the atherosclerotic process, and assess the added value of the two combined parameters for clinical decision making. Especially for follow-up studies, normalized WSS might be an attractive marker to investigate. Clinical studies have demonstrated a relationship between this marker and enhanced plaque vulnerability [6, 12, 13], and we recently demonstrated that this marker is relatively insensitive to the imposed boundary conditions [28].

In the present study, the most representative and largest plaque was selected in each pullback, and five fibrocalcific plaques were analyzed in addition to seven fibrous plaques. Therefore, the impact of fibrous cap thickness per se was not systematically assessed. Nonetheless, the principal aim of this study was to demonstrate the feasibility of image co-registration between OCT and angiography; therefore, these patients with fibrocalcific plaques were included as well. The 3D volume was generated using a pair of angiography images at two different angles. In such procedure, the cross-sectional lumen geometry was approximated to an elliptic shape. Nevertheless, this assumption had a limited effect on the lumen area, as indicated by the good correlation between the actual lumen area quantified in OCT and the approximated measurements derived from angiography (\(R=0.89\), Fig. 3). This is in good agreement with previously reported values [25, 30, 36]. Furthermore, several other studies have shown that angiography-based reconstructions of coronary arteries can be used to provide a good representation of the WSS patterns [15, 29, 34].

Other investigators concluded that WSS from steady-state simulations is in close agreement with time-averaged WSS results from transient computations [8, 16, 21]. Therefore, steady-state computations were performed in this study in order to speed up computation time and make online application feasible. However, this omits the possibility of performing fluid–structure interactions to inspect wall stresses. Future studies, where direct clinical feasibility is less pressing, should focus on correlating wall stresses and rupture risk.

In the present study, 3D reconstruction of the lumen geometry was performed using two angiograms. Therefore, the luminal contour was approximated to an elliptical shape and then used to compute the WSS. Such approximation could result in significant errors (i.e., overly smoothed shear stress maps) when analyzing severely diseased vessels. In these cases, accurate geometry using invasive imaging such as OCT-derived reconstructions would provide more reliable results [25]. Nevertheless, the scope of the present study concerns mildly diseased (i.e., yet untreated) arteries. In such cases, a good agreement was demonstrated between shear maps derived from angiography and OCT [34].

In the scope of the present work, WSS computation took approximately 2 h. In this study, we focus on relevant clinical markers extracted from off-line WSS computations. Nevertheless, optimizing the implementation would enable to decrease the processing time down to a couple of minutes. Therefore, by performing angiography imaging first, subsequent WSS computations could be performed, while the patient is imaged with OCT, and the two imaging parameters could be co-registered as proposed in the current study, while the patient is still being investigated in the cath lab.

Plaque vulnerability and plaque mechanics can be assessed with fluid–structure interaction (FSI). Although FSI would yield added value to the analysis, it is also computationally extremely expensive. Moreover, it was demonstrated that cap thickness is the most important determinant of peak cap stress and can therefore be regarded as a geometrical surrogate marker for peak cap stress [1, 19, 24]. For these reasons, FSI analyses were not performed in this study, as cap thickness was used as an imaging biomarker reflecting the short-term rupture risk.

Between the two side branches used as landmark for the registration procedure, the stack of OCT frames was linearly rotated to match the orientation of the 3D volume. Due to the absence of landmarks between these two side branches, it was not possible to assess the accuracy of this interpolation. Another limitation lies in the fact that, although the 3D reconstruction of the vessel was performed using two angiography views in an identical cardiac phase, the OCT-derived information was acquired all along the cardiac cycle and did not systematically match the same cardiac phase. Accordingly, spatial shift could have been introduced between the parameters derived from angiography and OCT. The change in curvature of the artery is likely to induce, up to a certain extent, different WSS values during the cardiac cycle. Additionally, the thickness of the cap may also undergo some variation due to the mechanical forces compressing the tissues. This issue could be addressed with an ECG-triggered acquisition of the OCT pullback, in such a way that the ROI is imaged during the diastolic phase to match the angiography information. The recently introduced heartbeat OCT technique [40] that can scan a complete coronary artery within one cardiac cycle at 3200 frames/s could also be used to reduce such motion and deformation artifacts.