Background
Intracranial aneurysms (IA) affect up to 5% of the US population [
1]. Rupture of IAs leads to subarachnoid hemorrhage, the most severe form of stroke with high rates of mortality (> 50% at 30 days) and disability (> 50% permanent disability among survivors) [
2]. The traditional IA treatment entails clipping, an open-skull, maximally invasive surgery with significant morbidity and mortality. Since early 1990s, endovascular intervention of coil embolization is widely used as a minimally invasive alternative that has revolutionized IA treatment [
3,
4]. Coil embolization obliterates an IA by filling the sac with platinum coils to reduce aneurysmal inflow and induce aneurysmal thrombosis. For wide-necked IAs (defined as having a neck of ≥ 4 mm or a dome-to-neck ratio of < 2), neurovascular stents are often deployed across the orifice of aneurysms in the parent vessel, typically a high-porosity neuro-stent is deployed across the aneurysm to reduce coil herniation [
5]. This treatment strategy is called stent-assisted coiling. Approximately one-third of IAs use a stent to assist coiling. Though increasing reports of successful endovascular intervention, many events of failure to occlude of IAs [
6] and permanent neurological procedure-related complications [
7] occur.
Moreover, prospective randomized multicenter trials comparing clipping with coiling in both ruptured and unruptured IAs have demonstrated better outcomes with endovascular therapy [
4,
8,
9]. However, the major drawback for endovascular treatment remains high recanalization (recurrence) rates (30%) [
10‐
12] and the need for retreatment in coiled IAs [
4]. Patients experiencing these negative outcomes are subjected to increased risk of IA rupture and complications from retreatment and generally have fewer treatment options available. Unfortunately, there is no way for clinicians to predict outcome of coiling intervention [
9].
In this study, we develop reliable and practical methods for virtual coiling and stenting. As the purpose of deployment of coils and stents is to provide the geometries for accurate post-treatment computational fluid dynamics (CFD) analysis, methods of this study do not require capturing all the details such as the stress and force distribution for the devices and vessel walls. Our philosophy for developing these methods is to balance accuracy and practicality. In this study we develop reliable and practical methods using simplified finite element method (FEM) for virtual coiling and stenting. We consider the mechanical properties of the devices and recapitulate the clinical practice using a FEM approach. At the same time, we apply some simplifications for FEM modeling to make our methods efficient. Once these two methods are developed, standard CFD procedure can be applied to simulate post-treatment hemodynamics in patient-specific IAs to investigate the association and build a statistical prediction model between the flow dynamics and clinical treatment outcomes using large number of treated cases by coiling and stenting in the future. This prediction model will be able to help assess different treatment strategies and choose the optimal treatment option.
Discussion
In this paper, we developed a reliable and efficient surgical planning procedure for intracranial aneurysms. For virtual surgical planning of stent and coil deployment, though some simplifications were applied, the main feature of this process were captured, including the final configurations and positions of the stent and coil, and these results were sufficient for providing the geometries of coils and stents for subsequent accurate post-treatment CFD analysis. The steps to link these modeling techniques with ultimate surgical outcome includes: (1) develop modeling techniques from current study; (2) validate these modeling techniques; (3) once these two methods are developed and validated, standard CFD procedure can be applied to simulate post-treatment hemodynamics in large number of patient-specific IAs; (4) investigate the association and build a statistical prediction model between the flow dynamics and clinical treatment outcomes using large number of treated cases by coiling and stenting. This prediction model will be able to help assess different treatment strategies and choose the optimal treatment option.
Currently, both porous media methods [
25,
26] and fast virtual methods such as dynamic path planning for coiling [
27] and simplex mesh expansion for stenting [
28] are available. They are fast; however, their accuracy is greatly compromised. By representing coils [
25,
29] and stents [
26] as homogeneous porous media, they do not sufficiently capture specific flow interventions. Furthermore, it is tedious to determine patient-specific coefficients for porous media models. The existing expansion-based fast virtual stenting includes the collision detection process [
28], which is inherently unstable, while the existing dynamic path planning method for fast virtual coiling [
27], based on an artificial potential field, is not realistic as it does not take into account the coil pre-shape. The FEM has been applied to virtually deploy stents and coils [
30,
31]. FEM-based techniques are accurate owing to explicit mechanical representation of device deployment. However, as they were used to capture all the details of the mechanical properties and behaviors of the devices and vessels, they are computationally expensive and time-consuming, and thus not practical. For example, a FEM-based HiFiVS technique for flow diverter (FD) deployment takes 100 h for single stent deployment [
30]. Coils alter aneurysmal hemodynamics by reducing aneurysmal inflow, which initiates subsequent thrombosis within the aneurysm, leading to occlusion of the aneurysm and its eventual exclusion from the blood flow circulation. However, flow impingement acting on coil mass or on the aneurysm wall is believed to be responsible for the coil recanalization of coil-treated IAs. Therefore, knowledge about the impact of coils or stent-assisted coils on hemodynamics is critical for predicting coiling treatment outcomes. Image-based CFD analysis can provide detailed information on post-treatment hemodynamics, but it requires realistic representation of coils and stents in deployed states. This presents challenges to the numerical simulation of coil and stent implantation, as previous methods do not efficiently capture realistic coil and stent deployment.
This study focuses on generating accurate deployed coil and stent geometries for subsequent post-treatment flow simulation, taking no account of capturing all the details, i.e., the stress or force distribution. The simplifications used in this paper include: (1) coils are modeled under the assumptions of 3D Euler–Bernoulli beam elements as coils are slender in shape with the length dimension that is much larger than the diameter. It is computationally efficient and provides good geometry representation. The complex-shaped coils (free-stress status without strain energy) will firstly be “pulled” into the catheter to package them (with strain energy) and then “pushed” into the aneurysm sac via a catheter. Though the 3D configuration of coils may be different from that of the clinical treatment, the packing density will be almost the same. Morales et al. have demonstrated that the simulated flow field was independent of coil deployment, when packing density is greater than 22% [
37]. (2) During the stent deployment process, we directly transform the stent–catheter system according to the centerline of the parent vessel since the final configuration of the stent is not as dependent on the deployment history as the FD. (3) The aneurysm and vessel are assumed to be rigid and are fully constrained during the simulation, because the deformation of aneurysm and vessel is minimal after the stent devices were deployed as demonstrated in this study (Fig.
5b). Neuro-stents to assist coiling are self-expandable stents with small radial force, thus rigid wall assumption is sufficient. This is different from the balloon-expandable stent deployment for stenosis-like coronary artery disease, where the contact stress applied to the stenosis from the opening stent, and thus rigid wall assumption can’t capture the deformable arterial wall impacted by stent. Moreover, the angle of aneurysm vessel may change from a sharp-angle configuration to a straight line shape when the radial force applied to the vessel wall from the expansion of stent. This change will have a certain effect to the overall shear stress, compression stress and velocity of the blood flow when the CFD analysis is conducted [
32].
Typically during FEM simulation procedures, packaging a coil into the catheter take around 1 h and crimping a stent could take up to 5 h. Building a library of packaged coils and crimped stents with different dimensions will significantly reduce the simulation time for future cases. After building the library, for virtual coiling we only need to simulate pushing the coil out of the catheter to deploy it into the aneurysm sac; while for virtual stenting, we will only need to transform the system and retract the catheter to release the stent. We expect both coiling and stenting simulation procedures to take around 1 h based on the library. Compared with the previous HiFiVS technique of FD deployment, which takes around 100 h [
30], our proposed methods are very efficient and practical. Even compared with current fast virtual intervention methods, which typically take minutes, our current methods are quite reasonable. Since post-treatment CFD simulations take up to hours for flow simulations even on computational clusters regardless of the device deployment methods used, the additional time needed for FEM-based over fast virtual deployment methods is insignificant. However, our simplified FEM-based coiling and stenting methods intrinsically capture the mechanical features and promise to more accurately recapitulate the device development.
Limitation
There are several limitations of this study. First, we did not conduct the validation of stent and coil deployment results in vitro and in vivo. We will conduct the validation in the future by both in vitro and in vivo data as follows:
1.
Validate the flow field based on the deployed geometry using in vitro data:
We will choose 12 patient-specific aneurysm 3D angiographic images for the algorithm validation based on different aneurysm types and locations. Two-thirds of the patients (8 aneurysms) have narrow-neck aneurysms treated by coiling alone, while the rest one-third of the patients (4 aneurysms) have wide-neck aneurysms treated by stent-assisted coiling to match the ratio of clinical coiling treatment strategies. These 3D IA images will be segmented to obtain the 3D geometries with STL files. From the STL files, we will build aneurysm phantoms made of elastomer as described [
33] and deploy coil and stent devices as done in the real patients. We will connect the phantoms to a flow loop mimicking cerebral blood flow and use stereoscopic Particle Image Velocimetry (PIV) to measure the post-treatment 3D pulsatile aneurysmal flow fields (at peak and end of diastole) for flow validation. Specifically, we will use PIV to measure the 3D velocity field only at the IA neck plane as the flow field inside the sac could not be measured by PIV due to the reflection of densely packaged coils. Measured PIV images will be automatically uploaded into the PIV post-processing system to generate in-plane velocity vectors, magnitude contour and out-of-plane velocity vectors. Flow patterns at the IA neck plane will be qualitatively compared for each case between PIV measurement and CFD simulation.
2.
Validate the flow field based on the deployed geometry using in vivo data:
For in vivo validation, we will generate virtual angiograms from the CFD results of these 12 cases by simulating scalar transport and contrast density projection [
34,
35], and compare them with clinical angiograms. Comparison will be done both qualitatively (flow patterns including jets, recirculation zones) and quantitatively (contrast residence time).
We expect that in vitro and in vivo testing will validate the accuracy of our simplified FEM-based virtual coiling and stenting methods in calculating hemodynamic parameters of patient-specific IAs. Especially for the primary coiling treatment (current study scope), coil packing density is always around 27–28% clinically [
36]. Frangi et al. have demonstrated that the simulated flow field was independent of coil deployment, when packing density is greater than 22% [
37].
Second, the material parameter values chosen for capturing deformable cerebral arterial wall was based on the values from carotid or abdominal arteries, which may be not appropriate for the cerebral artery. We will obtain the HGO parameter values from tensile tests of human cerebral arterial vessels from autopsy study by cooperating with hospitals in the future. We will then do the simulation by replacing the material parameter values from experiments. Third, we simulated the stent and coil deployment process to only one patient-specific aneurysm, which is not sufficient to cover the different shapes of aneurysms.
Conclusion
In this study, two types of surgical planning simulation were performed, the stent deployment and coiling. A simplified modeling approach to simulate and understand these processes were developed. In particular, the use of 3D Euler–Bernoulli beam elements for modeling coils and transformation according to the centerline of the parent vessel for the delivery of stent–catheter system were carried out. The aneurysm and vessel are assumed to be rigid and are fully constrained during the simulation, and all the stent–catheter systems and coil–catheter systems were prepared and packaged as a library which contained all types of stents and coils. It is computationally efficient and provides good geometry representation, which can provide the geometries for subsequent accurate post-treatment CFD analysis.
Authors’ contributions
JPX and XLZ conceived and designed the study. YQJ conducted the literature search. XCL and YW were involved in performing the finite element analysis of stent and coil deployment. JPX and JX supervised the simulations. XCL and JPX analyzed the results and wrote the manuscript. All authors read and approved the final manuscript.