Software-based simulation for pipeline vantage flow diverter preprocedural assessment: Method and validation study

Abstract

Background

Flow diverters (FDs) have revolutionized intracranial aneurysm (IA) treatment. Proper FD sizing is crucial for aneurysm occlusion and complication prevention. Ankyras (Mentice, Gothenburg, Sweden) is a device-specific sizing software. We evaluated Ankyras’ performance in predicting the final deployment of the Pipeline™ Vantage Embolization Device with Shield Technology™ (PVST) (Medtronic Neurovascular, Irvine, California, USA).

Methods

We analyzed with Ankyras software the three-dimensional rotational angiography (3DRA) images of 10 consecutive patients with unruptured IA treated with PVST. Conventional digital subtraction angiography (DSA) measurements were used for FD sizing, FD was implanted, and postprocedure DSA and 3DRA images were obtained. Ankyras software generated an aneurysm model and simulated the FD size used in actual procedures. We compared the simulated length (SL) from Ankyras and the labeled length (LL) provided by the vendor to the real-case postdeployment measured length. We also compared the expansion from Ankyras simulation (SE) with the real case measured expansion (ME).

Results

Our analysis revealed a mean accuracy for SL across all cases of 92.05% (SD: 4.93%; range: 81.60%–99.10%), while the LL accuracy was 78.71% (SD: 12.43%; range: 62.20%–98.38%). A Pearson R² test indicated a strong correlation for SL at 0.9818, compared to 0.8625 for LL. Furthermore, the mean accuracy for expansion prediction was 86.28% (SD: 4.88%; range: 79.34%–92.46%).

Conclusion

The Ankyras software shows promise as a viable tool for sizing PVST. Enhancing the accuracy of expansion predictions may further improve the precision of device specific simulation.

Graphical abstract.

This is a visual representation of the abstract.

Introduction

Intracranial aneurysms (IAs) are well-known as common and potentially lethal conditions. ¹ When ruptured, IAs lead to subarachnoid hemorrhage, which may cause severe complications, such as vasospasm and hydrocephalus. ² In this context, as early detection and management of this disease is key, head imaging and proper device election are crucial. ³

When it comes to treatment, the endovascular approach has largely replaced once-standard surgical treatment for ruptured and unruptured aneurysms. ⁴ Moreover, flow diverters (FDs) have become a feasible treatment option in patient management in the last two decades, as these devices exclude aneurysms from circulation and reduce blood circulation inside the sac. This mechanism leads to stasis and thrombosis, fostering the genesis of a neointimal endothelial layer across the aneurysm neck.^4,5

In essence, FD treats IA without manipulating the aneurysm sac, making it unique among other open surgical and endovascular options. ⁶ Moreover, successful endovascular treatment of IA depends on the appropriate device choice and deployment. Several models have tried to predict FD sizing, such as manual three-dimensional rotational angiography (3DRA) or virtual software. ⁷ So far there is no software described in the literature that analyzes Pipeline™ Vantage Embolization Device with Shield Technology™ (PVST) (Medtronic Neurovascular, Irvine, California, USA). We present the first method and validation study of the preprocedural assessment of PVST FD sizing and placement using Ankyras software (Mentice, Gothenburg, Sweden).

Methods

Ten consecutive patients treated with PVST FD for unruptured IAs were enrolled in this retrospective study between January and October 2024 in our institution. The operating physician selected the treatment, device, and technique at the time of the intervention. This study aimed to validate the Ankyras PVST simulation final length and expansion.

Data acquisition and software

Each case had a pre- and postprocedural 3DRA using the Siemens Artis Icono biplane (Siemens HealthCare, Erlangen, Germany). The deidentified pre-3DRA images have a 512, 512, 380 matrix size and a pixel spacing of 0.478 × 0.478 × 0.478 mm. Images were processed in Ankyras software) for 3D-virtual FD simulation and retrospective comparison with the implanted device, measured from the postprocedural image as explained in the following sections. Consent was not required for the study as it was a retrospective review of completely deidentified data.

FD simulation in Ankyras

Ankyras software is a real-time sizing tool to assist clinicians in selecting FD size for IA treatment. Starting from the collected preprocedural 3DRA image and as shown in Figure 1, Ankyras allows the first reconstruction of the vessel model with the aneurysm (Figure 1A), then selects and measures the proximal and distal device targeted positions in vessel (Figure 1B) and finally simulates the desired FD model and size to assess whether the device is suitable for the patient treatment based on two main parameters (Figure 1C): the simulated FD foreshortening and the FD diameter expansion.

Figure 1.

Ankyras pathway to FD sizing showing the reconstruction of the vessel model with the aneurysm (A), selection and measurement of the proximal and distal device targeted positions (B), and device simulation and suitability (C).

Abbreviation: FD: flow diverter.

Measured length and simulated length

To assess the accuracy of Ankyras PVST foreshortening, three lengths were compared for each case: (1) FD labeled length (LL, the length of the stent as stated by the vendor); (2) the measured length (ML, the length of the stent as measured after deployment); (3) the simulated length (SL, the length of the stent as predicted by the software).

The ML of the implanted FD inside the patient was obtained from the postprocedural 3DRA image. Firstly, the FD model was manually segmented using the 3D Slicer software (Figure 2A). Secondly, the segmented FD model was registered with the preprocedural vessel model, previously segmented in Ankyras. Then, synthetic tubular extensions were added at the FD model ends to later obtain a centerline through the complete segmented device using a VMTK filter with a resolution (distance between consecutive points) of 0.25 mm. Finally, the centerline was used to extract the ML as the length between the proximal and distal device ends projected in the centerline. In the case shown in Figure 2B, an ML of 18.976 mm was obtained.

Figure 2.

Methods to extract both the ML and SL. Firstly, the FD model was manually segmented using the 3D Slicer software (A); ML was the distance between proximal and distal device ends projected in the centerline (B). The SL was readily obtained in Ankyras by selecting the size of the FD implanted while keeping the distal end unchanged (C–D) to finally compare SL and ML.

Abbreviations: FD: flow diverter; ML: measured length; SL: simulated length.

The SL was readily obtained in Ankyras by selecting the size of the FD implanted while keeping the distal end unchanged (Figure 2C) for the proper comparison with the ML (Figure 2D). In the case shown in Figure 2, a PVST with a labeled diameter of 4.0 mm and an LL of 16 mm was selected. After simulation, a SL of 18.104 mm was obtained.

Length analysis and error assessment

For each case, the Ankyras SL's error and accuracy were defined considering the ML as the reference value:

$$Simulatedlengtherror=|SL-ML|/ML$$

$$Simulatedlengthaccuracy[\text{\%}]=[1-Simulatedlengtherror]\times 100$$

The final SL accuracy was calculated as the average of all the SL accuracies from the dataset. Similarly, the average error was computed.

Additionally, the FD LL, was also used to calculate the LL accuracy by substituting SL with LL in the length's error and accuracy equations above. Finally, the SL and LL accuracies were compared. All statistical analyses were performed with NumPy and SciPy, two widely used Python packages for data analysis.

Measured expansion and simulated expansion

This study also assessed the accuracy of Ankyras expansion simulation. Similarly to the length assessment methodology, two average expansion measurements were used to compare in each case: (1) the measured expansion (ME, the expansion of the stent as measured after deployment); (2) the simulated expansion (SE, the expansion of the stent as measured by the software).

The ME of the implanted FD in the patient was also obtained from the postprocedural 3DRA image, using the segmented FD 3D model and its centerline, mentioned in the measured length and simulated length section (Figure 3A). As the expansion varies along the length of the device, a set of values was retrieved from each case. First, the centerline was used to slice the FD 3D model and obtain the set of cross-sections along the device. Second, the local expansion was measured as the diameter of the circumference that best fitted that cross-section (represented in yellow in Figure 3B–C) divided by the FD maximum diameter (provided by the device manufacturer). Therefore the ME was defined as a percentage at each cross-section.

Figure 3.

Method to extract the ME at each device cross-section. The 3D segmented device (continuous overlay) is sliced by the centerline (continuous line). Each cross-section of the device shape is fitted to a circumference shape (discontinuous circumferences). The diameter of the fitted circumference is used to extract each ME percentage value.

Abbreviation: ME: measured expansion.

The SE was readily obtained in Ankyras once the corresponding dimensions of the implanted FD were simulated keeping the distal end unchanged (Figure 2C). For each case, the SE was a set of local values along the device (calculated at each device cross-section) measured as percentages, where 100% expansion corresponds to the FD maximum diameter value provided by the FD device manufacturer (+0.25 mm in case of the PVST for the labeled diameter).

Expansion analysis and error assessment

To assess the Ankyras expansion, each cross-sectional of SE was compared to the ME from its corresponding cross-section of the segmented FD model. The assignment of cross-section pairs was possible as the measured FD model was previously registered with the preprocedural vessel and simulated FD (as can be seen in Figure 2D). The paired cross-sections allowed a point-to-point evaluation to assess the accuracy of the Ankyras simulated expansion, considering the ME as the reference value:

$$Expansionerror=|SE-ME|/ME$$

$$Expansionaccuracy[\text{\%}]=[1-Expansionerror]\times 100\text{\%}$$

The accuracy of each case was determined by averaging its cross-section expansion samples. As not all the cases had the same amount of samples due to the different device lengths (Table 1), the global expansion accuracy was calculated as the average of all expansion samples to ensure equitable weighting of all cases across the dataset. The statistical analysis was performed with NumPy and SciPy Python packages.

Table 1.

Dataset and final length and expansion accuracies.

Case ID	FD diam (mm)	FD labeled length (mm)	Aneurysm location	Laterality	Measured length (mm)	Simulated length (mm)	Simulated length accuracy (%)	n expansion samples	Expansion accuracy (%)
1	2.5	10	MCA	Left	15.194	13.470	88.652	40	81.691
2	3.5	20	ICA	Left	21.385	21.194	99.107	62	90.850
3	3.25	16	ICA	Right	19.657	16.042	81.609	47	86.749
4	3	16	VA	Right	22.811	20.128	88.237	59	88.610
5	4	16	ICA	Right	18.976	18.104	95.406	53	79.559
6	3	16	BA	NA	24.575	22.575	91.862	66	79.350
7	5	16	VA	Left	17.256	17.452	98.862	51	92.466
8	2.5	14	MCA	Right	14.230	13.216	92.37	39	86.232
9	4	20	ICA	Left	27.207	25.165	92.496	73	92.170
10	6	30	ICA	Left	48.224	52.335	91.476	151	85.134

Abbreviations: MCA: middle cerebral artery; ICA: internal carotid artery; BA: basilar artery; VA: vertebral artery.

Results

A total of 10 preprocedural and postprocedural angiographic data of patients with unruptured aneurysm cases treated with Pipeline Vantage FD were retrospectively analyzed. We examined aneurysm vessel location and laterality; FD labeled, simulated, and MLs; and FD expansion as shown in Table 1.

The aneurysms were distributed across several major arteries: the internal carotid artery (ICA) (50%), middle cerebral artery (MCA) (20%), basilar artery (BA) (10%), and vertebral artery (VA) (20%). Laterality showed 50% of cases on the right side and 40% on the left, one BA case was categorized as nonapplicable (NA). FD diameters ranged from 2.5 mm to 6 mm, while lengths spanned from 10 mm to 30 mm.

ML and simulated expansion accuracy

The mean SL accuracy across all cases was 92.05% (SD: 4.93%; range 81.60%–99.11%) and the LL accuracy was 78.71% (SD: 12.43%; range 62.20%–98.38%). Table 1 shows the results for each case. When evaluated by aneurysm location, MCA aneurysms achieved a mean SL accuracy of 90.76% (n = 2), ICA aneurysms 92.02% (n = 5), VA aneurysms 93.55% (n = 2), and BA aneurysms 91.86% (n = 1). Besides, LL accuracy was 82.10% in the MCA location (n = 2), 78.99% in the ICA location (n = 5), 81.43% in the VA location (n = 2), and 65.11% in the BA location (n = 1). A Pearson R² test indicated a strong correlation for SL at 0.9818, compared to 0.8625 for LL (Figure 4).

Figure 4.

SL (upper trend) and LL (lower trend) against ML, with their respective correlation coefficients.

Abbreviations: SL: simulated length; LL: labeled length; ML: measured length.

The mean simulated expansion accura cy across the 10 cases (a total of 641 samples) was 86.28% (SD: 4.89%; range 79.35–92.47), detailed in Table 1. Notice that this accuracy result is different from the average that can be obtained from the accuracy values in the table. The global expansion accuracy is calculated as the average of simulation expansion accuracy by location revealed values of 83.96% for MCA (n = 2), 86.89% for ICA (n = 5), 90.54% for VA (n = 2), and 79.35% for BA (n = 1).

Discussion

In this retrospective pilot analysis of a PVST sizing software, Ankyras, we found a mean SL accuracy across all cases of 92.05% (SD: 4.93%; range 81.60%–99.10%), and the LL accuracy was 78.71% (SD: 12.43%; range 62.20%–98.38%). A Pearson R² test indicated a strong correlation for SL at 0.9818, compared to 0.8625 for LL, suggesting that the Ankyras simulation aligns more closely with actual measurements than the labeled size. Moreover, we found a mean expansion prediction of 86.28% (SD: 4.88%; range 79.34%–92.46%). The expansion accuracy data had a slightly less substantial result, with the lowest stratified accuracy of 79.3% in a BA aneurysm case.

FDs, with their braided stent structure, provide high metal coverage and pore density to restore blood flow and promote thrombosis within aneurysms. Their flexible design adapts to the tortuous intracranial vasculature but poses challenges like foreshortening. A short FD may lack proper anchoring and risk dislodgment, while a long one could obstruct collateral vessels. Traditionally, FD sizing relies on measuring vessel length and diameters from 2D angiograms or 3D renderings, combined with manufacturer specifications—an imprecise approach that depends heavily on clinician expertise. Virtual stenting methods now offer a more precise, efficient solution, reducing the learning curve for FD sizing. ⁸

In industrial and research settings, the finite element method (FEM) is widely used. This technique may provide the most accurate results, including push or pull operations, if the precise mechanical data of the device and vessel is imputed. Nevertheless, the calculation and result may take several hours, making it inappropriate in intraoperative clinical settings.^9–11 In the search to improve simulation time effectiveness, researchers developed a spring-mass model, in which an FD braided structure is modeled into a tube consisting of mass points and springs. Although it can save a lot of computational costs, this model cannot properly represent the FD foreshortening behavior when compared to FEM.^12–14

The geometrical deployment method may be the most promising in clinical settings. The main point of this model is to fractionate the FD into a series of tubular segments and align them along the centerline with each segment inscribed in the vessel lumen. This method has a faster sizing process compared to the spring-mass model, and it includes a foreshortening mechanism, recommending an FD based on the chosen landing zone. Commercially available options for this method are Sim&Size (Sim&Cure, Montpellier, France), AneuGuide^TM (ArteryFlow Technology, Hangzhou, China), and Ankyras (Mentice, Gothenburg, Sweden).^8,15–19

In a retrospective study with 189 patients enrolled, neuro interventions that used Sim&Size software were reported to have a lower rate of corrective intervention (9% vs 20%, p = .036), shorter procedure duration (46 min vs 52 min, p = .002), lower median radiation dose (1150 mGy vs 1558 mGy, p < .001), and shorter stent length (14 mm vs 16 mm, p < .001). ¹⁶ Despite being a retrospective, small sample, and nonrandomized, this study suggests that FD sizing simulation may have some benefit in workflow, but validation of the accuracy of this particular software has not yet been revealed.

Similar to Ankyras, the AneuGuide™ algorithm provides a real visualization, allowing the user to manipulate the model for computational fluid dynamics, which maintains both high accuracy and low computational cost. Both softwares also provide a simulation of porosity, which is an important complement to the FD coverage ratio. The mean error between the ML and the calculated length was 6.6% (range 0.32%–21.2%). However, SL by the software remained a more accurate predictor of actual length than LL, ⁸ which is the same conclusion we had in this study.

Previous Ankyras study with a cohort of 82 aneurysms cases treated with four different FD brands, such as DERIVO^® Embolization Device (Acandis GmbH, n = 20), Surpass Streamline™ FD (Stryker Corp, Kalamazoo, Michigan, n = 11), p64^® Flow Modulation Device (Phenox GmbH, Germany, n = 22), and Pipeline™ Embolization Device (Medtronic Inc, n = 29). The average error obtained was 7.6% (median of 5.93%; standard deviation 6.02%). The subgroup analysis of Pipeline (Medtronic Inc.) had a simulation error of 8.47% with a mean standard deviation of 7.22%, ¹⁷ which is similar to our result with an error of 7.95% (accuracy 92.05%–SD: 4.93%; range 81.60%–99.10%).

To the authors’ best knowledge, this is the first study that addressed expansion accuracy using FD software sizing, making it hard to make a direct comparison with the literature on this topic. However, compared to this study length simulation, our simulated expansibility data showed a less accurate analysis (expansion accuracy 86.28%–SD: 4.88%; range 79.34%–92.46%). Thus, a simulation improvement of the device to vessel apposition might also contribute to a more precise length sizing prediction in future versions of the simulation software.

Conclusion

Ankyras software was quite accurate on the sizing of PVST, especially regarding forshortening. The lack of expansion accuracy in previous FD simulation software validation studies might represent a gap in improving simulation precision and future FD sizing algorithms. Addressing the FD expansion simulation and its validation may lead to more reliable simulation results.

Appendix

Abbreviation list

3DRA

Three-dimensional rotational angiography

BA

Basilar artery

DSA

Digital subtraction angiography

FD

Flow diverter

FEM

Finite element method

IA

Intracranial aneurysm

ICA

Internal carotid artery

LL

Labeled length

ME

Measured expansion

ML

Measured length

MCA

Middle cerebral artery

NA

Nonapplicable

PVST

Pipeline™ vantage embolization device with shield technology™

R²

Pearson correlation coefficient

SAH

Subarachnoid hemorrhage

SE

Simulated expansion

SL

Simulated length

VA

Vertebral artery