Over the last decade, computational modeling of hemodynamics in cerebral aneurysms has grown tremendously as a field of research. The expectation is that one day patient-specific hemodynamic data will complement the currently available information and improve clinical decision-making. Although in vivo techniques that measure hemodynamic quantities are getting better, computational modeling provides superior temporal and spatial resolution. It also has the distinct advantage of enabling the exploration of hypothetical conditions, which is essential for treatment planning and for noninvasively evaluating the hemodynamics at different levels of physical activity.

The theory and techniques of computational fluid dynamics are well-established. Yet, when applying them to study a new problem, especially one as complex as blood flow, we need to carefully choose appropriate assumptions and understand the sensitivities to input parameters. What numeric resolution is needed to capture all relevant physical phenomena? Should vessel wall motion be incorporated? Do we need to simulate the pulsatility of blood flow, or does a steady flow simulation suffice? How sensitive is the computed flow field to vascular geometry and flow rate boundary conditions?

Computational models have helped us identify associations between hemodynamic factors such as wall shear stress (WSS) and aneurysm initiation, growth, and rupture. They have contributed to our understanding of hemodynamic changes due to endovascular treatment with coils or flow diverters. It has, however, also become apparent that the reproducibility and accuracy of the models are still limited. Accordingly, much research has been dedicated to characterizing these limitations, and there is an ongoing effort to reduce them.

Our paper in the American Journal of Neuroradiology addressed one of the key input parameters of hemodynamic simulations, vascular geometry. Simulations are in part personalized by extracting 3D models of the vasculature from angiographic images. Depending on the stage of the patient care cycle, a different trade-off is made between the invasiveness of the modality and the quality of the image. This motivated us to study the reproducibility of hemodynamic simulations across 3D rotational angiography (3DRA) and CTA.¹ Compared with CTA, 3DRA produces vascular models with superior anatomic accuracy, but its relative invasiveness restricts use to treatment and pretreatment planning. CTA is used for diagnosis and follow-up. We found that CTA-derived vascular models overestimated the neck size and often lacked small vessels. The main flow characteristics of aneurysms were reproduced, yet there were substantial discrepancies for quantitative hemodynamic variables, such as the average WSS on the aneurysm.

In response to our paper, Dr. Kallmes pointed out that even 3DRA should not be "trusted" to produce anatomically accurate models.² Just as CTA overestimates the neck size with respect to 3DRA, 3DRA overestimates the neck size with respect to higher resolution digital subtraction angiography (DSA) images. DSA is a 2D imaging technique and therefore not suitable to reconstruct 3D vascular models. However, by digitally reducing the neck size of 3DRA-derived vascular models using DSA as a guide, Schneiders et al recently revealed that neck overestimations in 3DRA can have nonnegligible consequences for the computed hemodynamics.³ Clearly, further improvements in anatomic accuracy are required.

Another key input parameter of hemodynamic simulations is the flow rate boundary conditions. In 2014, we published a paper in the Journal of Biomechanics in which we assessed the potential of steady flow simulations to approximate the time-averaged WSS.⁴ As part of this study, we varied the flow rate at the vasculature's inlet. Even for variations within a physiologic range, the changes in WSS on the aneurysm were very large. Fortunately, these changes were mostly restricted to the WSS magnitudes, not the distributions. Given the common lack of patient-specific flow rate boundary conditions, this suggests analysis should focus on normalized WSS fields — highlighting the distribution rather than the magnitude — until patient-specific data become readily available. Perhaps even more so than with improving the accuracy of vascular models, which will occur at the "periphery" of our field through advances in imaging and segmentation, hemodynamic modelers should play an active role in retrieving flow and pressure information from the patient. Moreover, rather than using a single set of boundary conditions, we could employ stochastic simulation techniques to obtain a space of possible hemodynamic conditions and account for the great variability observed under different levels of physical activity.

References

1. Geers AJ, Larrabide I, Radaelli AG, et al. Patient-specific computational hemodynamics of intracranial aneurysms from 3D rotational angiography and CT angiography: an in vivo reproducibility study. AJNR Am J Neuroradiol 2011; 32:581–86, 10.3174/ajnr.A2306
2. Kallmes DF. Identifying "truth" in computational fluid dynamics research. AJNR Am J Neuroradiol 2011;32:E122, 10.3174/ajnr.A2537; author reply, E123
3. Schneiders JJ, Marquering HA, Antiga L, et al. Intracranial aneurysm neck size overestimation with 3D rotational angiography: the impact on intra-aneurysmal hemodynamics simulated with computational fluid dynamics. AJNR Am J Neuroradiol 2013; 34:121–28, 10.3174/ajnr.A3179
4. Geers AJ, Larrabide I, Morales HG, et al. Approximating hemodynamics of cerebral aneurysms with steady flow simulations. J Biomech 2014;47:178–85, 10.1016/j.jbiomech.2013.09.033

 

Read this article at AJNR.org …