Published ahead of print on April 14, 2011
doi: 10.3174/ajnr.A2537

American Journal of Neuroradiology 32:E122, June-July 2011
© 2011 American Society of Neuroradiology

D.F. Kallmes^a
aMayo Clinic
Department of Radiology
Rochester, Minnesota

I read with interest the article by Geers et al¹ regarding the differences in estimated flow between CT angiography and 3D rotational angiography (3DRA). In that study, the authors noted almost a 50% difference in calculated mean wall shear stress between the 2 imaging modalities in a series of 10 patients. The remarkable variability in output as a function of imagingtechnique alone is enough to cast doubt on the utility of parameters such as wall shear stress in computational fluid dynamics (CFD) research. However, the authors never shared with us perhaps the most relevant imaging question of all: What is the “truth?”

The article never mentions which of the 2 imaging studies is considered the standard of reference, so we have no idea which of the 2 outputs we should trust. While I may be misreading the literature, it is my sense that most CFD researchers inherently trust 3DRA as the imaging technique with the highest fidelity. That would be incorrect, however, because the reconstructed 3DRA image is itself just a computer programmer’s rendition of the truth. Indeed, it is well known to experienced interventionalists that the 3DRA should not be trusted, especially when defining the aneurysm neck, because the reconstructed image systematically overestimates neck breadth compared with 2D digital subtraction angiography (DSA) (Fig 1).² I would like to go out on a limb and claim that the 2D image is the truth until proved otherwise.

In short, the authors have done an excellent job of raising further doubt about the clinical utility of CFD but have unfortunatelyfailed to point us in the direction of the truth.

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 Neuroradiol2011;32:581–86[Abstract/Free Full Text]
2. Brinjikji W, Cloft H, Lanzino G, et al. Comparison of 2D digital subtraction angiography and 3D rotational angiography in the evaluation of dome-to-neck ratio. AJNR Am J Neuroradiol 2009;30:831–34. Epub 2009 Jan 8[Abstract/Free Full Text]

Reply

Published ahead of print on April 14, 2011
doi: 10.3174/ajnr.A2562

American Journal of Neuroradiology 32:E123, June-July 2011
© 2011 American Society of Neuroradiology

A.J Geers^a, I. Larrabide^a, A.G. Radaelli^a, H. Bogunovic^a, M. Kim^a and A.F. Frangi^a
aUniversitat Pompeu Fabra
Barcelona, Spain

H.A.F. Gratama van Andel^b, C.B. Majoie^b and E. VanBavel^b
bUniversity of Amsterdam
Amsterdam, the Netherlands

We greatly appreciate the comments by Dr Kallmes regarding our article on the differences in computational fluid dynamics (CFD) simulations of aneurysmal blood flow due to the choice of imaging technique between CT angiography and 3D rotational angiography (3DRA).¹ We found large quantitative differences in the estimation of hemodynamic variables, but qualitative variables that describe the main flow characteristics were reproduced across imaging modalities.

In our population, the estimated mean wall shear stress on the aneurysm differed on average 44.2%. Although this result indeed encourages us to be prudent when analyzing quantitative data, we think that Dr Kallmes’ doubt about the utility of CFD-derived hemodynamic variables is not fully justified. The main flow characteristics that reproduced well in our study have also been found to compare well with in vivo data,^2,3 and they seem to provide valuable information regarding aneurysm development. Last year, the American Journal of Neuroradiology published 2 articles by Cebral et al^4,5 demonstrating the potential of CFD simulations in a study of 200 cases imaged with 3DRA. Theauthors found associations between aneurysmal rupture risk and both qualitative and normalized quantitative variables, suggesting that despite inaccuracies in the estimated magnitude of hemodynamic forces, valuable information can be derived from flow patternsalone.

We never set out to answer the question “Which imaging technique is the standard of reference?” in our study. With its higher spatial resolution and lower visibility of bone, we can expect 3DRA to provide superior anatomic accuracy in comparison with CTA and, therefore, superior accuracy in the CFD simulation. However, without data on the true geometry of the vasculature and the true hemodynamics, we were not in the position to support statements about which of the imaging modalities in our patient data produced estimations closer to the “truth.” As mentioned before, other studies did make comparisons with in vivo data and found the main flow characteristics of CFD simulations to agree well.

We thank Dr Kallmes for providing new images related to the study by Brinjikji et al⁶ that argued in favor of 2D digitalsubtraction angiography over 3DRA in performing anatomic measurements. 2D imaging techniques are not an option when constructing 3D vascular models for CFD simulations, but the findings of Brinjikji et al illustrate clearly that better spatial resolution will naturally lead to improved neck characterizations⁷ and, more generally, that advances in imaging techniques will naturally lead to more accurate vascular models. However, we would like to emphasize that the vascular models we used were not threshold segmentations that depend strongly on the choice of threshold value (in the way that the size of vascular structures in the visualization of 3DRA images depends on window/level settings). The vascular models were instead obtained by using a completely automatic geodesic active region segmentation algorithm. More details on this method and its accuracy are provided by Hernandez and Frangi.⁸

We hope that we have shed some light on the reproducibility of CFD simulations across imaging modalities. Despite the inaccuracies in quantitative hemodynamic variables, we genuinely believe that CFD simulations have proved and will continue to prove useful in understanding the initiation, growth, and rupture of aneurysms and will 1 day find their way into clinical practice to provide the clinician with patient-specific accurate information on the hemodynamic condition of an aneurysm.

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[Abstract/Free Full Text]
2. Ford MD, Stuhne GR, Nikolov HN, et al. Virtual angiography for visualization and validation of computational models of aneurysm hemodynamics. IEEE Trans Med Imaging 2005;24:1586–92[CrossRef][Medline]
3. Karmonik C, Klucznik R, Benndorf G. Comparison of velocity patterns in an AComA aneurysm measured with 3D phase contrast MRI and simulated with CFD. Technol Health Care 2008;16:119–28[Medline]
4. Cebral JR, Mut F, Weir J, et al. Association of hemodynamic characteristics and cerebral aneurysm rupture.AJNR Am J Neuroradiol 2011;32:264–70[Abstract/Free Full Text]
5. Cebral JR, Mut F, Weir J, et al. Quantitative characterization of the hemodynamic environment in ruptured and unruptured brain aneurysms. AJNR Am J Neuroradiol 2011;32:145–51. Epub 2010 Dec 2[Abstract/Free Full Text]
6. Brinjikji W, Cloft H, Lanzino G, et al. Comparison of 2D digital subtraction angiography and 3D rotational angiography in the evaluation of dome-to-neck ratio. AJNR Am J Neuroradiol 2009;30:831–34. Epub 2009 Jan 8[Abstract/Free Full Text]
7. Kallmes DF, Layton K, Marx WF, et al. Death by nondiagnosis: why emergent CT angiography should not be done for patients with subarachnoid hemorrhage. AJNR Am J Neuroradiol 2007;28:1837–38[Free Full Text]
8. Hernandez M, Frangi AF. Non-parametric geodesic active regions: method and evaluation for cerebral aneurysms segmentation in 3DRA and CTA. Med Image Anal 2007;11:224–41. Epub 2007 Feb 25[CrossRef][Medline]