Anu
Anu
Anu
Comprehensive validation of computational fluid dynamics simulations of in-vivo blood flow in patient-specific
cerebral aneurysms
Med. Phys. 39, 742 (2012); 10.1118/1.3675402
[DOI: 10.1118/1.3652917]
Key words: blood flow, aneurysm, phase-contrast magnetic resonance angiography, computational
fluid dynamics
6300 Med. Phys. 38 (11), November 2011 0094-2405/2011/38(11)/6300/13/$30.00 C 2011 Am. Assoc. Phys. Med.
V 6300
6301 Jiang et al.: A comparative study of Aneurismal Hemodynamics 6301
II.B. Description and calibration of the accelerated II.C. Description of “animal-specific” CFD simulations
PC-VIPR MRA technique
Temporally-averaged MRA velocity magnitude data from
The 3D phase-contrast Vastly undersampled isotropic pro- the acquisitions described above were transferred to an offline
jection reconstruction (PC-VIPR) MR acquisition is a highly personal workstation (Dell Precision 390, Dell Inc., Austin,
undersampled 3D radial acquisition that provides isotropic TX). These data were then interpolated to achieve isotropic
voxel dimensions.12 PC-VIPR MRA exploits the sparsity of voxel sizes of 0.225 0.225 0.225 mm3 from their native
the vascular data sets and allows for accelerated imaging as resolutions (0.525 0.525 0.525 mm3 and 0.62 0.62
compared to conventional Cartesian PC-MRA.12 All MR 0.62 mm3 for Aneurysms A and B, respectively). Segmenta-
imaging was performed with a 1.5 T clinical scanner (Signa tion of the volume of interest (VOI) was accomplished with a
HD, GE Healthcare, Waukesha, WI, USA). A standard eight- commercially available package, ScanIP (Simpleware Inc.,
channel knee coil was used for imaging. An optimized Exeter, United Kingdom).29 To verify the realism of the geo-
PC-VIPR MRA method28 provided us with time-resolved 3D metries reconstructed from MRA data, these were visually
voxel data, each of which has three-directional flow velocity compared to the ones derived from 3D-DSA data using an
components, using the following parameters: a TE=TR of open source visualization package (Paraview, Kitware Inc.,
3.6=14.0 ms, retrospective ECG gating, a bandwidth of NY). Besides ensuring visual consistency, geometrical meas-
631.25 kHz, temporal resolution of approximately 35 ms, urements (aneurysm height, width, surface area, and volume)
and scan time of 8 min. The scan time was short because an were taken from both the MRA-based and 3D-DSA-based
accelerated PC-MRA technique was used. More details can geometries. The averaged differences are less than 10% for
be found in previous publications.12,28 The field of views both aneurysms. The reconstructed geometry was imported to
(FOVs) were 13 13 13 cm3 and 16 16 16 cm3 for a voxel-based mesh generator30 ScanFE (Simpleware Inc.,
aneurysms A and B, respectively, thereby resulting in iso- Exeter, United Kingdom) to create an unstructured mesh.
tropic spatial resolution of 0.525 0.525 0.525 mm3 and To compute velocity waveforms in and around the aneur-
0.62 0.62 0.62 mm3 for aneurysms A and B, respectively. ism, we solved the time-dependent incompressible, 3D
The above-mentioned 3D PC-VIPR scans were performed Navier–Stokes equations [i.e., Eqs. (1) and (2)] for the 3D
twice using both a high (1.5 m=s) VENC and a low VENC meshed vessel geometry using FLUENT software, a commer-
(0.5 m=s) in order to better capture both the high velocity cial computational fluid dynamic (CFD) solver (version
jets and areas of slow flow known to be present in and 12.0; ANSYS-FLUENT Inc., Lebanon, NH). The equations
around these experimental aneurysms. Both VENCs were for velocity are written as
chosen along each of the three principal axes so that all three
components of the velocity vector fields were measured by ~ ~
r u ¼ 0; (1)
converting measured phase angles to velocity magnitudes.
All data acquisition was directed by experienced MR physi- @~
u
q þq ~ ~ ~
ur ~ þ lr2~
u ¼ rp u; (2)
cists (K. J. or O. W.). @t
TABLE I. Basic geometrical and flow parameters of Aneurysms A and B. The aspect ratio is defined as the ratio between the aneurysm height and the neck
width.
Aneurysm Inlet diameter (mm) Averaged flow rate (mL=s) Height (mm) Width (mm) Aspect Ratio Volume (mm3) Ostium area (mm2)
where ~ u is the three-dimensional velocity vector, q is the simulated velocity values and wall shear stresses at selected
blood density, p is the pressure and l is the viscosity. The locations were performed similar to methods described by
pressure–velocity coupling in Eqs. (1) and (2) is obtained Valencia et al.36
using the SIMPLEC algorithm. The explicit time-marching Since a steady state computation was used by ANSYS FLUENT
second-order scheme with a time step 2 103 s (approxi- 12.0 as the initial guess for the transient simulation, multiple
mately 350 steps per cardiac cycle) was used for the compu- cardiac cycles were necessary for the result to be considered
tations. Although this time step is relatively coarse, it seems as fully periodic in time. For each simulation, we computed
adequate to capture gross flow features.8 For both aneurysms flow over a total of three cardiac cycles and results corre-
A and B, approximately 0.75 million tetrahedral cells were sponding to the last cycle simulated will be reported below.
used to simulate transient flow [see Eqs. (1) and (2)] using We verified that the periodicity was reached after the second
the finite volume method by the FLUENT Solver. The average cycle.
size of tetrahedral cells is approximately 0.0035 mm3. We
also verified that sufficiently fine (approximately 0.03 mm II.D. Off-line visualization and data analysis
layer thickness) boundary layers were created by the SCANFE
II.D.1. Comparison between PC-MRA measured and
software.
CFD-simulated velocity fields
In Eqs. (1) and (2), blood was modeled as an incompressi-
ble laminar fluid with a density of 1050 kg=m3. Both non- As shown in Fig. 1, we first verified that the segmented
Newtonian and Newtonian viscosity models were used for geometries (i.e., meshed surfaces) were spatially-registered
CFD simulations. In the Newtonian viscosity model, a con- with the PC-MRA signal magnitude data as well as the
stant dynamic viscosity of 0.0035 kg=m s was assumed. Non- phase-contrast velocity data. Then, for point-to-point and
Newtonian blood behavior, which could have important temporally-matched comparisons between the PC-MRA
effects on the flow in low shear rate (c)_ regions, was taken measurements and the CFD-calculated velocity fields, all ve-
into account by use of a Carreau-Yasuda viscosity model as locity data were interpolated onto a rectilinear grid at an iso-
follows: tropic spatial resolution of approximately 0.6 mm (which
l l1 h in1=2 corresponds to the spatial resolution of MRA velocity meas-
¼ 1 þ ðkc_ Þ2 ; (3) urements) using MATLAB (Mathworks Inc., Natick, MA). As
l0 l1
described before, we repeated velocity data acquisition for
where l0 and l1 are the zero and infinite shear rate limit vis- each animal using two VENC settings to capture both low-
cosities, respectively, k is the relaxation time constant and n velocity and high velocity flows. In this step, we simply used
is the power law index. The Carreau model values adopted31 MRA measurements with the low VENC value for compari-
are as follows: l0 ¼ 0:0035 kg=m=s (which corresponds to sons with simulated velocity values by the CFD unless ve-
the value of viscosity used in the Newtonian flow simula- locity aliasing was identified. In the other words, with the
tions), l1 ¼ 0:056 kg=m=s, k ¼ 3:313 s, and n ¼ 0.3568. presence of velocity aliasing in certain spatial locations,
Hereafter, we referred non-Newtonian and Newtonian CFD matched MRA velocity measurements acquired under the
simulations as to CFD1 and CFD2, respectively. high VENC were then used to replace these aliased velocity
An inlet flow boundary condition was chosen from PC- measurements in these locations. The potential beat-by-beat
MRA measurements made in the artery upstream (approxi- variations (e.g., slight heart rate change) accompanied by
mately 4 cm) from the aneurysms. Time-dependent flow rates two data acquisitions were not accounted for in this study.
were obtained by integrating the through-plane PC-MRA ve- To quantify the degree of similarity between the MRA-
locity component over the artery lumen on a slice perpendic- measured and the CFD-synthesized velocity fields, we
ular to its long axis. It is reported that uncertainties of in vivo adopted a similarity metric between vector fields related to
conventional PC-MRA in terms of flow rate range from 3 to that developed by Li and Shen.37 This similarity metric is nor-
13%.32–34 This accuracy remains adequate for use in estab- malized between 0 and 1, 1 being the highest degree of simi-
lishing boundary conditions for CFD calculations.35 Outlet larity. More details are provided in Appendix A. Similarity
boundary conditions of the flow volume ratios of the two out- values were calculated for three pairs of velocity vector fields:
let branches were based on the PC-MRA measurements. MRA versus non-Newtonian CFD simulations (CFD1), MRA
More specifically, user defined functions (UDFs) (ANSYS- versus Newtonian CFD simulations (CFD2) and CFD1 versus
FLUENT Inc., Lebanon, NH) were used to define time- CFD2. We also estimated the absolute magnitude (unit: m=s)
varying flow rates based on PC-MRA measurements at the and angular (unit: degree) differences for these three pairs of
inlet and at one of the outlets by assuming parabolic velocity velocity vector fields.
profiles. Therefore, the flow volume division ratio between Prior to quantitative comparisons of the CFD-simulated
the two outlet branches was a variable and consistent to the and the MRA-measured velocity fields, CFD-simulated veloc-
MRA-measured flow rates (approximately 1:1). A “no-slip” ity data were first interpolated from an unstructured=irregular
boundary condition and rigid wall were assumed. Conver- grid onto a 3D uniform rectilinear grid (0.1-mm voxel size).
gence criteria for continuity and velocity were both set to These re-interpolated velocity fields were then convolved
103 (relative to the largest value in the first five iterations) with a 3D isotropic Gaussian kernel (1.2 1.2 1.2 mm with
and were satisfied for each simulation. Sensitivity tests on a sigma of 0.7) to mimic the “averaging=smoothing” effect
mesh size, step size, and convergence thresholds in terms of by a point-spread function of the MRA system as reported in
the literature.38 Finally, the “convolved” velocity data by the II.D.4. Statistical analysis
above-mentioned Gaussian Kernel were interpolated to a desir-
Pearson’s correlation coefficients were calculated
able resolution for similarity comparison as described above.
between quantities derived from MRA and CFD velocity
Of note, the velocity comparisons between CFD1 and
data to determine their linear correlation. Student’s t-tests
CFD2 were done in a fine grid (0.1-mm voxel size consistent
were performed to assess the statistical significance of the
to the CFD mesh density), while the velocity comparisons
observed correlation values. In this study, a p-value of 0.05
for the other two pairs were performed at a coarse grid
or less indicated statistical significance.
(approximately 0.6-mm consistent to the MRA voxel size).
spatially-registered and temporally-matched (between MRA each figure, three components (Vx, Vy, Vz) and the velocity
and CFD) velocity fields during diastole for one of the trans- amplitude map for this plane are compared, respectively.
verse and one of the longitudinal planes from aneurysm A. In Quantitatively, the similarity metric defined by Eq. (A1)
in Appendix A as well as the absolute magnitude and angular
differences were calculated at the same transverse and longi-
tudinal planes illustrated in Figs. 2(a) and 2(b) between the
MRA-measured and CFD-simulated (both Newtonian and
non-Newtonian) velocity fields as well as between CFD-
simulated velocity fields. Then, these spatially-varying simi-
larity and, absolute magnitude and angular difference values
were displayed in Figs. 3(a) and 3(b). It is easy to see, from
Figs. 2 and 3, that spatial velocity distributions between the
MRA measurements and CFD simulations were in favorable
agreement. We also calculated spatially-averaged similarity
values within the aneurysm dome (i.e., the outlined areas in
the respective similarity maps of Figs. 3(a) and 3(b)), the
spatially-averaged similarity values within the dome of the
transverse plane are 0.67 and 0.66 between the MRA-
measured and the CFD1, and between the MRA-measured
and CFD2, respectively, in Fig. 3(a), while the spatially-
averaged similarity values with the dome of the longitudinal
plane are 0.72 and 0.70 between the MRA-measured and the
CFD1, and between the MRA-measured and CFD2, respec-
tively, in Fig. 3(b). However, the similarity values between
the non-Newtonian (CFD1) and Newtonian (CFD2) are
higher as clearly illustrated in Figs. 3(a) and 3(b). The
spatially-averaged similarity values with the dome of the
transverse and longitudinal planes are 0.89 and 0.91, respec-
tively, between the CFD1 and CFD2 simulations.
The averaged absolute magnitude differences between the
CFD-simulated and the MRA-measured velocities are approx-
imately 0.10 m=s, while the averaged absolute angular differ-
ences range from 20.0 to 33.8 , as shown in Figs. 2 and 3.
Similarly, the absolute magnitude (roughly 0.02 m=s) and
angular (roughly 6 ) differences between the non-Newtonian
(CFD1) and Newtonian (CFD2) velocities are lower.
Similar to what illustrated in Figs. 2 and 3, we analyzed
the similarity values within the domes of the two aneurysms
for all phases of the cardiac cycle at four transverse and four
longitudinal planes. Results are listed in Table II below.
Overall, there was a moderately high similarity (>0.6)
between the MRA-measured and CFD-calculated velocity
vector fields. The absolute magnitude and angular differences
were approximately 0.09 m=s and 40 , respectively. Consist-
ent with the visual perception shown in Figs. 3(a) and 3(b),
FIG. 3. The estimated similarity values [see Eq. (A1)], absolute magnitude the averaged similarity values of velocity vector fields
and angular differences between three pairs of velocity fields [MRA vs. between the CFD1 and CFD2 simulations were 0.85 6 0.04
CFD1, MRA vs. CFD2, and CFD1 vs. CFD2] for (a) the transverse and (b) and 0.78 6 0.06 for Aneurysms A and B, respectively. Con-
the longitudinal planes in aneurysm A illustrated in Figs. 2(a) and 2(b),
respectively. The velocity data correspond to a phase at the diastole. Arrows sistent to this observation, the absolute magnitude differences
on 3D velocity vector plots [top plots of both (a) and (b)] indicate a border between the CFD1 and CFD2 simulations were approximately
region between the velocity jet and the low-velocity recirculation zone. In 0.03 m=s and therefore were significantly lower (p < 0.05), as
this figure, CFD1 and CFD2 denote non-Newtonian and Newtonian CFD compared to these between the MRA-measured and CFD-
simulations, respectively. Manually-delineated contours on both (a) and (b)
indicate regions within the aneurysm dome. Mag (m=s), Ang (degree) and calculated velocity vector fields. The similarities between the
Sim [0-1] denote the absolute magnitude and angular differences, and the CFD1 and CFD2 simulations were significantly higher
similarity metric between two velocity vectors, respectively. The arrows in (p < 0.05). Interestingly, the absolute angular differences
the similarity plots (left lower three columns) point to low similarity values,
where arrows in the absolute magnitude (right lower three columns) and
between the CFD1 and CFD2 simulations in aneurysm B were
angular (middle lower three columns) difference plots point to large discrep- significantly (p < 0.05) higher than these in aneurysm A.
ancies in both magnitude and direction of two sets of comparing vectors. However, in both aneurysms, there was no statistically
TABLE II. Estimated absolute magnitude and angular differences, and similarity values with two aneurysm sacs for three pairs of velocity vector fields: MRA
versus non-Newtonian CFD simulations (CFD1), MRA versus Newtonian CFD simulations (CFD2) and CFD1 versus CFD2. The mean values (6one standard
deviation) were estimated within the manually-segmented aneurysm domes [see Figs. 3(a) and 3(b)].
Aneurysm A Aneurysm B
Magnitude (m=s) Angular (degree) Similarity Magnitude (m=s) Angular (degree) Similarity
MRA-CFD1 0.09 6 0.03 38.4 6 7.4 0.64 6 0.04 0.08 6 0.03 40.0 6 7.4 0.64 6 0.04
MRA-CFD2 0.09 6 0.03 40.7 6 7.5 0.62 6 0.04 0.07 6 0.03 39.6 6 8.0 0.65 6 0.04
CFD1-CFD2 0.03 6 0.01 11.2 6 3.0 0.85 6 0.04 0.03 6 0.01 22.9 6 7.4 0.78 6 0.06
significantly difference among similarity values in the diastole 5(a) and 5(b) above. As compared to values during diastole
and systole. [Figs. 3(a) and 3(b)], the spatially-averaged similarity values
Spatially-registered and temporally-matched velocity within the aneurysm dome between the MRA-measured and
fields at the peak systole for the same transverse and longitu- the CFD1 velocity fields were comparable (0.65 and 0.68 for
dinal planes shown in Fig. 2(a) and 2(b) are shown in Figs. 5(a) and 5(b), respectively), whereas the spatially-
Figs. 4(a) and (b).
The corresponding similarity, and absolute magnitude and
angular difference plots at the peak systole are shown in Figs.
FIG. 5. The estimated similarity values [see Eq. (A1)], absolute magnitude
and angular differences between three pairs of velocity fields [MRA vs.
CFD1, MRA vs. CFD2, and CFD1 vs. CFD2] for (a) the transverse and (b)
the longitudinal planes in aneurysm A illustrated in Figs. 2(a) and 2(b),
FIG. 4. Comparison of 3D velocity vector fields at the peak systole (three respectively. The velocity data correspond to the peak systole. In this figure,
components Vx, Vy and Vz and the velocity amplitude V) obtained from CFD1 and CFD2 denote non-Newtonian and Newtonian CFD simulations,
subject-specific CFD simulations and PC-VIPR MRA measurements for (a) respectively. Mag (m=s), Ang (degree) and Sim [0-1] denote the absolute
a transverse plane and (b) a longitudinal plane in aneurysm A. In both plots, magnitude and angular differences, and the similarity metric between two
CFD1 and CFD2 denote non-Newtonian and Newtonian CFD simulations, velocity vectors, respectively. The arrow in the similarity plots (left lower
respectively. Color bars corresponding to three plots of each column are three columns) points to low similarity values, where arrows in the absolute
shown at the end of each column. The unit used for velocity is m=s for all magnitude (right lower three columns) and angular (middle lower three col-
plots in (a) and (b). The arrow in (b) points to possible MRA measurement umns) difference plots point to large discrepancies in both magnitude and
errors. direction of two sets of comparing vectors.
FIG. 6. Plots of streamlines (i.e., lines of tangent to instantaneous velocity vectors) of CFD simulated and PC-MRA measured velocity vectors at diastole for
(a) Aneurysms A and (b) B, respectively. Streamlines were color encoded using velocity amplitude (0–0.8 m=s). In this figure, CFD1 and CFD2 denote non-
Newtonian and Newtonian CFD simulations, respectively.
averaged similarity values within the aneurysm dome between When all (normal) velocities were integrated over inflow
the MRA-measured and the CFD2 velocity fields were or outflow areas for the four transverse planes, in CFD simula-
reduced to 0.67 and 0.66 in Figs. 5(a) and 5(b), respectively. tions made with a rigid wall assumption, the differences
In Figs. 5(a) and 5(b), the similarity values of velocity vector between inflow and outflow volumes were approximately 1%.
fields between the CFD1 and CFD2 were 0.84 and 0.94, There were, however, larger imbalances in the PC-VIPR
respectively. It is also worth noting that, consistent to Figs. MRA measured values between the ratios of the inflow
3(a) and 3(b), discrepancies between the MRA measurements and outflow volumes [35.8 6 23.4% (mean 6 standard
and CFD simulations shown in Figs. 5(a) and 5(b) also deviation) for aneurysm A and 24.2 6 17.1%, for aneurysm
occurred at the border zone between the velocity jet and the B, respectively].
low-velocity recirculation area where the velocity gradient
and shear rate are also high. Similarly, the effect of the non-
III.B. Visualization of 3D streamlines
Newtonian properties of blood also seemed important in these
border regions with high velocity gradients as shown in Figs. In Figs. 6(a) and 6(b), instantaneous streamlines at the
3 and 5. peak systole were calculated using MRA-measured and
In any aneurysm, two major flow areas can be defined: CFD-simulated velocity fields for both aneurysms. Although
flow coming into the aneurysms, i.e., the inflow into the aneu- Figs. 6(a) and 6(b) only show the streamline flow patterns
rysm and flow leaving the aneurysm, i.e., the outflow out of during the peak systole, these patterns were found to be sta-
the aneurysm (hereafter referred as to inflow and outflow for ble over the entire cardiac cycle. As would be expected,
simplicity). Using transverse planes oriented nearly perpen- since both aneurysms were geometrically similar, the gross
dicular to the flow direction [Figs 3(a) and 5(a)] we estimated, flow patterns were also quite similar. There was a jet-like
during each phase of the cardiac cycle, the size of the inflow stream oriented parallel to the upstream segment of the par-
and outflow zones (hereafter referred as to inflow or outflow ent artery immediately below the aneurysm ostium which
area) for four such transverse planes in each of the two aneur- enters the aneurysm along its left side, reaches the top of the
ysms. The averaged absolute percent area differences dome, then spins clockwise around and across the dome to
(AAPAD) between the MRA measurements and the CFD pre- finally exit from the aneurysm almost exclusively into the
dictions were calculated as follows: abs[(measured value— branch on the right side. Qualitatively, these streamline visu-
CFD prediction)=measured value]. Relative differences alizations show good correspondence between major flow
between the inflow areas were relatively large (average 13.9 features in and around both aneurysms as determined from
and 15.1% for inflow areas for aneurysms A and B, respec- the MRA measurements and the CFD calculations.
tively) between the MRA-measurements and the CFD1 pre- Two main differences were, however, noted. First, a 3D
dictions. However, the discrepancy was even larger (average streamline illustrates the trajectory a zero-mass particle takes
17.2 and 20.4% for inflow areas for aneurysms A and B, through an instantaneous flow field. Because of the presence
respectively) between the MRA-measurements and the CFD2 of MRA-measurement noise, there were fewer trajectories
predications. Our values were slightly higher but comparable calculated when the MRA-measured velocity fields were
to 12% of average differences reported by Karmonik et al.19 used as compared to both the CFD1 and CFD2 simulations.
FIG. 7. Side-by-side comparisons for TA-WSS values derived from PC-MRA-estimated and CFD. Arrows points to the likely locations where the inflow jets
impinge the walls.
IV. DISCUSSION
Our results (Figs. 2, 4, and 6) show, in both aneurysms,
gross aneurismal flow patterns obtained by the MRA-
measurements and the CFD simulations are similar. Espe-
cially, CFD models may exhibit different velocity magnitudes
than the corresponding MRA data (on average 0.1 m=s within
the aneurysm sac). However, regions of low and high blood
FIG. 8. Selected images of the estimated WSS under noise-free for three dif-
flow, location of velocity jets are in good agreement. This
ferent voxel sizes (5, 250 and 750 lm) and their estimated correlation finding is consistent with what were reported in Rayz et al.18
values. where they compared CFD-simulated velocity with in vivo
We recognize that CFD-calculated flow is not sufficient to agreement. Results, however, suggest that there are still large
serve as the gold standard21 to validate such 4D PC-MRA discrepancies in the WSS values derived from the PC-MRA
measurements. However, they do, in our opinion, serve as a velocity fields and the CFD simulations largely because of
tool for performing valid comparisons of measured flow the combination of noise and limited PC- MRA spatial reso-
parameters. For instance, by comparing to CFD-simulated lution. Similar to what had already been reported,44 we also
velocity fields, we, among others,21 found that MRA- found that similar bulk flow patterns were obtained from
measurement velocity fields resemble=capture less secondary CFD simulations with Newtonian and non-Newtonian flow
flow (rotational and nonaxial) effects and therefore are less for data investigated. Although there were subtle differences
disturbed. As a result of insufficient spatial and temporal reso- between the Newtonian and non-Newtonian CFD simula-
lution, nonaxial components of the velocity field that often tions, the bulk flow patterns and overall distributions of
have lower amplitudes might suffer more an averaging proc- blood velocities were nearly the same in two aneurysms
esses both in spatial and time domains. investigated. These results are encouraging and warrant fur-
Tremendous developments in computer techniques and ther study involving a larger number of subjects. Combining
medical imaging (e.g., 3D DSA and MRA) have greatly PC-MRA measured and CFD calculated hemodynamic val-
enhanced the ability using CFD simulations to produce ues has promise to enhance the utility of both techniques.
clinically-relevant subject-specific calculations of blood
flow.8 One advantage of the CFD simulation technique is that, ACKNOWLEDGMENTS
once the complex 3D unsteady velocity fields associated with We are grateful to our colleague Mr. Dan Consigny, B.S.,
a particular vascular abnormality are obtained, then, many he- from the Radiology Department at the University of Wiscon-
modynamic parameters such as WSS, pressure, and particle sin’s School of Medicine of Public Health and Mr. Kevin
residence time can be easily extracted, assessed and visual- Royalty, M.S., from Siemens Healthcare (USA) Inc. for their
ized using the same numerical model. Other advantages of help with animal studies. This project is funded, in part, by a
CFD include the ability to observe the differences in hemody- NIH Grant No. (2R01HL072260-05A1) and a Wallace H.
namics that occur before and after embedding of a virtual de- Coulter Foundation phase one grant awarded to the Univer-
vice intended for treatment of a particular vascular lesion, sity of Wisconsin, and a Center of Excellence Grant from
e.g., flow diverter50 and, to use simulations done at varying the Norwegian Research Council awarded to the Simula
hear rate to assess the importance of heart rate on hemody- Research Laboratory.
namic stresses.51 These advantages are not available when
flow is examined with the 4D PC-MRA technique. Because
of the complexities of the vascular tree and flowing blood APPENDIX A: SIMILARITY METRIC FOR COMPARING
CFD calculations will remain dependent on numerous VECTOR FIELDS
assumptions regarding boundary conditions, vessel wall struc- In this study, we adopted a similarity metric between vec-
ture and blood properties. Unfortunately, the potential errors tor fields similar to that was developed by Li and Shen37.
caused by the inadequacy of a subject-specific CFD model The similarity metric S accounts both the angular and magni-
based on such assumptions are still largely unknown.18,52,53 tude differences between two vectors Vi and Vj
Comparisons of patient specific CFD results to those meas-
ured by 4D PC-MRA seem well suited to improve the utility SðVi ; Vj Þ ¼ b e2a þ c em ; (A1)
of both techniques in a clinical setting.
We believe that combining 4D PC-MRA with CFD simu- where b and c are positive weighting constants, b þ c ¼ 1,
lations, will enhance the ability to analyze, interpret, and vis- and a and m represent the normalized angular and relative
ualize complex but disturbed areas of flow such as occur in magnitude differences between two velocity vector Vi and
and around intracranial aneurysms. In our future work, we Vj , respectively. The angular difference m in Eq. (A1) was
will also enforce a compromise between CFD-simulated and normalized by using the antisine function value of the angu-
MRA-measured velocity fields using certain mathematical lar difference between two vectors. In this study, we selected
frameworks.54,55 both b and c to be 0.5. It is easy to see this metric [Eq. (A1)]
In our study, two velocity fields were compared in terms is on a scale from 0 to 1, 1 being the highest. This metric
of the relative angular and magnitude differences by an em- will be estimated in a point-to-point fashion and be displayed
pirical equation37 to obtain a single summary measure of the as a spatially-registered image.
overall similarity. In future study, more rigorous measures of
similarity between vector data need to be further explored.
APPENDIX B: WSS CALCULATION
Other limitations to the present study include the small num-
ber of animals and lack of a gold standard for comparison. WSS is calculated by its definition in Newtonian fluid: the
multiplication of fluid viscosity and shear rate at the vessel
wall.56 Similar to that was used by Isoda,16 our WSS estima-
V. CONCLUSION tion method consists of: (1) determination of the vessel wall
Four-dimensional Velocity fields obtained from two and (2) approximation of the derivative of the blood velocity
experimental bifurcation aneurysms using PC-VIPR MRA function by curve fitting. Specifically, in step 1, the STL file
and subject-specific CFD simulations were in favorable representing tri-angularized vessel surface extracted from a
45
imaging of blood flow with a phase subtraction technique. in vitro and E. L. William and E. C. Harvey, “Marching cubes: A high resolution 3D
in vivo validation,” Invest. Radiol. 28, 109–115 (1993). surface construction algorithm,” SIGGRAPH Comput. Graph. 21, 163–169
34
J. Lotz, C. Meier, A. Leppert, and M. Galanski, “Cardiovascular Flow (1987).
46
Measurement with Phase-Contrast MR Imaging: Basic Facts and A. D. Augst, D. C. Barratt, A. D. Hughes, F. P. Glor, G. T. S. A. Mc, and
Implementation,” Radiographics 22, 651–671 (2002). X. Y. Xu, “Accuracy and reproducibility of CFD predicted wall shear
35
U. Kohler, I. Marshall, M. B. Robertson, Q. Long, X. Y. Xu, and P. R. stress using 3D ultrasound images,” J. Biomech. Eng. 125, 218–222
Hoskins, “MRI measurement of wall shear stress vectors in bifurcation (2003).
47
models and comparison with CFD predictions,” J. Magn. Reson. Imaging A. Caprihan, S. A. Altobelli, and E. Benitez-Read, “Flow-velocity imag-
14, 563–573 (2001). ing from linear regression of phase images with techniques for reducing
36
A. A. Valencia, A. M. Guzman, E. A. Finol, and C. H. Amon, “Blood flow eddy-current effects,” J. Magn. Reson. (1969) 90, 71–89 (1990).
48
dynamics in saccular aneurysm models of the basilar artery,” J. Biomech. A. Chernobelsky, O. Shubayev, C. R. Comeau, and S. D. Wolff, “Baseline
Eng. 128, 516–526 (2006). correction of phase contrast images improves quantification of blood flow
37
H. Li and I. F. Shen, “Similarity Measure for Vector Field Learning,” in in the great vessels,” J. Cardiovasc. Magn. Reson. 9, 681–685 (2007).
49
Advances in Neural Networks—ISNN 2006 (2006), pp. 436–441. F. B. Meyer, J. Huston III, and S. S. Riederer, “Pulsatile increases in aneu-
38
M. Markl and J. Hennig, “Phase contrast MRI with improved temporal re- rysm size determined by cine phase-contrast MR angiography,” J. Neuro-
solution by view sharing: k-space related velocity mapping properties,” surg. 78, 879–883 (1993).
50
Magn. Reson. Imaging 19, 669–676 (2001). J. R. Cebral, F. Mut, M. Raschi, E. Scrivano, R. Ceratto, P. Lylyk, and C.
39
M. Zockler, D. Stalling, and H.-C. Hege, “Interactive visualization of 3D- M. Putman, “Aneurysm rupture following treatment with flow-diverting
vector fields using illuminated stream lines,” Proceedings of the 7th con- stents: Computational hemodynamics analysis of treatment,” AJNR Am J
ference on Visualization ’96, (IEEE Computer Society Press, San Fran- Neuroradiol 32, 27–33 (2011).
51
cisco, California, 1996). J. Jiang and C. Strother, “Computational fluid dynamics simulations of in-
40
H. Gudbjartsson and S. Patz, “The Rician distribution of noisy MRI data,” tracranial aneurysms at varying heart rates: a “patient-specific” study,”
Magn. Reson. Med. 34, 910–914 (1995). J. Biomech. Eng. 131, 091001 (2009).
41 52
P. Lylyk, C. Miranda, R. Ceratto, A. Ferrario, E. Scrivano, H. R. Luna, A. S. Jin, J. Oshinski, and D. P. Giddens, “Effects of Wall Motion and Com-
L. Berez, Q. Tran, P. K. Nelson, and D. Fiorella, “Curative endovascular pliance on Flow Patterns in the Ascending Aorta,” J. Biomech. Eng. 125,
reconstruction of cerebral aneurysms with the pipeline embolization de- 347–354 (2003).
53
vice: The Buenos Aires experience,” Neurosurgery 64, 632–642; discus- J. R. Cebral, M. A. Castro, S. Appanaboyina, C. M. Putman, D. Millan,
sion 642–633; quiz N636 (2009). and A. F. Frangi, “Efficient pipeline for image-based patient-specific anal-
42
J. Bock, A. Frydrychowicz, A. F. Stalder, T. A. Bley, H. Burkhardt, ysis of cerebral aneurysm hemodynamics: technique and sensitivity,”
J. Hennig, and M. Markl, “4D phase contrast MRI at 3 T: Effect of stand- IEEE Trans. Med. Imaging 24, 457–467 (2005).
54
ard and blood-pool contrast agents on SNR, PC-MRA, and blood flow vis- M. Watanabe, R. Kikinis, and C. F. Westin, “Level set-based integration
ualization,” Magn. Reson. Med. 63, 330–338 (2008). of segmentation and computational fluid dynamics for flow correction in
43
F. M. Box, R. J. van der Geest, J. van der Grond, M. J. van Osch, A. H. phase contrast angiography,” Acad. Radiol. 10, 1416–1423 (2003).
55
Zwinderman, I. H. Palm-Meinders, J. Doornbos, G. J. Blauw, M. A. van T. Schenkel, M. Malve, M. Reik, M. Markl, B. Jung, and H. Oertel,
Buchem, and J. H. Reiber, “Reproducibility of wall shear stress assess- “MRI-based CFD analysis of flow in a human left ventricle: methodol-
ment with the paraboloid method in the internal carotid artery with veloc- ogy and application to a healthy heart,” Ann Biomed Eng 37, 503–515
ity encoded MRI in healthy young individuals,” J Magn. Reson. Imaging (2009).
56
26, 598–605 (2007). W. R. Milnor, Hemodynamics (Williams & Wilkins, Baltimore, 1982).
44 57
A. Valencia, A. Zárate, M. Gálvez, and L. Badilla, Non-Newtonian T. A. Driscoll and A. R. H. Heryudono, “Adaptive residual subsampling
Blood Flow Dynamics in a Right Internal Carotid Artery a Saccular Aneu- methods for radial basis function interpolation and collocation problems,”
rysm (John Wiley & Sons, New York, 2006). Comput. Math. Appl. 53, 927–939 (2007).