Nicolas Aristokleous1, Petter Frieberg1, Johannes Töger 1, Petru Liuba2, Pia Sjöberg1, Einar Heiberg 1, and Marcus Carlsson1
1Department of Clinical Sciences Lund, Clinical Physiology, Skåne University Hospital, Lund University, Lund, Sweden, 2Department of Clinical Sciences Lund, Pediatric Heart Center, Skåne University Hospital, Lund University, Lund, Sweden
Synopsis
Computational
fluid dynamics (CFD) modelling may help patients with heart defects by
predicting blood flow after interventions thus aiding surgical planning.
Current CFD-techniques are not clinically feasible due to long computational
times. We compared a simplified CFD technique to the state-of-the-art advanced
CFD and validated the results with Magnetic Resonance Imaging. We could show
that the simplified CFD method on a conventional laptop saved between 75%-99%
computing time compared with advanced CFD on a dedicated computation server
with comparable accuracy in results of flow distributions. This simplified CFD
method may facilitate the implementation of CFD predictions in a clinical
setting.
Introduction
Congenital heart disease (CHD) is the leading cause of
mortality from birth defects [1]. The reported prevalence of
CHD in the general population varies between 4 and 10 per 1000 live
births [2].
Patients with complex CHD may need to
be surgically transformed to a Fontan circulation with one ventricle pumping
blood to the body and the returning venous blood directly connected to the
pulmonary arteries. However, complications are frequent and life expectancy decreased.
Computational Fluid Dynamics (CFD) can be used for patient-specific hemodynamic
predictions in patients with univentricular hearts after Glenn (stage II) or
Fontan (stage III) palliation and may improve the surgical results, but the
required resources may be prohibitive for routine clinical use [3].
Therefore, our purpose is to reduce the
complexity in this process so that more centers can adopt this method in
clinical routine. The aim of the current study is therefore to compare a
simplified CFD solver on a basic laptop with an advanced CFD solver on a powerful
computation server, in terms of pulmonary flow distribution to the left
pulmonary artery (%PFD), hepatic flow distribution to the left pulmonary artery
(%HFD) and total computation time and validate the results with MRI.Methods
MRI flow
measurements from 13 patients (median age 6.7 years, range 3 to 17 years) with
Glenn (n=4) and Fontan (n=9) circulation were performed on a 1.5T Philips
Achieva scanner (Philips Healthcare, Best, The Netherlands) (n=8) or on a 1.5T Siemens Aera (Siemens Healthcare,
Erlangen, Germany) (n=4). Segmentation and modelling were performed using
Segment (Medviso AB, Lund, SE) and Creo Parametric (PTC, Boston, MA, USA). MRI parameters
were as follows: Two-dimensional phase contrast MRI (2D PC-MRI) flow
measurements were acquired using a phase velocity encoded fast field echo
sequence (TR/TE/flip angle: 5-10 ms/3-7 ms/15-20° acquired in-plane resolution
2 mm x 2 mm). Flow measurements of the superior and inferior vena cava, and the
pulmonary veins were typically velocity encoded at 80 cm/s and for the aorta at
200 cm/s.
All image
analysis was done using the freely available software Segment
(http://segment.heiberg.se).
CFD
simulations were carried out using FloEFD for Creo (Mentor Graphics, Portland,
OR, USA) and Star-CCM+ (v2019.1, Siemens PLM Software, Plano, TX, USA) as the
simplified and advanced solver, respectively. For the simplified CFD
simulations, hexagonal immersed boundary meshes of approximately 150.000
elements were constructed by FloEFD. For the advanced CFD simulations, tetrahedral
meshes of approximately 1.2×106 elements size were constructed by
ICEM CFD v19.2 (Ansys Inc., Canonsburg, PA). A systematic grid convergence
study was performed to estimate the error in numerical simulation. More details
on meshing are provided elsewhere [4]. Blood was modeled as an incompressible Newtonian fluid (ρ =
1050 kg/m3, μ = 0.004 kg/mˑs) and patient-specific flow rates from MRI were applied at each inlet. Pulmonary vascular resistance was calculated as
according to a previous study [5]. For the pulsatile simulations, results were obtained from
the 3rd cycle to avoid transient effects and the
convergence criteria were set to residual errors <10−5. The simplified CFD simulations were performed
on a Dell XPS 15 laptop with a 4-core Intel i7 processor and 16GB RAM. Advanced
steady CFD was performed on an HP Z240 Workstation with 4-core Intel(R) Xeon(R)
CPU E3-1230v5 @ 3.40GHz and 32GB RAM, and pulsatile CFD on 32 cores of a Dual
18-core Intel(R) Xeon(R) Gold 6154 CPU running @ 3GHz x 72 and 376.6 GiB RAM.
Quantification of hemodynamic metrics were computed as follows:
1. Flow distribution to the left pulmonary artery (%PFD)
%PFDLPA = (QLPA / (QLPA + QRPA)) ·100%
2.Hepatic Flow Distribution to the left pulmonary artery (%HFD)
%HFDLPA = (Q IVC to LPA / QIVC) · 100%
The methods
were compared using bias according to Bland-Altman (mean difference±SD) and linear correlation analysis. Statistical
significance was assumed for p<0.05.Results
Figure 1
illustrates the streamlines color coded with velocity magnitude, for four
representative cases calculated by an advanced and a simplified solver.
There was a strong correlation (Figure 2 Right) between the
two methods in terms of %PFD, R2=0.96) and %HFD (R2=0.94),
respectively and the bias were low (0.9±2.2 % and -1.8±6.5 % respectively,
(Figure 2 Left).
Table 1
presents the hemodynamic comparison of %PFD between simplified and advanced CFD
and the validated results with MRI and from a previously published study [6]. Furthermore, the table includes the %HFD and the total
computation time. The simplified CFD on a laptop took between 1-25% of the
computing time of the advanced CFD on a computation server.Discussion
Simplified and
advanced CFD modelling yielded similar results. This may help more centers deploy CFD for hemodynamic assessment of Glenn
and Fontan patients in clinical practice and enable results from CFD to be
presented to surgeons the same day as MRI.Conclusion
We have shown
that using a simplified CFD program on a conventional laptop can save between
75%-99% computing time compared with advanced CFD on a computation server with
comparable accuracy in results, with patient-specific MRI measurements as
reference.Acknowledgements
The work was supported by
Swedish Research Council,
Marianne and Marcus Wallenberg foundation, Swedish Heart and
Lung foundation.References
[1] Triedman JK, Newburger JW (2016) Trends
in Congenital Heart Disease. Circulation.
https://doi.org/10.1161/CIRCULATIONAHA.116.023544
[2] Marelli AJ, Mackie AS, Ionescu-Ittu R, et
al (2007) Congenital Heart Disease in the General Population. Circulation
115:163–172. https://doi.org/10.1161/circulationaha.106.627224
[3] Trusty PM, Slesnick TC, Wei ZA, et al
(2018) Fontan surgical planning: Previous accomplishments, current challenges,
and future directions. J. Cardiovasc. Transl. Res.
[4] Aristokleous N, Seimenis I, Georgiou GC,
et al (2014) Impact of head rotation on the individualized common carotid flow
and carotid bifurcation hemodynamics. IEEE J Biomed Heal Informatics 18:.
https://doi.org/10.1109/JBHI.2014.2305575
[5] Frieberg P, Sjöberg P, Revstedt J, et al
(2018) Simulation of aortopulmonary collateral flow in Fontan patients for use
in prediction of interventional outcomes. Clin Physiol Funct Imaging
38:622–629. https://doi.org/10.1111/cpf.12457
[6] Tang E, Restrepo M, Haggerty CM, et al
(2014) Geometric characterization of patient-specific total cavopulmonary
connections and its relationship to hemodynamics. JACC Cardiovasc. Imaging