Download Anatomically Realistic Patient-Specific Surgical Planning of

Anatomically Realistic Patient-Specific Surgical Planning of
Complex Congenital Heart Defects Using MRI and CFD
Kartik S. Sundareswaran1, Diane de Zelicourt1, Kerem Pekkan1, Gopinath Jayaprakash1, David Kim1, Brian
Whited2, Jarek Rossignac2, Mark A. Fogel3, Kirk R. Kanter4, and Ajit P. Yoganathan1
1.) Wallace H. Coulter Department of Biomedical Engineering, Georgia Tech, Atlanta, GA
2.) College of Computing, Georgia Tech, Atlanta, GA
3.) Children’s Hospital of Philadelphia, Atlanta, GA
4.) Pediatric Cardiothoracic Surgery, Children’s Healthcare of Atlanta, Atlanta, GA
Abstract— Single ventricle congenital heart defects, which
are characterized by cyanotic mixing between the oxygenated
and de-oxygenated blood, afflict 2 per every 1000 live births.
These defects are surgically treated by connecting the superior
and inferior vena cava to the pulmonary arteries. However,
such a configuration (also known as the total cavopulmonary
connection), results in high energy losses and therefore the
optimization of this connection prior to the surgery could
significantly improve post-operative performance. In this
paper, a surgical planning framework is proposed. It is
exemplified on a patient with pre and post surgical MRI data.
A pediatric surgeon performed a “virtual surgery” on the
reconstruction of the patient’s anatomy prior to the actual
surgery. Post-operative hemodynamics in the virtually designed
post-surgical anatomy and in the actual one are computed
using computational fluid dynamics and compared to each
other. This framework provides the surgeon to envision
numerous scenarios of possible surgical options, and
accordingly predict the post operative hemodynamics.
Index Terms – Surgical Planning, Magnetic Resonance
Imaging, Computational Fluid Dynamics
S
side of the heart. This pressure build up may be a possible
explanation for the complications typically observed in these
patients including congestive heart failure, liver dysfunction,
protein losing enteropathy, limited exercise capacity,
systemic venous hypertension, hepatic and pulmonary
congestion, valvular and myocardial dysfunction. Among
the parameters that come into play, one over which the
surgeons have some control is the design of the surgically
created connection. Significant work has been done in
understanding the fluid dynamics of the TCPC, underscoring
the tight relationship between design and associated pressure
drops and energy dissipation. However, no work to date has
been done towards a systematic surgical planning of the
TCPC, trying to optimize its design for every single patient,
in an effort to reduce pressure losses and improve their longterm outcome.
I. INTRODUCTION
ventricle congenital heart defects afflict 2 per
every 1000 live births. These defects are characterized
by the mixing between the oxygenated blood from the
systemic circulation and the de-oxygenated blood from the
pulmonary circulation, leading to acute cyanosis. “Fontan
repairs”, which designate the current surgical procedure of
choice, are performed in three stages with the anastomosis
of the superior (SVC) and inferior (IVC) vena cava onto the
pulmonary arteries in the 2nd and 3rd stage, respectively (see
Fig.1 and Fig.2). The resultant vascular anatomy is called
the total cavopulmonary connection (TCPC). It establishes a
complete bypass of the right side of the heart thus restoring
the vital separation between pulmonary and systemic
circuits. Unfortunately, these repairs are palliative and not
curative. Survivors often require a lifetime of intensive
medical attention. Clinicians report that over 50% of their
time must be devoted to the 20% fraction of their patients
having this complex cardiovascular physiology. Due to the
absence of the right ventricular pump, Fontan patients are
reported with higher pressures than normal in the systemic
return so as to provide the pressure head necessary to drive
the flow through the pulmonary circuit and back to the left
INGLE
a.)
b.)
Fig 1. Anatomic reconstruction of a stand alone preFontan (a) and with all the surrounding cardiac
structures included (b).
A large anatomic database of patient-specific TCPC
anatomies was compiled in our laboratory. Through a global
analysis of the main geometrical parameters and detailed
analysis of selected TCPC geometries1-5, several parameters
have been identified that can be used towards the
optimization such as the offsets between the SVC and the
IVC3, 6, size of the IVC baffle2, and the type of connection
from the IVC to the pulmonary arteries (extracardiac or
intraatrial)7. Based on this knowledge, we established a
surgical planning methodology that allows the surgeons to
envision different surgical scenarios for the 3rd and final
stage of the TCPC procedure and assess their hemodynamic
performances.
This paper describes the surgical planning methodology
in detail, taking the example of a patient with known pre and
post-Fontan cardiac anatomy. Pre- and post-cardiac
anatomies were reconstructed from magnetic resonance
images (MRI) using an in-house segmentation and
reconstruction scheme. A pediatric cardiac surgeon was
blinded to the actual post-operative anatomy and asked to
virtually connect the IVC to the pulmonary arteries.
Computational fluid dynamic (CFD) simulations were
conducted in the virtually designed and actual post-operative
anatomy and the computed energy losses were used as a
metric for comparing the performance of both designs.
Fig 2. The actual post-Fontan geometry of the patient
based on post-operative MRI
B. Surgical Planning and Free Form Anatomy
Deformation
We have previously developed a surgical planning
environment called SURGEM12, which provides intuitive
operations to edit complex vascular anatomies as needed for
our surgical planning purposes. The real-time shape editing
capabilities of SURGEM are based on a mathematical model
of free-form deformations (FFD) that are weighted averages
of pose-interpolating screw motions13, 14, originally
developed for editing motions that interpolated userspecified position and orientation constraints. This in-house
mathematical tool is integrated with a new Human-Shape
Interaction (HSI) methodology that uses two commercially
available six degrees of freedom haptic devices, to control
the parameters of the FFD by simple and natural gestures of
both hands. As a result, SURGEM supports the intuitive
manipulation of shape with both hands through natural
gestures that grab, move, bend, and twist the desired portion
of the shape.
a.)
b.)
II. METHODOLOGY
A. Anatomic Reconstruction and Model Generation
A stack of axial MRI slices was acquired in a patient who
had undergone the second stage Bidirectional-Glenn
procedure and was scheduled for the third stage Fontan
operation. The anatomy of the bidirectional Glenn
connection was segmented and reconstructed using our inhouse bouncing ball segmentation algorithm8, 9. Because it
does not depict anatomic constraints that a surgeon
experiences in the operating room, solely segmenting the 2nd
stage anatomy by itself is insufficient for surgical planning
purposes. Surrounding cardiovascular structures were thus
segmented as well including heart, pulmonary veins, aorta,
and IVC stump, which serve as landmarks for the surgeon to
connect the artificial conduit. All structures are segmented
individually and merged together using Raindrop Geomagic
Studio (Triangle Park, North Carolina) to obtain a
representation of the single ventricle anatomy (Fig 1). This
combined reconstruction was then used as an input to an
anatomy modifying tool10, 11 that was used to identify the
best spatial arrangement of the IVC conduit given the
cardiac anatomy of the patient.
c.)
d.)
Fig. 3: The process of surgical planning. a.) and b.) Two
different views of the artificial conduit connected to the preFontan geometry using the anatomical constraints. c.) The
heart, pulmonary veins, and the aorta, subtracted out of the
reconstruction. d.) The stitched anatomical model ready for
CFD.
The complete cardiovascular anatomic reconstruction of
the single ventricle anatomy was imported into the tool. A
pediatric cardiac surgeon was blinded to the actual postoperative anatomy and was asked to connect the IVC from
the stump included in the segmentation to the pulmonary
arteries given the anatomic space constraints (Fig. 3b). The
heart, pulmonary veins, and the aorta were then subtracted
from the anatomy, and only the simulated TCPC was
retained (Fig. 3c). The artificial TCPC conduit was then
stitched to the PAs using Raindrop Geomagic (Triangle
Park, North Carolina) (Fig. 3d), and this geometry was used
for performing the computational fluid dynamic (CFD)
simulations in order to compare the hemodynamics of the
simulated surgical planning model with the actual Fontan
model.
C. Computational Fluid Dynamics
The simulated anatomic lumen surfaces and the actual
post-Fontan anatomic lumen surfaces were exported to
GAMBIT (Fluent Inc., Lebanon, NH) for mesh generation.
Mesh independency was achieved at around 600,000 4-node
tetrahedral elements. Grid verification runs were done
previously for similar anatomies in previous studies4, 5. For
both the models, steady-state flow fields were computed
using the parallelized segregated finite-element CFD solver
FIDAP (Fluent Inc., Lebanon, NH).
The pressure projection algorithm both with the standard
first-order and stream-wise upwinding scheme was used,
which also demonstrated good agreement with the in vitro
experiments in our previous studies with more complex
anatomical models. In order to prescribe in vivo flow rates in
the CFD simulations, phase contrast MRI (PC MRI) data
were acquired concurrently to the stack of images used for
the anatomical reconstruction. Accordingly, the preoperative anatomy was run using the mean pre-operative in
vivo flow rates of 0.7, 0.28, and 0.42 L/Min at the SVC,
LPA and RPA, respectively, while both the actual and the
virtual post-operative anatomies were run using the mean in
vivo post-operative flow rates of 1.05, 0.79, 0.86, 0.73
L/min at the IVC, SVC, LPA and RPA, respectively.
a.)
b.)
Fig. 4: Stream traces of the flow fields obtained from the
CFD results of the pre-Fontan model. a.) vector fields
along a plane of the mode; b.) streamlines color-coded by
velocity magnitude
III. RESULTS
Shown in Fig.1 are the reconstructions of the original
bidirectional Glenn (pre-Fontan) model as well as the
surrounding anatomic structures that were used as anatomic
landmarks for performing the surgical planning. Fig 2 shows
the 3D anatomic reconstruction of the actual post-operative
TCPC anatomy. Fig. 3 shows a step by step demonstration
of the artificial baffle being connected onto the bidirectional
Glenn.
Fig 4 and 5 show sample results from the CFD
simulations that were conducted in the pre- and postoperative anatomies. The power losses were of 0.58 mW for
the actual post-Fontan anatomy and 1.36 mW for the
virtually designed one.
a.)
b.)
c.)
d.)
Fig. 5: CFD results depicting a comparison of flow fields
between the TCPC obtained using the presented surgical
planning methodology (a-b), and the actual post-Fontan
anatomy. As can be seen there are differences between the
flow fields between the modified TCPC and the actual
TCPC. The stream traces on the left are color coded
according to the vessel from where they were released: red
corresponds to the IVC, while blue corresponds to the
SVC. The stream traces on the right are color coded
according to the velocity magnitude.
The total flow rate through the connection was of
0.7L/min pre-operatively, and 1.84L/min post-operatively.
This increase was mainly due to the IVC flow being brought
into the connection. The actual post-operative TCPC is an
extracardiac baffle with a mean IVC diameter of 1.73 cm
and hardly any IVC-SVC offset. On the opposite, the
surgeon who performed the virtual surgery created an extracardiac connection using a baffle of size 1.4 cm in diameter
and slightly offset the IVC baffle towards the RPA. Both
anatomies display helical flow patterns in the PAs and fairly
streamlined flow in the SVC. Rather uncommonly, the in
vivo flow data for that patient revealed a pulmonary flow
ratio of 56/44 LPA/RPA that favored left lung perfusion.
Due to the IVC-SVC offset, the IVC flow of the virtual
anatomy impinged on the superior RPA wall before being
redistributed to both lungs, while the SVC flow solely went
to the RPA. In the actual post-Fontan anatomy, SVC and
IVC streams collided head on, mixed and were then
redistributed to both lungs.
IV. DISCUSSION
In his virtual design, the surgeon chose to include an
offset, which had previously been shown to be beneficial in
terms of power losses3. This however is not the case here
and the energy losses computed in the virtual extra-cardiac
anatomy (1.36mW) were actually higher than those
computed for the actual post-operative anatomy (0.58mW).
The primary reason for this increased energy dissipation is
the smaller sized baffle used in the virtual design when
compared to the actual one. Energy losses in the virtual
design could have been lowered further by increasing the
baffle size. However, it has also been shown that important
mismatches between the baffle size and the diameters of the
connecting vessels yielded detrimental flow features such as
regions of flow stagnation or recirculations.
For a proper surgical-planning scenario, the surgeon
should thus have tested-out several baffle sizes and used the
results of the corresponding CFD simulations to decide on
which one was the optimal, as there is a tradeoff between
reduced power losses and increased adverse flow features.
In addition it has been previously shown that small
differences under resting conditions may actually have a
large impact in exercise situation when increased cardiacoutput would result in increased energy losses.
V. CONCLUSION AND FUTURE WORK
A novel surgical planning methodology applicable for
children born with congenital heart defects is presented.
This methodology allows the surgeons to envision different
surgical scenarios and test their hemodynamic efficiency. If
implemented clinically, this tool would be the perfect
platform for surgeons to test implementation techniques they
do not know, or even visualize procedures envisioned by
others. Work is currently underway for studying the effects
of conduit size, angle, and curvature in addition to semiautomatic optimization strategies that could provide the
most efficient surgical connection prior to the surgery itself.
VI. ACKNOWLEDGMENTS
This study was funded by the National Heart Lung and
Blood Institute, grant HL67622.
VII. REFERENCES
1. de Zelicourt DA, Pekkan, K.,Parks, J.,Kanter, K.,Fogel,
M.,Yoganathan, A. P. Flow study of an extracardiac
connection with persistent left superior vena cava. J Thorac
Cardiovasc Surg 2006; 131(4):785-91.
2. de Zelicourt DA, Pekkan K, Wills L, et al. In vitro flow
analysis of a patient-specific intraatrial total cavopulmonary
connection. Ann Thorac Surg 2005; 79(6):2094-102.
3. Ensley AE, Lynch P, Chatzimavroudis GP, et al. Toward
designing the optimal total cavopulmonary connection: an in
vitro study. Ann Thorac Surg 1999; 68(4):1384-90.
4. Pekkan K, de Zelicourt D, Ge L, et al. Physics-driven
CFD modeling of complex anatomical cardiovascular flowsa TCPC case study. Ann Biomed Eng 2005; 33(3):284-300.
5. Pekkan K, Kitajima HD, de Zelicourt D, et al. Total
cavopulmonary connection flow with functional left
pulmonary artery stenosis: angioplasty and fenestration in
vitro. Circulation 2005; 112(21):3264-71.
6. Ensley AE, Ramuzat A, Healy TM, et al. Fluid mechanic
assessment of the total cavopulmonary connection using
magnetic resonance phase velocity mapping and digital
particle image velocimetry. Ann Biomed Eng 2000;
28(10):1172-83.
7. Sundareswaran K.S. Fogel M.A., Pekkan K.P., Kitajima
H.D., Parks W.J., Sharma S., Yoganathan A.P. Viscous
Dissipation Power Loss of the Total Cavopulmonary
Connection Evaluated Using Phase Contrast Magnetic
Resonance Imaging. Proceedings of the American Heart
Association Annual Scientific Session 2006 2006.
8. Frakes DH, Conrad CP, Healy TM, et al. Application of
an adaptive control grid interpolation technique to
morphological vascular reconstruction. IEEE Trans Biomed
Eng 2003; 50(2):197-206.
9. Frakes DH, Smith MJ, Parks J, et al. New techniques for
the reconstruction of complex vascular anatomies from MRI
images. J Cardiovasc Magn Reson 2005; 7(2):425-32.
10. Pekkan K. SD, Parks W.J., Kitajima H., Salee D.,
Fogel M., Yoganathan A.P. Pre-Fontan surgery
computational fluid dynamic analysis of three Glenn stage
anatomies, Effects of innominate vein and upper-lobe RPA
branch. The American Society of Mechanical Engineering
Conference. Vail, CO, 2005.
11. Pekkan K. ZD, Sorensen D, Kitajima H, Yoganathan
A.P. Surgical Planning of the Total Cavopulmonary
Connection Using MRI, Computational and Experimental
Fluid Mechanics. 3rd European Medical and Biological
Engineering Conference, EMBEC Prague, Czech Republic,
2005.
12. Rossignac J, Pekkan K, Whited B, Kanter K,
Yoganathan A , Proceedings of . Surgem: Next Generation
CAD tools targeting anatomical complexity for patientspecific surgical planning.
ASME-Bio2006 Summer
Bioengineering Conference. Florida, 2006.
13. Rossignac J. Compressed Piecewise Circular
Approximation of 3D Curves. Computer-Aided Design
2003; 35(6):533-547.
14. Rossignac. J. “Dynapack: Space-Time compression of
the 3D animations of triangle meshes with fixed
connectivity”,.
Tech Report GIT-GVU-03-08.ACM
Symposium on Computer Animation (SCA), 2003.