MR-Based Real Time Path Planning for Cardiac
Operations with Transapical Access
Erol Yeniaras1, Nikhil V. Navkar1,2, Ahmet E. Sonmez1, Dipan J. Shah3,
Zhigang Deng2, and Nikolaos V. Tsekos1
1
Medical Robotics Lab
Computer Graphics and Interactive Media Lab,
Department of Computer Science, University of Houston, Houston, TX 77004, USA
3
Methodist DeBakey Heart & Vascular Center, Houston, TX 77030, USA
{yeniaras,nvnavkar,aesonmez,ntsekos,zdeng}@cs.uh.edu,
djshah@tmhs.org
2
Abstract. Minimally invasive surgeries (MIS) have been perpetually evolving
due to their potential high impact on improving patient management and overall
cost effectiveness. Currently, MIS are further strengthened by the incorporation
of magnetic resonance imaging (MRI) for amended visualization and high
precision. Motivated by the fact that real-time MRI is emerging as a feasible
modality especially for guiding interventions and surgeries in the beating heart;
in this paper we introduce a real-time path planning algorithm for intracardiac
procedures. Our approach creates a volumetric safety zone inside a beating
heart and updates it on-the-fly using real-time MRI during the deployment of a
robotic device. In order to prove the concept and assess the feasibility of the
introduced method, a realistic operational scenario of transapical aortic valve
replacement in a beating heart is chosen as the virtual case study.
Keywords: Real time MRI, Image Guided Surgeries, and Beating Heart.
1 Introduction
Contemporary improvements in the field of medical robotics, and a series of
successful clinical applications, have led to the emergence of interventional robots by
the clinical and technical community. The inclusion of real-time image guidance in
robotic-assisted interventions may further elevate the field by offering improved
information-rich visualization, as well as option of assessing the tissue before, during
and after a procedure [1]. Considering the challenges associated with the continuous
cardiac motion, real-time image guidance can provide a number of benefits especially
for robot-assisted surgeries in a beating heart [2, 3].
Among the emerging clinical paradigms in the area of minimally invasive
procedures in a beating heart is magnetic resonance imaging (MRI) guided prosthetic
aortic valve implantation via transapical access; a procedure that has been
demonstrated manually [4] and with robotic assistance [2]. For such off-pump
procedures, an important factor is the efficacy of using images to assess the dynamic
environment during operation. One effective method is the extraction of dynamic
G. Fichtinger, A. Martel, and T. Peters (Eds.): MICCAI 2011, Part I, LNCS 6891, pp. 25–32, 2011.
© Springer-Verlag Berlin Heidelberg 2011
26
E. Yeniaras et al.
access corridors from images [5]. MRI was selected by numerous investigators for its
high soft-tissue contrast, absence of ionizing radiation, and inherent robustness and
versatility [4, 5]. With the current state-of-the-art MRI, dynamic images can be
collected at a rate of 40-50 ms/image. Practically, only a single imaging plane can be
collected with such a high, for MRI standards, rate. Thus the question is how a 3D
corridor inside the beating heart can be updated using a single plane.
To address this issue, in this work, we evaluate a method that combines
preoperative multislice dynamic MRI (i.e., cine MRI) with single-slice real-time MRI
to update an access corridor from the apex to the aortic annulus. Cine MRI is used to
generate a preoperative 3D corridor in the left ventricle (LV) which is updated on-thefly by registering it onto the intraoperative real-time MR images. The method was
assessed for accuracy of the corridor registration and simulated for the deployment of
a virtual robot for 12 subjects.
2 Methodology
In a typical transapical scenario, the robotic manipulator enters LV via a trocar
affixed at the apex, T(t); and maneuvers toward the targeted center of aortic annulus,
A(t) as depicted in Fig. 1. Preliminary analyses of the cine data from 12 healthy
volunteers indicated that LV can be transversed with a cylindrical corridor and the
deployment path from T(t) to A(t) is not a straight line. For a precise orthogonal
approach to aortic annulus, a dynamic bending point, B (t), near the base of LV is
needed. The characteristics of the corridor, as well as the aortic annulus diameter,
coronary ostial anatomy and apical entrance point were determined from cine MRI,
whereas real-time MRI was used to update the corridor and follow the operation.
Fig. 1. A long axis MR image shows a typical transapical approach to a beating heart at time t
2.1 Preoperative Planning
For any given cardiac phase (i.e., at time frame t), the access corridor CR(t) is defined
as the largest cylindrical volume that lies along LV from the apex toward the base of
the heart. Then, an appropriate-sized surgical device should be able to deploy inside
CR(t) from T(t) to A(t) without colliding or injuring the endocardium, papillary
muscles or chordae tendinae. The cylindrical corridor was generated from cine
MR-Based Real Time Path Planning for Cardiac Operations with Transapical Access
27
datasets (n=12) collected with a true fast imaging, steady-state precession pulse
sequences (TrueFISP) with TR/TE = 2.3 ms/1.4 ms, flip angle = 80o, slice thickness =
6.0 mm, and acquisition matrix = 224x256. Each dataset included 19 short axes (SA)
and 5 long axes (LA) slices, capturing heart motion with 25 frames over a complete
cardiac cycle. In order to determine the transient positions of the corridor and the
deployment points (i.e., T(t), B(t), and A(t)), SA and LA images were segmented
using a region-growing algorithm based on Insight Toolkit (ITK) filters to extract the
apex, LV and aortic annulus. As shown in Fig. 2(a) and 2(b), to realistically model the
corridor, papillary muscles and chordae tendinae were also considered. For every
single heart phase t (t=1 to 25), CR(t) is constructed as follows:
1. The SA slices with visible blood pool are determined by checking the inside
surface areas of LV segmentations, i.e., selecting the non-zero ones as in Fig. 2(b);
2. LV segmentation contours of these SA slices are projected onto a virtual plane
along their common orthogonal axis to find their intersection polygon (INPT) by
2D polygon clipping. This projection is based on the fact that SA slices are parallel
to each other and collected with the same field of view;
3. The largest circle (ST) that fits into INPT is determined. Since a circle can be
created with a center point and a radius, let`s define CT as the Center of ST and, RT
as the Radius of ST. Then the centroid of INPT is chosen as the center of ST, shown
in Eq. (1) where N is the number of edges of INPT:
C T (x, y) =
1
N
N INP (x , y )
∑ i=1
T i
i
(1)
Thus, RT can be safely defined as “the minimum distance from CT to the edges of
INPT” as formulated in Eq. (2):
R T = min1...N (|| CT (x, y) - INPT (x i , y i ) ||)
(2)
4. Finally, ST is stretched from the apex to the base of the heart, through all SA slices
to generate a circular straight access corridor as shown in Fig. 2(c).
Fig. 2. Selected segmentations of LA (a), SA (b) slices from diastole and systole phases highlight
LV blood pool and boundaries, and a sample corridor with endocardial contours (c)
Since a unique 3D access corridor is generated for every single heart phase, CR(t)
is a 4D dynamic entity defined for a full heart cycle consisting of 25 time frames. The
final step of preoperative planning is to set the initial positions of deployment points
as follows: (1) T(t) is selected manually by a cardiovascular surgeon as the tip of the
LV (the apex) on the central LA slice (and verified computationally that it belongs to
28
E. Yeniaras et al.
a SA slice that shows only myocardium with no LV blood pool); (2) A(t) is
determined from the segmentation contours of two LA and one SA slices that include
the aortic valve annulus at the level of aortic valve leaflets. Note that, the initial
selection of the exact target point is also made by the surgeon (or cardiologist); (3)
B(t) is assigned as the intersection of the aortic annulus midline and the top face of
the safe corridor initially (as also depicted in Fig. 1).
2.2 Intraoperative Guidance
In this phase, the access corridor and deployment points, generated preoperatively for
a complete heart cycle, are updated on-the-fly. Intraoperative guidance is based on
continuous real-time acquisitions of a central LA slice taken from the very same
healthy volunteers (n=12) but this time spanning 30 full heart-beats with breathing
(TrueFISP parallel imaging with effective repetition time = 48.4 ms, TE=0.95 ms,
alpha=65o, slice thickness=6.0 mm, acquisition matrix = 160x66). After
comprehensive analyses of MRI data and different imaging planes, we observed that
heart mainly translates along and rotates around its long axis without significant outof-plane motion with respect to the real-time LA slice under consideration. Moreover,
breathing adds an extra vertical motility with respect to the MR table, as denoted with
R in Fig. 3, which can also be followed on the same LA slice effectively. To the end,
we choose this single real-time LA slice for intraoperative guidance.
The most challenging task is to register the preoperative corridor onto the real-time
LA slice on-the-fly during the operation. This is done via two major steps: (1)
Determine the heart phase in which the real-time slice was collected (and thus match
it with the corresponding corridor); (2) Adjust the position and orientation of this
corridor to account for heart motion due to respiration, arrhythmias, etc.
Fig. 3. Illustration of LA segmentations depicting the blood-pool area, R: respiratory motion;
AJ: Apex points; CJ: Midpoints; LJ: Vertical lines, VJ: CJ Æ AJ LV directional vectors for nonreal-time and real-time LA slices respectively; J = 1 and 2
First, LV is segmented in the real-time LA image using the same region-growing
algorithm and area of the blood-pool is calculated. Then, this area is compared to each
of the 25 preoperative LV areas of the same LA slice that is extracted with the same
parameters, to find the closest. This comparison is done by a conservative approach,
i.e., selecting the closest one with the minimum size to guarantee the aforementioned
safety criterion. Once it is found, the corridor corresponding to this heart phase is
selected as the one to be registered to the real-time LA slice. Fig. 3 shows a
segmented cine (non real-time) LA slice and its real-time counterpart respectively.
MR-Based Real Time Path Planning for Cardiac Operations with Transapical Access
29
Next, the corridor needs to be correctly positioned onto the real-time slice. To achieve
that, a vertical line (L2) crossing the base of the heart is defined. This line is defined
such that segmentation includes most of LV blood pool but not the aortic valve. Then,
the intersection points of this line to the endocardial wall are determined and their
midpoint (C2) is calculated as shown in Fig. 3. The same operation is performed for
the corresponding preoperative image to compute L1 and C1, and thus we can compare
C1 with C2. It should be noted that, A1 and A2 denote the apex points, while V1 and V2
are directional vectors for the LVs.
Then, the relative displacement between C1 and C2 is calculated and applied to the
top-center point of the corridor, P11, to find P21 as shown in Fig.4 (a). To adjust the
orientation, the angle between V1 and V2 is calculated and applied to the direction
vector of the preoperative corridor V1C to find V2C. Finally, the resultant corridor is
registered to its new position as depicted in Fig.4 (b) and it is ready for the robotic
manipulator. The above process is highly efficient and real-time (i.e., all the
computing takes less than 48.4 ms, which is less then effective repetition time).
Fig. 4. (a) Corridor registration using vectors; and (b) the registered trajectory on LA slices
Finally, initial deployment points are tracked during the (virtual) operation using
an efficient 2D fast-tissue-tracking algorithm [6]. In an in vivo scenario, the target and
the bending points can be tracked with such an algorithm whereas appropriate MRI
methods should provide the position of the apex point, T(t) in real-time (e.g.,
miniature RF coil beacons on the trocar [7]).
2.3 Experimental Studies
The proposed method was tested for the registration accuracy of the corridor onto the
real-time MRI. Specifically, the registered corridors were compared to the groundtruths that were created by manually locating them onto their correct positions in all
the real-time images for 30 full heart cycles (n=12). Fig. 5 shows three examples of
corridor positioning errors.
We also simulated the deployment of a six degrees-of-freedom (DOF) RRPRRP
(R: Rotational joint, P: Prismatic joint) virtual robot inside the registered corridor to
assess the possible collisions with the endocardium or the aortic walls. The proximal
end of the robot was assumed to be attached to T(t) with a continuous actuation to
follow the motion of the apex and Denavit-Hartenberg convention was used in inverse
30
E. Yeniaras et al.
kinematics to deploy the robot along the corridor. The inputs for robot control were
the dynamic coordinates of deployment points as well as the initial conditions
specified by the operator, e.g., the time frame when the robot initiates its
maneuvering, and whether and for how long it may hold a certain position along its
path. Motion of the virtual robot entails the following three steps: deployment of the
first link from the apex toward B(t), extension of the second link toward A(t), and the
holding of the position. Since the device must stay inside CR(t), turn can occur before
or after the initial B(t) provided that the first section remains in the corridor and the
tip heads to the center of aortic annulus. During the maneuvering process, the robot
control supplies the values of the updated DOFs for each time instance. Robot
deployment was visually simulated along with the surgical field using OpenGL.
Fig. 5. Registered points: P21, P22 and ground-truth: PG1, PG2 for three different samples
3 Results and Discussion
Application of the method on all preoperative cine sets (n=12) demonstrated that a
dynamic cylindrical corridor can be defined and tracked for safe deployment inside
the LV of the beating heart. The average base diameter of this corridor for 12 subjects
was 9 mm in systole and 22 mm in diastole. In regard to the code for on-the-fly
processing of the real-time MRI, corridor selection was practically error-free as a
reflection of the conservative approach in selecting them (i.e., the minimal size). The
average distal error for the starting point of the corridor (P21) was 1.3 mm while it was
2.0 mm for the ending point (P22) as formulated in Eq. (3) and listed in Table 1.
εS =|| P21 - PG1 || ; ε Ε =|| P22 − PG 2 ||
(3)
This difference is mainly caused by the fact that the ending point is nearer to the apex
which is the most dynamic point of a heart, and the starting point is nearer to the
aortic annulus which undergoes a relatively slower motion. In order to guarantee a
safe deployment, let’s assume that the registration errors take the maximum value of
2.4 mm for both the starting and ending points in either direction (i.e., total error of
4.8 mm at each side), then omitting the outer parts, the diameter of the corridor drops
to 4.2 mm in systole and 17.2 mm in diastole. Since the robot always follows the
centerline of the corridor, any device with diameter less than 4.2 mm can be deployed
safely within such a corridor.
MR-Based Real Time Path Planning for Cardiac Operations with Transapical Access
31
For all the 12 subjects, the simulated deployment of the virtual robot through
dynamically registering the corridors with real-time MRI showed no collision with the
inner boundaries of LV. The tested diameter of the robotic link was 4 mm. As
depicted in Fig. 5 after an initial user-defined idling period, there are two deployment
phases.
Table 1. The registration errors for the updated corridors on the real-time slices (n=12)
Subject/Error (mm)
1
2
3
4
5
6
7
8
9
10
11
12
End point (εE)
Max
Min
1.9
1.6
2.1
1.8
2.4
2.2
2.0
1.7
2.1
1.9
2.1
1.8
2.2
1.8
2.4
1.9
2.0
1.7
2.2
1.7
2.2
2.0
2.3
2.0
Average
1.8
1.9
2.3
1.8
2.0
1.9
2.0
2.3
1.8
1.9
2.1
2.2
Start point (εS)
Max
Min
1.6
1.2
1.4
1.1
1.5
1.1
1.7
1.3
1.5
1.2
1.3
1.0
1.4
1.1
1.5
1.2
1.6
1.1
1.4
1.0
1.2
1.0
1.5
1.2
Average
1.5
1.2
1.2
1.6
1.4
1.2
1.3
1.3
1.4
1.2
1.1
1.3
Fig. 5. Deployment is simulated (the images have the same field of view): (a) and (b) spans
Phase I. Phase II starts at (c) and ends in (d).
The Phase I is the extension of the first link from the apical entrance toward the
bending point, which entails the actuation of the first two rotational DOFs (i.e., R1
and R2) to maintain the deployed part inside the corridor. Once the distal end of the
first link reaches near B(t), the Phase II starts with the extension of the second link
toward the targeted aortic annulus. Concurrently, the third and fourth rotational DOFs
(i.e., R3 and R4) are also actuated to maintain the second link along the aortic annulus
midline. After reaching the target, the robot maneuvers to hold the position: the base
of the robot at T(t), the robot inside the corridor, and the tip of the second link at A(t).
32
E. Yeniaras et al.
It is noteworthy that the presented computational core is connected to the MR
scanner via a local area network (LAN) for a two-way communication: (i) real-time
data transfer from the scanner to the computational core (thereby achieving a
refreshing rate of 20 fps) and (ii) in reverse on-the-fly adjustment of the imaging
parameters from the control module of the core, as we have demonstrated before [8].
4 Conclusions
This paper introduces a novel computational methodology for planning and
performing real-time MRI-guided interventions in a beating heart. In all our studies it
was able to generate a dynamic cylindrical corridor and update it with real-time
single-plane MRI. Future work includes testing it on an actuated cardiac phantom [9]
and automatically tracking the aortic annulus centerline on LA real-time MRI.
Acknowledgments. Supported by the National Science Foundation (NSF) award CPS0932272. All opinions, findings, conclusions or recommendations in this work are
those of the authors and do not necessarily reflect the views of our sponsors.
References
1. Jolesz, F.A.: Future perspectives for intraoperative MRI. Neurosurg. Clin. N. Am. 16, 201–
213 (2005)
2. Li, M., Mazilu, D., Wood, B.J., Horvath, K.A., Kapoor, A.: A robotic assistant system for
cardiac interventions under MRI guidance. In: Proc. SPIE 7625, pp. 76252X (2010)
3. McRae, M.E., Rodger, M., Bailey, B.A.: Transcatheter and transapical aortic valve
replacement. Crit. Care Nurse 29, 22–37 (2009)
4. Horvath, K.A., Mazilu, D., Guttman, M., Zetts, A., Hunt, T., Li, M.: Midterm results of
transapical aortic valve replacement via real-time magnetic resonance imaging guidance. J.
Thorac. Cardiovasc. Surg. 139, 424–430 (2010)
5. Yeniaras, E., Deng, Z., Davies, M., Syed, M.A., Tsekos, N.V.: A Novel Virtual Reality
Environment for Preoperative Planning and Simulation of Image Guided Intracardiac
Surgeries with Robotic Manipulators. Stud. Health Technol. Inform. 163, 716–722 (2011)
6. Zhou, Y., Yeniaras, E., Tsiamyrtzis, P., Tsekos, N., Pavlidis, I.: Collaborative Tracking for
MRI-Guided Robotic Intervention on the Beating Heart. In: Jiang, T., Navab, N., Pluim,
J.P.W., Viergever, M.A. (eds.) MICCAI 2010. LNCS, vol. 6363, pp. 351–358. Springer,
Heidelberg (2010)
7. Elgort, D.R., Wong, E.Y., Hillenbrand, C.M., Wacker, F.K., Lewin, J.S., Duerk, J.L.: Realtime catheter tracking and adaptive imaging. J. Magn. Reson. Imaging 18, 621–626 (2003)
8. Christoforou, E., Akbudak, E., Ozcan, A., Karanikolas, M., Tsekos, N.V.: Performance of
interventions with manipulator-driven real-time MR guidance: implementation and initial in
vitro tests. Magn. Reson. Imaging 25, 69–77 (2007)
9. Sternberg, N.V., Hedayati, Y., Yeniaras, E., Christoforou, E., Tsekos, N.V.: Design of an
actuated phantom to mimic the motion of cardiac landmarks for the study of image-guided
intracardiac interventions. In: ROBIO, pp. 856–861 (2010)