Three-dimensional tracking of cardiac catheters using an inverse geometry
x-ray fluoroscopy system
Michael A. Speidela?
Department of Medical Physics, University of Wisconsin–Madison, Madison, Wisconsin 53705
Michael T. Tomkowiak
Department of Biomedical Engineering, University of Wisconsin–Madison, Madison, Wisconsin 53705
Amish N. Raval
Department of Medicine, University of Wisconsin–Madison, Madison, Wisconsin 53792
Michael S. Van Lysel
Department of Medicine and Department of Medical Physics, University of Wisconsin–Madison,
Madison, Wisconsin 53792
?Received 22 June 2010; revised 21 October 2010; accepted for publication 23 October 2010;
published 23 November 2010?
Purpose: Scanning beam digital x-ray ?SBDX? is an inverse geometry fluoroscopic system with
high dose efficiency and the ability to perform continuous real-time tomosynthesis at multiple
planes. This study describes a tomosynthesis-based method for 3D tracking of high-contrast objects
and present the first experimental investigation of cardiac catheter tracking using a prototype SBDX
Methods: The 3D tracking algorithm utilizes the stack of regularly spaced tomosynthetic planes
that are generated by SBDX after each frame period ?15 frames/s?. Gradient-filtered versions of the
image planes are generated, the filtered images are segmented into object regions, and then a 3D
coordinate is calculated for each object region. Two phantom studies of tracking performance were
conducted. In the first study, an ablation catheter in a chest phantom was imaged as it was pulled
along a 3D trajectory defined by a catheter sheath ?10, 25, and 50 mm/s pullback speeds?. SBDX tip
tracking coordinates were compared to the 3D trajectory of the sheath as determined from a CT
scan of the phantom after the registration of the SBDX and CT coordinate systems. In the second
study, frame-to-frame tracking precision was measured for six different catheter configurations as a
function of image noise level ?662–7625 photons/mm2mean detected x-ray fluence at isocenter?.
Results: During catheter pullbacks, the 3D distance between the tracked catheter tip and the sheath
centerline was 1.0?0.8 mm ?mean ?one standard deviation?. The electrode to centerline distances
were comparable to the diameter of the catheter tip ?2.3 mm?, the confining sheath ?4 mm outside
diameter?, and the estimated SBDX-to-CT registration error ??0.7 mm?. The tip position was
localized for all 332 image frames analyzed and 83% of tracked positions were inside the 3D sheath
volume derived from CT. The pullback speeds derived from the catheter trajectories were within
5% of the programed pullback speeds. The tracking precision of ablation and diagnostic catheter
tips ranged from ?0.2 mm at the highest image fluence to ?0.9 mm at the lowest fluence. Track-
ing precision depended on image fluence, the size of the tracked catheter electrode, and the contrast
of the electrode.
Conclusions: High speed multiplanar tomosynthesis with an inverse geometry x-ray fluoroscopy
system enables 3D tracking of multiple high-contrast objects at the rate of fluoroscopic imaging.
The SBDX system is capable of tracking electrodes in standard cardiac catheters with approxi-
mately 1 mm accuracy and precision. © 2010 American Association of Physicists in Medicine.
Key words: catheter tracking, inverse geometry, fluoroscopy, tomosynthesis, image guidance
Catheter ablation has become a widespread treatment for a
range of cardiac arrhythmias, including forms of atrial fibril-
lation and ventricular tachycardia. In the traditional electro-
physiology approach, a multielectrode catheter is guided in-
side the cardiac chambers under x-ray fluoroscopy. The
endomyocardial surface is probed to create a map of electri-
cal activation and then energy is delivered through the tip to
destroy triggers or pathways of arrhythmia propagation. Re-
cently, ablation strategies have been developed to electrically
isolate cardiac anatomy that is critical to arrhythmia propa-
gation. For example, in atrial fibrillation, there may be mul-
tiple trigger sites within the sleeves of atrial tissue extending
into the pulmonary veins.1Instead of targeting all of these
sites separately, operators ablate circumferentially around the
pulmonary vein ostia2or antrum.3These anatomic ablation
6377 6377 Med. Phys. 37 „12…, December 2010 0094-2405/2010/37„12…/6377/13/$30.00© 2010 Am. Assoc. Phys. Med.
strategies demand accurate real-time knowledge of catheter
tip position relative to moving 3D cardiac structures.
Conventional x-ray fluoroscopy offers high resolution
real-time imaging of metallic catheter electrodes. However,
the lack of depth information and poor soft tissue contrast in
a 2D x-ray projection limits the utility of fluoroscopy alone
for anatomic-based ablation. Biplane x-ray fluoroscopy im-
proves the ability to appreciate catheter position in 3D but
the ionizing radiation dose to the patient is a source of con-
cern. Radiofrequency catheter ablation ?RFCA? is associated
with long fluoroscopic imaging times and radiation-induced
skin injuries have been reported.4A study of 28 RFCA pro-
cedures reported the fluoroscopic time from both planes av-
eraged 120.8 min ??62.6 min?, with peak skin dose increas-
ing with both fluoroscopic time and patient weight.5The
mean effective radiation dose from biplane fluoroscopy dur-
ing ablation of atrial fibrillation has been reported as 15.2–
39.0 mSv, increasing with body mass index.6
Nonfluoroscopic 3D electromagnetic ?EM? catheter track-
ing systems have emerged as an important tool for catheter
ablation. These systems perform 3D tip tracking using a local
magnetic field emitter and a field-sensing catheter tip.7The
tracked positions can be used to generate 3D electroanatomic
maps of activation time and can be merged with preacquired
volume images of cardiac anatomy ?e.g., cardiac CT?.8EM
tracking facilitates catheter ablation but has several limita-
tions. Specialized catheters and external equipment are re-
quired. Trackable catheters are limited to those offered by the
tracking system vendor. The specialized hardware compo-
nents built into the catheter constrain device profile and me-
chanical performance. Since these systems do not provide
live imaging, they are typically used in combination with
conventional x-ray fluoroscopy. Low dose imaging methods
that could offer 3D localization and tracking of any catheter
device would be an appealing alternative.
Inverse geometry x-ray fluoroscopy, based on scanning
beam digital x-ray ?SBDX? technology, has the potential to
substantially reduce patient x-ray doses and also provide
three-dimensional tracking of unmodified catheters by using
real-time tomosynthesis.9–12SBDX performs fluoroscopy
and angiography at 15–30 frames/s using a rapidly scanned
narrow x-ray beam directed at a small-area photon-counting
detector array ?Fig. 1?. As detailed in Ref. 10, the use of a
narrow x-ray beam, distant detector, and thick CdTe x-ray
detector results in low levels of detected scatter and high
x-ray stopping efficiency over a range of kVps, which in turn
allow a given image signal-to-noise ratio ?SNR? to be
achieved with lower x-ray output and lower patient dose.
Inverse geometry spreads the source x-rays over a larger area
at the patient entrance, further reducing skin dose. Signal-to-
noise ratio and entrance exposure measurements on a proto-
type SBDX system have demonstrated the potential for 84%
entrance exposure reduction without loss of SNR compared
to a conventional cardiac angiographic system at equal
The inverted system geometry also gives SBDX a unique
real-time tomosynthesis capability. SBDX simultaneously re-
constructs multiple tomosynthetic images spaced throughout
the patient, each of which portrays in-plane objects in focus
and out-of-plane objects as blurred. A 2D multiplane com-
posite is generated for live display.10Recently, algorithms
were described that use the tomosynthetic images generated
by SBDX to perform 3D tracking of high-contrast objects
such as catheter electrodes.12Computer simulations of
simple geometric phantoms indicated 3D tracking with sub-
millimeter accuracy and precision was feasible, depending
on image SNR, object velocity, and system tomographic
In this paper, we present the first investigation of the
SBDX catheter tracking algorithm using images acquired
with an SBDX prototype. The catheter tracking algorithm is
detailed and two phantom studies are reported. In the first
study, the accuracy and precision of catheter tracking is
evaluated in the presence of anatomic background for differ-
ent catheter velocities. In the second study, the precision of
3D localization is determined for six different cardiac cath-
eter geometries as a function of image SNR.
The 3D catheter tracking algorithm is an extension of the
tomosynthetic image reconstruction and plane scoring tech-
nique that is used to generate the live multiplane composite
display for SBDX. We begin with a brief review of the
SBDX scanning and image reconstruction method and then
describe the catheter tracking algorithm.
II.A. Scanning and image reconstruction
The SBDX x-ray source consists of a magnetically de-
flected focal spot, a planar transmission target, and a multi-
hole collimator. The collimator holes define a set of narrow
overlapping x-ray beams directed at the x-ray detector. There
are 100?100 focal spot positions on a 2.3 mm pitch. The
detector is located 1500 mm above the target plane in the
x-ray source. The source and detector are mounted to a gan-
FIG. 1. The SBDX system uses a raster scanned focal spot, transmission
target, multihole collimator, and hardware based reconstructor. Boxes with
solid lines indicate the steps of image formation and image presentation in
the current SBDX prototype. The proposed catheter tracking steps are
shown with dashed lines.
6378 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy6378
Medical Physics, Vol. 37, No. 12, December 2010
try, with the isocenter located 450 mm above the source
plane and 400 mm above the collimator exit surface. The full
field-of-view can be continuously scanned and reconstructed
at up to 30 frames/s. In this study, imaging was performed at
15 frames/s using 71?71 collimator holes and the isocenter
field-of-view was 11.4 cm?11.4 cm. The effective pulse
width in this imaging mode is 9.4 ms.10The maximum beam
current is 220 mAp at 100 kVp and 203 mAp at 120 kVp.11
As the focal spot is scanned from hole to hole, detector
images are captured and sent to a reconstruction engine. The
reconstruction engine contains a set of identical hardware
channels which simultaneously perform shift-and-add digital
tomosynthesis in real-time at multiple planes parallel to the
target surface. The current hardware configuration provides
16 planes. The plane position reconstructed by each channel
is programmable. Typically, a stack of 16 planes separated by
12 mm and centered on the gantry isocenter is reconstructed.
Implementation details of the tomosynthetic reconstruction
may be found in Ref. 10. In the 71?71 hole scanning mode,
each image plane has 710?710 pixels. Pixel size is 0.161
mm in the isocenter plane and follows the relationship p
=??S/10??1−?z/ZD??, where z is the source plane to recon-
struction plane distance, ?Sis the focal spot pitch, and ZDis
source to detector distance.
Each reconstructed plane has the property that in-plane
objects appear in focus and out-of-plane objects are progres-
sively blurred the farther out of the plane they reside ?Fig. 2?.
A single tomosynthetic plane is inconvenient for interven-
tional cardiac procedures ?e.g., angioplasty? since it does not
display all anatomy and devices in focus simultaneously. To
obtain an image display analogous to that of a conventional
fluoroscopic system, with all features presented in focus, the
reconstructed planes are combined into a multiplane compos-
To form the multiplane composite image, first a “score
image” is produced for each reconstructed plane. A score
image indicates the degree of focus at each ?row, column?
position in the tomosynthetic image and is calculated from
the local image gradient magnitudes ?detailed below?. Next,
the score images are compared. At each ?row, column? posi-
tion, the plane containing the maximum score is located.
Last, the composite image is formed from the original tomo-
synthetic image stack using the pixel value from the plane of
highest focus at each pixel position. This technique has been
reported in anthropomorphic phantom, animal, and human
subjects.10,11,13Figure 3 compares an SBDX single plane im-
age to a multiplane composite image of the same object.
II.B. 3D catheter tracking algorithm
The general procedure for 3D localization of objects in a
frame period is: ?i? Reconstruct tomosynthetic images at
regularly spaced planes throughout the volume of interest.
?ii? Generate a score image for each tomosynthetic image.
?iii? Segment the stack of score images into regions occupied
by objects to be tracked. ?iv? For each detected object region,
analyze score data and calculate a 3D coordinate represent-
ing the object center. This is repeated for each scan frame,
resulting in a set of ?x,y,z? coordinates for each frame pe-
II.B.1. Plane scoring
The first two steps of tracking, reconstruction at a set of
fixed planes and plane scoring, are designed to reuse image
data and methods already used by the SBDX reconstructor to
produce a real-time multiplane composite display. The gen-
eral scoring procedure is outlined in Fig. 4?b?. The tomosyn-
thetic image is prefiltered with a rectangular 2D kernel ?K1?
of adjustable width. The intent of the prefilter K1is to sup-
press image gradients created by pixel noise while at the
same time preserving object gradient information. The K1
kernel is generally smaller than the object being tracked.
Vertical and horizontal gradient images are calculated by fil-
tering with Prewitt gradient kernels and then the absolute
gradient values are averaged at each pixel position. Option-
ally, the score image may be downsampled by averaging
over 2?2 blocks of pixels. Although downsampling is not
required for catheter tracking, it was performed in this study
to mimic the behavior of the current SBDX reconstructor.
FIG. 2. SBDX tomographic blur. ?a? A point in the patient is imaged by a
patch of the source array, typically 10?10 spots. Backprojecting the point
to planes away from the object produce tomographic blurring that is sym-
metrical about the object. ?b? The blurring geometry and image pixel defi-
nition are such that the central point of the blurred object falls at the same
pixel row and column in all image planes ?shown with a thick dark line?.
FIG. 3. Phantom demonstration of the SBDX multiplane composite display.
The phantom consists of lead numbers spaced at 2 in. intervals along the
source-detector axis ?number “0” closest to the source?. ?a? SBDX single
plane tomosynthetic reconstruction of a plane 4 mm above the number “4.”
?b? SBDX multiplane composite image generated from single plane images
with 12 mm spacing.
6379 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy6379
Medical Physics, Vol. 37, No. 12, December 2010
Gradient filtering results in an image with scores located
mostly along object edges ?with higher contrast objects gen-
erating higher scores?. The subsequent segmentation of the
score data into object-centered regions is aided if these
scores are spread over the object area. Therefore, in the final
step of plane scoring, a second adjustable rectangular 2D
filter ?K2? is applied to smooth and spread edge gradients
over the object. The size of the K2kernel is related to the
size of the object.
Figure 5 shows example tomosynthetic images and the
corresponding score images for the catheter motion study
reported in this paper. The prefiltering and postfiltering ker-
nel sizes are free parameters in the catheter tracking algo-
rithm. The tuning of these filters is described in Sec. III C.
Note that if a catheter tip is guided through a background
with varying intensity ?e.g., diaphragm, lung, and spine?,
then even though the tip may have fixed contrast, the image
gradient magnitude at the tip will vary in proportion to the
local background intensity. This poses a problem during the
segmentation step, which uses a fixed score threshold to de-
tect objects. To facilitate the use of a single threshold, inde-
pendent of background intensity, two intensity lookup tables
are added to the above scoring procedure. Before the first
filtering step, the tomosynthetic image values are passed
FIG. 4. Overview of the SBDX tracking algorithm. ??a?–?c?? Score images are generated from the tomosynthetic images reconstructed at planes spaced
throughout the volume of interest. ?d? A MIP of the score data is formed and connected component labeling is applied to the MIP. ?e? Each detected object
?connected component? defines a cylinderlike region in the imaging volume. The row and column coordinates of an object are calculated from the MIP. ?f? The
z-coordinate is calculated from the scores versus z-plane inside the object cylindrical region.
FIG. 5. Demonstration of SBDX tracking for an ablation catheter tip and a fiducial attached to the anterior chest surface ?arrows?. A total of 40 image planes
were reconstructed in this example, with 12 mm spacing. ?a?Afixed ROI at eight different reconstructed planes demonstrates progressive out-of-plane blurring
?plane numbers 14, 17, 20, 23, 26, 29, 32, and 35 are shown?. ?b? Score image regions corresponding to the image regions shown in part ?a?. ?c? Maximum
intensity projection of the full score stack, with values below the object detection threshold set to zero. ?d? Result of connected component labeling. ??e? and
?f?? Score versus z-plane distributions for the tip and fiducial, respectively, prior to subtraction of the baseline.
6380 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy6380
Medical Physics, Vol. 37, No. 12, December 2010
through a log lookup table y=ln?x?. This converts the image
intensity to units proportional to the x-ray attenuation line
integral and gives the subsequently calculated absolute gra-
dient image units of pixel-to-pixel difference in the attenua-
tion line integral. These pixel-to-pixel differences in attenu-
ation are then converted to units of “edge contrast,” ranging
from 0% to 100%, by passing the absolute gradient image
through a second lookup table y=100?1−exp?−x??.
II.B.2. Segmentation and localization
For the third and fourth steps of tracking, segmentation
and 3D localization, two different approaches have been
studied.12In the method employed in this paper and shown
in Fig. 4, it is assumed that the patient is imaged in a radio-
graphic projection such that catheter electrodes are not over-
lapping or viewed end-on in the live fluoroscopic display. A
maximum intensity projection ?MIP? of the score stack is
generated along the source-detector direction
M?u,v? = max
where S represents the score stack data, M is the score MIP,
and the discrete variables u, v, and z are, respectively, pixel
column number, pixel row number, and z-plane coordinate.A
matching binary image is generated with values of 1 where
the score MIP equals or exceeds a specified object detection
1 if M?u,v? ? T1
0 otherwise ?.
The object detection threshold T1rejects gradients due to
background noise, and also serves to separate neighboring
local maxima in the score MIP ?e.g., different catheter elec-
trodes? into distinct regions. Distinct groups of pixels in im-
age ?u,v?-space are then identified by applying 2D con-
nected component labeling to the binary image B?u,v?.14
Each grouping of pixels defines a 3D cylinderlike segment in
?u,v,z?-space, as shown in Fig. 4?e?.
To determine the object’s 3D coordinate, the scores inside
the segmented object region are treated as a probability dis-
tribution function. Coordinates in the column and row direc-
tions ?analogous to x and y coordinates? are determined by
calculating the center-of-mass of object pixels in the score
MIP. If a detected object occupies a set of N pixels in the
score MIP, with ?u,v?-coordinates designated as ?ui,vi?, i.e.,
?u1,v1?, ?u2,v2?,... ?uN,vN?, then the object coordinates in
the column and row directions are given by
The object z-coordinate is calculated from the distribution of
the scores versus z-plane at positions inside the segmented
object region ?Fig. 4?f??. To construct this distribution, first
the scores in the cylinderlike object region are extracted, a
specified baseline value T2is subtracted from each pixel
score, and negative results are set to zero.
S?ui,vi,z? − T2 if S?ui,vi,z? ? T2
The baseline T2determines how many planes contribute to
the z-coordinate and also rejects background gradients. Then
the results are averaged at each plane to produce the score
versus z-plane distribution for the object.
The object z-coordinate is the center-of-mass of this distri-
where the summation is over the discrete plane positions z,
measured from the target surface in the x-ray source to the
plane of reconstruction.
Figure 5 shows an example thresholded score MIP, the
result of connected component labeling, and two score ver-
sus z-plane distributions prior to subtraction of the baseline
T2. The value of the object detection threshold T1is tuned to
the objects being tracked ?see Sec. III C?. The baseline value
T2is also adjustable. In this study, we set T2to approxi-
mately 2/3 of the peak in the score versus z-plane distribu-
tion for an object. Baseline subtraction ensures that a reason-
ably low number of planes contribute to the calculation of
the z-coordinate ?typically 7–8 around the object? and that
the postsubtraction distribution is symmetrical in shape ?i.e.,
not influenced by the constant background level or asymme-
tries that may exist in the tails of the distribution?.
The algorithm exploits two properties of the SBDX tomo-
synthetic blurring geometry and image reconstruction ?Fig.
2?. First, tomosynthetic blurring is symmetric about the ob-
ject’s true position. z-positions can be determined with
higher precision than the plane spacing by calculating a
weighted average of object focus versus z-plane. Second, the
image of an object remains centered on a fixed column and
row, regardless of plane position. This allows segmentation
to be performed in ?u,v?-space.
After all the segmented object regions have been pro-
cessed in the manner described by Eqs. ?3?–?7?, the set of
?u,v,z? coordinates corresponding to detected objects is
mapped to an ?x,y,z? coordinate system. This is accom-
plished using 3D linear interpolation and the known ?x,y,z?
coordinates of all the pixels in the score image stack. The
?x,y,z? coordinate system is right handed, defined with the
origin at the center of the target surface, z increasing toward
the detector, x increasing to the right in the image display
?patient left in supine position? and y increasing to the top of
the image display ?patient superior?.
To evaluate the accuracy and precision of SBDX catheter
tracking, two phantom studies were conducted on the SBDX
6381Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy 6381
Medical Physics, Vol. 37, No. 12, December 2010
prototype at the University of Wisconsin-Madison. In the
first study, an ablation catheter inside a chest phantom was
tracked at different speeds as it was pulled along a fixed 3D
trajectory. Tracking results were registered to and compared
to a volume CT of the chest phantom. In the second study,
tracking was performed at different image intensities and
signal-to-noise ratios for six stationary catheters. Table I lists
the different catheter types and geometries used. Figure 6
shows an SBDX x-ray image of the catheters. The specifica-
tions of the SBDX system have been previously reported.10,11
III.A. Moving catheter tip in chest phantom
A phantom was constructed to mimic the anatomy and
catheter trajectory typical of catheter ablation for the treat-
ment of left atrial fibrillation. A hollow cardiac chamber
phantom ?Venous Sam, LFA Anatomicals, Lake Forest, IL?
was placed inside a chest phantom consisting of plastic with
embedded ribs and spine ?Fig. 7?. A trans-septal catheter
sheath ?Agilis 12 Fr, St. Jude Medical, St. Paul, MN? was
attached to points inside the cardiac phantom to define a
reproducible catheter trajectory from the femoral vein, up the
inferior vena cava, into the right atrium, across the septum,
and into the left atrium. Catheter “A” was advanced up the
sheath so that the tip was located in the left atrium and the
proximal end was attached to a linear motion stage to per-
form constant-speed catheter pullbacks.
SBDX tracking results for the catheter tip were compared
to the 3D volume of the confining catheter sheath determined
from a CT scan of the phantom. To facilitate the registration
of the SBDX and CT coordinate systems, fiducials were at-
tached to points on the anterior and posterior chest walls.
Each fiducial consisted of a nylon nut, attached to the phan-
tom, and a matching nylon screw with a 2.4 mm diameter
steel ball bearing at the center of the screw head.
The SBDX gantry was placed in posterior-anterior orien-
tation and the phantom was centered at the gantry isocenter.
Three imaging runs were performed with the catheter pull-
back speed programed to 10, 25, and 50 mm/s. For reference,
the majority of velocities observed at landmarks on the coro-
nary arteries fall within the range 10–50 mm/s.15Scanning
was performed in 71?71 hole, 15 frames/s mode, which
provided an 11.4 cm wide image at gantry isocenter. The
x-ray source was operated at 100 kVp. Beam current was
adjusted such that the detected fluence at the isocenter plane
ranged from 1121 to 2244 photons/mm2in the vicinity of
the catheter, depending on location within the field-of-view
?29.1–58.2 photons/pixel?. Image intensity was similar to
that previously measured with 26–30 cm acrylic attenuation
at maximum beam current ?220 mAp at 100 kVp?.11The
chest phantom itself provided 10–14 cm of acrylic attenua-
tion depending on catheter location. An additional 2.3 cm
acrylic plus 0.5 mm Cu was placed in front of the x-ray
TABLE I. Cardiac electrophysiology catheters used in this study.
Label Catheter typeNo. of electrodes
Tip electrode Proximal electrodes
Max, min gap
Ablation ?7 Fr?
Ablation ?8 Fr?
Ablation ?8 Fr?
Diagnostic ?5 Fr?
Coronary sinus ?6 Fr?
Circular mapping ?7 Fr?
FIG. 6. SBDX image of catheters A–F. Circles indicate regions that were
independently tracked in 3D space.
FIG. 7. ?a? Chest phantom placed on an x-ray table and supporting foam.
The SBDX x-ray source is below. ?b? Cardiac phantom and catheter sheath
prior to placement inside the chest phantom. ?c? Tip of ablation catheter A.
?d? CT images of the chest phantom showing catheter sheath.
6382Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy 6382
Medical Physics, Vol. 37, No. 12, December 2010
source for beam hardening. The total filtration was estimated
to produce an x-ray spectrum shape similar to that exiting at
least 30 cm acrylic.16The beam current was 21.3 mAp.
Raw SBDX detector images generated during each imag-
ing run were recorded to a memory board in the SBDX pro-
totype hardware reconstructor. After the experiment, the data
were transferred to an external workstation for offline image
reconstruction and catheter tracking. This method was used
because the catheter tracking algorithm was not part of the
hardware reconstructor. Image reconstruction, catheter track-
ing, and selection of adjustable tracking parameters are de-
scribed in Sec. III C.
The CT scan of the phantom was performed on a 64 de-
tector row scanner ?Lightspeed VCT, GE Healthcare, Wauke-
sha, WI? with the catheter removed and the sheath left intact
?120 kVp, 295 mA, 0.984:1 helical pitch, 0.5 s rotation?. To
minimize streaking in the CT images, the fiducial screw
heads were swapped out for identical ones containing 2.4
mm aluminum ball bearings. Axial images were recon-
structed with 0.625 mm slice thickness and 0.488 mm in-
plane voxel width. The CT images were exported in DICOM
format for analysis. The comparison of the SBDX-tracked
catheter tip trajectory to the CT-derived sheath volume is
described in Sec. III D.
III.B. Catheter tracking versus image SNR
Previous simulation studies12found that object tracking
precision along the z-axis ?source-detector direction? de-
pends on image noise magnitude, with the z-coordinate stan-
dard deviation increasing as the mean detected fluence de-
creases. A wide range of fluences are encountered clinically,
depending on patient size, anatomic region, radiographic
projection, and x-ray source technique. Figure 8 shows the
mean detected fluence in photons/pixel as a function of kVp
and acrylic phantom thickness for the SBDX prototype in
71?71 hole, 15 frames/s scanning mode ?max power?.11
Furthermore, tracking precision can depend on the contrast
and size of the electrodes/markers in a catheter.
To investigate the relationship between tracking precision,
image noise level, and catheter element contrast and size,
SBDX imaging runs were acquired at different image inten-
sities with six different catheters affixed to the x-ray table.
Three of the catheters were four-electrode EP ablation cath-
eters ?“A,” “B,” and “C”?. The other three catheters were a
four-electrode EP diagnostic catheter ?“D”?, a ten-electrode
coronary sinus mapping catheter ?“E”?, and a ten-element
circular mapping catheter ?“F”?. The gantry was in posterior-
anterior orientation and the table surface was approximately
at gantry isocenter. The experimental setup placed the cath-
eters in a common plane approximately parallel to the target
in the x-ray source.
The x-ray source was operated with the same scan mode,
kVp, and mAp that was used in the moving catheter chest
phantom study ?Sec. III A?. Six imaging runs of 30 frames
each were acquired, using uniformly attenuating acrylic
phantoms of 7.0, 9.3, 11.7, 14.0, 16.3, and 18.6 cm acrylic
?plus 0.5 mm Cu filtration in all cases?. The acrylic was
distributed above and below the plane of the catheters. De-
pending on phantom thickness, the mean detected image flu-
ence ranged from 7625 to 662 photons/mm2at isocenter
?198–17.2 photons/pixel?. Figure 8 shows the mean image
fluences in this image SNR study and also the range of flu-
ences measured from the moving catheter chest phantom
study ?darker spine region versus bright atrial region?.
III.C. Image reconstruction and catheter tracking
In this first study, SBDX catheter tracking was performed
in software written in MATLAB ?MathWorks, Natick, MA? at
the University of Wisconsin-Madison. The software accepted
the raw detector data acquired during SBDX imaging, recon-
structed a stack of tomosynthetic images for each frame pe-
riod ?emulating the SBDX reconstruction hardware?, and
then performed the plane scoring, segmentation, and local-
ization steps described in Sec. II B. The tomosynthetic plane
stack, score stack, and 3D coordinates of detected objects
were saved for each frame period.
Tomosynthetic images were reconstructed at fixed plane
positions spaced by 12 mm and centered on the phantom. As
discussed in Sec. II, the hardware reconstructor of the SBDX
prototype can reconstruct 16 planes simultaneously in real-
time. During tracking, typically seven to eight planes about
an object will contribute to the object’s 3D coordinate calcu-
lation. In the study of stationary catheter tracking precision
versus SNR ?Sec. III B?, 16 plane positions were recon-
structed. In the study of a moving catheter in a chest phan-
tom ?Sec. III A?, extra planes were added to enable simulta-
neous tracking of the fiducials placed on the anterior and
posterior chest surfaces. Therefore, the moving catheter ex-
periment used a stack of 40 planes centered on the chest
phantom. All plane stacks were positioned so that one plane
coincided with the isocenter plane.
The catheter tracking algorithm has two adjustable filters
in the plane scoring stage ?K1and K2? and two adjustable
thresholds in the segmentation and localization stage ?T1and
T2?. Although these parameters could be tuned to individual
FIG. 8. Mean image fluence during catheter tracking studies ?shaded re-
gions? compared to the mean image fluence measured versus acrylic phan-
tom thickness and kVp on the SBDX prototype system ?lines; Ref. 11?.
6383Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy6383
Medical Physics, Vol. 37, No. 12, December 2010
catheter element shapes and orientations, an effort was made
to identify a single combination of parameters that would
enable tracking for all elements on a catheter. An effort was
also made to use common tracking parameters across differ-
ent image scenes and different catheter types where feasible.
Initial investigations were conducted using a range of
square filter kernel sizes ?1–22 pixels wide? and a range of
threshold values ?5%–55%? with sample image frames. The
performance of each filter combination ?K1and K2? was
evaluated by examining two effects: ?i? The peak score value
at the object relative to the scores in a background region and
?ii? the number of valid object detection thresholds that the
filtering enabled, where a threshold is considered valid if it
results in detection of the correct number of catheter ele-
ments. For prefiltering, it was decided to use K1=5?5 ?0.8
mm wide at isocenter?, which showed good performance for
a number of catheters. The postfiltering kernel K2was ad-
justed depending on which catheters were in the imaging
field. It was found that segmentation was aided if the post-
filtering kernel size was similar to the object size and not
significantly larger than the smallest dimension of the object
as it appeared in the image. The selected value for T1was the
midpoint of the valid range of thresholds for the selected
filter combination. The T2value was set to approximately 2/3
of the peak in the score versus z-plane distribution.
Note the postfiltering step was performed on score images
that were downsampled via 2?2 block averaging. For the
sake of consistency, all filter kernel sizes are reported in units
of input tomosynthetic image pixels. The width of the kernel
in mm at the isocenter plane is also reported for reference.
III.D. Evaluation of tracking performance
III.D.1. Moving catheter tip tracking
The set of 3D points generated by SBDX catheter tip
tracking during a pullback was compared to the known 3D
volume of the confining catheter sheath determined from the
CT scan. Details of the SBDX-to-CT registration, sheath seg-
mentation, and sheath centerline calculation are provided be-
low. For each image frame, the 3D tip-to-centerline distance
was calculated, defined as the distance between a tracked
catheter tip position and the nearest point on the sheath cen-
terline. Results are reported as the average value over all
frames ?one standard deviation. Tracking results were ana-
lyzed when the tip was moving and within the central 92% of
the image field-of-view. The fraction of all SBDX-tracked
points falling within the sheath volume was also calculated
and for points falling outside of the sheath, the tip-to-sheath
surface distance was calculated. Last, the average speed of
the catheter along its tracked 3D trajectory was calculated
and compared to the programmed speed. The average speed
was calculated after performing five-frame moving average
on the catheter trajectory ?0.33 s temporal window? to mini-
mize the influence of coordinate fluctuations.
A rigid 3D rotation/translation transformation matrix be-
tween the SBDX and CT systems was calculated from seven
fiducial positions.17The 3D position of a fiducial in SBDX
coordinates was determined by averaging tracking results
within a cylindrical volume of interest around the fiducial.
The corresponding 3D position in CT coordinates was deter-
mined by locating the fiducial within the CT image volume
and calculating its intensity-weighted center-of-mass, ignor-
ing values below 300 HU.
The CT voxels contained within the sheath wall and lu-
men were identified using a combination of edge detection
and intensity-based segmentation. The 3D centerline of the
sheath was determined by skeletonizing the segmented vol-
ume ?sheath plus lumen? using a homotopy-preserving flux-
driven algorithm.18A 3D smoothing spline curve was fit to
the skeleton voxel positions to obtain a sheath centerline
representation with subvoxel precision. The smoothing
spline was evaluated at points spaced evenly by 0.075 mm
along the 3D curve.
III.D.2. Catheter tracking versus image SNR
Cylindrical ROIs were defined around elements in a cath-
eter and the 3D tracking coordinates versus frame were ex-
tracted for each element. Mean and standard deviation in the
tracked position were calculated in the x, y, and z directions.
As seen in Fig. 6, the tip electrode and first proximal elec-
trode on a catheter are often bridged by an internal high-
contrast component or are very close together. Therefore, the
tip and first proximal electrode were tracked as a single ob-
ject ?this was also done for the chest phantom study?.
z-coordinate results are reported for each tip ROI. For the
remaining proximal elements in a catheter, the z-coordinate
mean and z-coordinate standard deviation measured from the
individual ROIs were averaged across ROIs. Averaging was
performed since proximal elements in a catheter were of
identical shape. Standard deviation in the x and y directions
were similar and results are reported as the maximum stan-
dard deviation in either direction.
IV.A. Moving catheter tip in chest phantom
SBDX catheter tip tracking during a 10 mm/s pullback is
compared to the CT-derived catheter sheath volume in Fig. 9.
The figure shows a single SBDX tomosynthetic image frame
during pullback and the 3D tip tracking results for all frames
from three different perspectives. All tip tracking results
have been registered to the CT coordinate system using a
transformation derived from the tracked fiducial positions.
Note the example tomosynthetic image planes and score im-
ages shown in Fig. 5 are a region of interest from the same
image frame shown in Fig. 9?a?. The 3D tip trajectory pro-
duced by the SBDX tracking algorithm closely followed that
of the catheter sheath volume determined from CT.
Figure 10 shows the components of the tip trajectory
along the x-axis ?phantom right-to-left?, y-axis ?inferior-to-
superior?, and z-axis ?posterior-to-anterior, also the source-
detector axis of SBDX? for 10, 25, and 50 mm/sec catheter
pullback speeds. The 3D distance between the tracked tip
position and the sheath centerline is shown next to each tra-
jectory figure. For the 10 mm/s pullback, the tip-to-centerline
6384 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy 6384
Medical Physics, Vol. 37, No. 12, December 2010
distance was 0.9?0.7 mm and 83% of all tracked tip posi-
tions were located inside the catheter sheath volume. The 25
mm/s pullback yielded 1.1?0.9 mm tip-to-centerline dis-
tance and 78% of tip positions inside the sheath volume. The
corresponding 50 mm/s pullback results were 0.9?0.6 mm
and 88%. The tip position was localized in every image
frame analyzed ?332 frames total for three imaging runs?.
The tip trajectory spanned 55 mm along the z-axis, 76 mm
along the x-axis, and 65 mm along the y-axis. The trajectory
was intersected by five of the reconstructed planes. Results
clearly demonstrate that 3D tracking coordinates can be ob-
tained with precision finer than the plane-to-plane spacing
Averaged over 10–50 mm/s pullback speeds, the tip-to-
centerline distance was 1.0?0.8 mm and 83% of tracked
positions were inside the sheath. Tip positions outside the
sheath were within 0.6?0.5 mm of the sheath surface. Tip-
to-centerline distances were comparable to the diameter of
the ablation catheter tip ?2.3 mm?. Note that tip-to-centerline
results include errors introduced by the SBDX-to-CT coordi-
nate transformation. In this study, a simple rigid transforma-
tion was derived from seven fiducial positions on the chest
wall. Registration error was estimated to be ?0.7 mm, cal-
culated as one standard deviation in the individual distances
between the transformed SBDX fiducial position and the CT
fiducial position. Frame-to-frame tracking precision also de-
pends on the image SNR in the vicinity of the catheter tip.
Image intensity ranged from 58.2 photons/pixel at the start of
the trajectory ?atrium region? to 29.1 photons/pixel at the end
of the trajectory ?spine region?. This variation in image in-
tensity along the trajectory is consistent with the observation
that tip-to-centerline distance tended to increase and show
more frame-to-frame deviations near the end of the catheter
pullbacks ?see Fig. 10?.
The average speeds calculated from the tip trajectories
were 10.5, 24.8, and 47.6 mm/s. The SBDX-tracked tip
speeds were all within 5% of with the programmed catheter
pullback speeds. The tip tracking parameters for all catheter
pullbacks were: Prefilter K1=5?5 ?0.8 mm?, postfilter K2
=22?22 ?3.5 mm?, detection threshold T1=25%, and
z-distribution baseline T2=25%. The fiducial tracking pa-
rameters were: K1=5?5 ?0.8 mm?, K2=14?14 ?2.3 mm?,
T1=20%, and T2=15%.
IV.B. Tracking precision versus image noise and
The z-coordinate for each of the six stationary catheter
tips is shown in Fig. 11 as a function of the mean background
FIG. 9. ?a? Example tomosynthetic image during a 10 mm/s catheter pullback. ??b?–?d?? Tip tracking coordinates for all image frames ?blue points? compared
to the catheter sheath volume ?green?. The comparison is shown from three perspectives. The right-left ?R-L?, inferior-superior ?I-S?, and posterior-anteior
?P-A? axes are shown for reference.
6385 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy 6385
Medical Physics, Vol. 37, No. 12, December 2010
image intensity. The symbols represent the mean position
over 30 frames; the error bars represent ?one standard de-
viation. The catheters were coplanar but not all at the same
z-coordinate. All analysis ROIs are indicated on Fig. 6. Table
I lists the measured length, diameter, and in-plane image
contrast for the catheter electrodes. The tip electrodes of the
ablation, diagnostic, and coronary sinus catheters ?A–E? had
95%–97% contrast, lengths from 2.0 to 8.0 mm, and diam-
eters from 1.7 to 2.7 mm. Catheter F, a circular mapping
catheter, had smaller, lower contrast ?52% to 61%? elec-
The same set of adjustable tracking parameters was used
for all elements within a catheter. Furthermore, a single set
of parameters was used for catheters A–E, which enabled
simultaneous tracking of these catheters. Figure 11 shows an
example of simultaneous 3D tracking of the five catheters.
The tracking parameters were: K1=5?5 ?0.8 mm?, K2=10
?10 ?1.6 mm?, T1=30%, and T2=20%. Since catheter F
contained lower contrast electrodes, the threshold values for
this catheter were set to T1=12% and T2=8% and tracking
was performed separately.
Figure 12 shows the standard deviation in catheter tip
z-coordinate precision improved as the detected image flu-
ence increased. With catheter A ?also used in the moving
catheter chest phantom study?, the z-coordinate standard de-
viation ranged from 0.8 to 0.2 mm for background intensities
from 17.2 to 198 photons/pixel. The tracking precision was
similar to that observed in the chest phantom study. The
trend in tip tracking z-precision versus image intensity was
similar for catheters A–E. For these catheter tips and all im-
aging conditions, the highest and lowest z-coordinate stan-
dard deviations were 0.9 and 0.2 mm. The tracking of cath-
eter tip F was less precise, with the z-coordinate standard
deviation ranging from 1.6 mm to 0.4 mm.
The z-coordinate standard deviations for the proximal
electrodes are shown in Fig. 13. As with the tips, there was a
trend of improving precision as image intensity increased
and there was a difference between catheters A–E and cath-
eter F. In any given catheter, the proximal electrodes were
tracked with somewhat lower z-coordinate precision than the
tip. The ratio of proximal deviations to tip deviations ranged
FIG. 10. Tip tracking results for 10 ?top row?, 25 ?middle row?, and 50 mm/s ?bottom row? catheter pullbacks along a fixed 3D trajectory in a chest phantom.
The left column shows the tracked tip position versus image frame in the right-left ?R-L?, inferior-superior ?I-S?, and posterior-anterior ?P-A? directions. The
right column shows the 3D distance from catheter tip-to-sheath centerline, versus image frame.
6386 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy 6386
Medical Physics, Vol. 37, No. 12, December 2010
from 1.33 ?catheter B? to 2.25 ?catheter C?. The difference is
believed to be due to the fact that proximal electrodes are
smaller than the tip and yield fewer gradient estimates over
the electrode perimeter. The proximal electrodes also had
slightly lower image contrast ?e.g., 84%–92% for catheters
The maximum standard deviation in the x and y directions
was 0.1 mm for any element of cathetersA–E in any imaging
scenario. For catheter F, the maximum standard deviation in
x or y was 0.4 mm. Therefore the greatest uncertainty in 3D
coordinate is in the z-direction ?source-detector axis?. This
can be understood as the result of the relatively narrow to-
mographic angle of the SBDX system.
Fluoroscopic x-ray imaging is the foundation of many in-
terventional procedures but it carries with it the potential to
deliver high x-ray dose. SBDX is designed to provide fluo-
roscopic imaging at a substantially reduced dose rate com-
pared to conventional fluoroscopic systems.11Dose reduction
is achieved, in part, by the inverse geometry configuration of
a large-area scanned source and small-area detector. The re-
sulting reconstructed image is planar tomographic in nature
and this fact can be exploited to provide the depth informa-
tion valuable in many interventional procedures, such as the
cardiac RF ablation procedures that are the focus of this
study. A previous simulation work predicted that the SBDX
catheter tracking algorithm described in this paper would
enable 3D catheter tracking with accuracy and precision of
approximately 1 mm or better.12The phantom studies re-
ported here, which were conducted on an SBDX prototype
system with a variety of catheter geometries, trajectories, ve-
locities, background features, and image intensities, confirm
SBDX catheter tracking is compatible with standard cath-
eters containing high-contrast markers, can follow multiple
markers and catheters in the field-of-view, has temporal sam-
pling equal to the imaging rate ?15–30 Hz?, uses a single
gantry view angle, and does not require geometric calibra-
tions. Since the location of each pixel in each tomosynthetic
image plane is precisely known relative to the source and
detector, the 3D coordinates are automatically registered to
the SBDX gantry. These coordinates can be transformed to a
patient-based coordinate system using gantry and table ori-
FIG. 11. Tip tracking in the z-direction ?source-detector axis? versus mean image fluence for six different catheters. Symbols indicate the mean z-coordinate
over 30 image frames; error bars indicate ?one standard deviation in the individual z-coordinates. The figure on the far right demonstrates simultaneous
tracking of elements in catheters A–E.
FIG. 12. Standard deviation in catheter tip z-coordinate versus mean image
6387Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy6387
Medical Physics, Vol. 37, No. 12, December 2010
The tracking algorithm is based on the concept that object
blurring versus plane position can be treated as a distribution
function whose center-of-mass represents the true object po-
sition. It is designed to make use of the tomosynthetic recon-
structions and plane scoring techniques that were developed
for real-time fluoroscopic image display. Tomosynthesis-
based 3D localization has been reported in other applica-
tions, for example, brachytherapy seed localization.19Track-
ing in cardiac applications is challenging, however, due to
the high object velocities involved. To our knowledge,
SBDX is the only tomosynthesis system with the scanning
speed, x-ray output, and continuous imaging capability re-
quired for both live x-ray fluoroscopy and frame-by-frame
3D cardiac catheter tracking.
The SBDX study of a moving ablation catheter in a chest
phantom demonstrated 3D tip tracking with 1.0 mm mean
error after registering and comparing tracked positions to CT.
Imaging was conducted at an intensity similar to that which
may be encountered clinically. The study of stationary cath-
eters versus image intensity demonstrated z-coordinate stan-
dard deviations ranging from 0.2 to 0.9 mm for ablation and
diagnostic catheter tips depending on background image in-
tensity. Smaller deviations were observed in the x- and
y-coordinates. Our experiments show that SBDX may offer
similar catheter tip tracking accuracy to commercially avail-
able nonfluoroscopic EM tracking systems which are in use
for radiofrequency catheter ablation procedures. EM tracking
systems have been shown to provide ?1 mm relative dis-
tance error and location standard deviation.20
Tracking precision depended on image intensity, catheter
electrode size, and electrode contrast. The improvement in
tracking precision with background intensity was consistent
with expectations. This may be understood as the result of
improved object signal-to-noise ratio in each tomosynthetic
image, which yields more reliable gradient estimates at any
object point, and also a reduction in background gradient
values, which gives more freedom in the selection of detec-
tion threshold. A previous work has shown that z-axis preci-
sion is also dependent on the tomographic angle of the
SBDX system.12In a detector redesign currently underway,
the detector width has been doubled. The increase in both
tomographic angle and image SNR anticipated for this new
detector is expected to improve tracking precision compared
to the SBDX prototype used in this study.
Tracking performance depends on the proper selection of
filtering kernels and score thresholds. In this study, several
guidelines for the selection of these parameters are pre-
sented. Although the algorithm could be optimized to yield
the best precision for individual electrodes and electrode ori-
entations, there is value in finding a set of parameters that
work for a variety of objects simultaneously. In this study, it
was possible to simultaneously track the electrodes on five
different catheters across a range of image intensities. Mul-
tiple catheter tracking may be useful in a scenario where a
navigated catheter tip is tracked relative to a second catheter
positioned at an anatomic landmark ?e.g., in the coronary
Several limitations of the tracking algorithm should be
noted. The tracking algorithm used here assumes that elec-
trodes are nonoverlapping in the image field-of-view. An al-
ternative algorithm that enables tracking of overlapping ob-
jects separated by several cm has been described; however, it
is more computationally expensive.12We note that the non-
overlapping condition is often satisfied during clinical fluo-
roscopy since the interventionist generally avoids end-on
views. Although a variety of velocities and 3D trajectories
were tested in phantoms, true 3D catheter motion in the clini-
cal setting may be more complicated and will include respi-
ratory motion. Studies of tracking performance in animal
models are underway.
Ultimately, a real-time implementation of the tracking al-
gorithm is desired. Presently, only multiplane tomosynthesis
and plane scoring may be performed in real-time SBDX
hardware. Tracking was performed offline in software using
detector data recorded and downloaded from the reconstruc-
tor. We note that the design of a next generation SBDX sys-
tem is underway and the real-time reconstruction hardware
for that system has been designed to enable both imaging
and independent analysis of tomosynthetic plane stacks ?e.g.,
catheter tracking?.21For a clinical implementation of SBDX
catheter tracking, a method for registering tracked 3D coor-
dinates to preacquired volume images ?e.g., CT or MR? is
also desired. The external fiducial technique used in this
study to register SBDX-to-CT is one possibility, as is the use
of internal landmarks or internally placed catheters. This is
left as a topic for further investigation.
High speed multiplanar tomosynthesis with an inverse ge-
ometry x-ray fluoroscopy system enables accurate and pre-
cise 3D tracking of high-contrast electrodes in standard car-
diac catheters. Phantom studies with the SBDX prototype
system demonstrate catheter tip tracking accuracy and preci-
sion of 1 mm or better is feasible. The combination of low
dose fluoroscopic imaging and 3D catheter tracking offered
by SBDX technology is well-suited to anatomically targeted
interventional procedures such as radiofrequency catheter ab-
lation of cardiac arrhythmias.
FIG. 13. Standard deviation in proximal electrode z-coordinate versus mean
6388 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy 6388
Medical Physics, Vol. 37, No. 12, December 2010
Financial support for this work was provided by NIH
Grant No. R01 HL084022. The authors wish to thank Dr.
Douglass Kopp and Dr. Andrew Klein for their helpful dis-
cussions and assistance. The authors also thank Triple Ring
Technologies, Inc. and NovaRay Medical, Inc. for technical
support for the SBDX prototype system.
a?Author to whom correspondence should be addressed. Electronic mail:
1M. Haïssaguerre, P. Jaïs, D. C. Shah, A. Takahashi, M. Hocini, G. Qui-
niou, S. Garrigue, A. Le Mouroux, P. Le Métayer, and J. Clémenty,
“Spontaneous initiation of atrial fibrillation by ectopic beats originating in
the pulmonary veins,” N. Engl. J. Med. 339, 659–666 ?1998?.
2C. Pappone, S. Rosanio, G. Oreto, M. Tocchi, F. Gugliotta, G. Vicedo-
mini, A. Salvati, C. Dicandia, P. Mazzone, V. Santinelli, S. Gulletta, and
S. Chierchia, “Circumferential radiofrequency ablation of pulmonary vein
ostia—A new anatomic approach for curing atrial fibrillation,” Circula-
tion 102, 2619–2628 ?2000?.
3M. Ouyang, D. Bansch, S. Ernst, A. Schaumann, H. Hachiya, M. Chen, J.
Chun, P. Falk, A. Khanedani, M. Antz, and K.-H. Kuck, “Complete iso-
lation of left atrium surrounding the pulmonary veins: New insights from
the double-lasso technique in paroxysmal atrial fibrillation,” Circulation
110, 2090–2096 ?2004?.
4T. R. Koenig, D. Wolff, F. A. Mettler, and L. K. Wagner, “Skin injuries
from fluoroscopically guided procedures: Part 1, characteristics of radia-
tion injury,” AJR, Am. J. Roentgenol. 177, 3–11 ?2001?.
5K. Chida, H. Saito, H. Otani, M. Kohzuki, S. Takahashi, S. Yamada, K.
Shirato, and M. Zuguchi, “Relationship between fluoroscopic time, dose-
area product, body weight, and maximum radiation skin dose in cardiac
interventional procedures,” AJR, Am. J. Roentgenol. 186, 774–778
6J. Ector, O. Dragusin, B. Adriaenssens, W. Huybrechts, R. Willems, H.
Ector, and H. Heidbuchel, “Obesity is a major determinant of radiation
dose in patients undergoing pulmonary vein isolation for atrial fibrilla-
tion,” J. Am. Coll. Cardiol. 50, 234–242 ?2007?.
7S. A. Ben-Haim, D. Osadchy, I. Schuster, L. Gepstein, G. Hayam, and M.
E. Josephson, “Nonfluoroscopic in vivo navigation and mapping technol-
ogy,” Nat. Med. 2, 1393–1395 ?1996?.
8J. Dong, H. Calkins, S. B. Solomon, S. Lai, D. Dalal, A. Lardo, E. Brem,
A. Preiss, R. D. Berger, H. Halperin, and T. Dickfield, “Integrated elec-
troanatomic mapping with three-dimensional computed tomographic im-
ages for real-time guided ablations,” Circulation 113, 186–194 ?2006?.
9E. G. Solomon, B. P. Wilfley, M. S. Van Lysel, A. W. Joseph, and J. A.
Heanue, “Scanning-beam digital x-ray ?SBDX? for cardiac angiography,”
Proc. SPIE 3659, 246–257 ?1999?.
10M. A. Speidel, B. P. Wilfley, J. M. Star-Lack, J. A. Heanue, and M. S.
VanLysel, “Scanning-beam digital x-ray ?SBDX? technology for interven-
tional and diagnostic cardiac angiography,” Med. Phys. 33, 2714–2727
11M. A. Speidel, B. P. Wilfley, J. M. Star-Lack, J. A. Heanue, T. D. Betts,
and M. S. Van Lysel, “Comparison of entrance exposure and signal-to-
noise ratio between an SBDX prototype and a wide-beam cardiac angio-
graphic system,” Med. Phys. 33, 2728–2743 ?2006?.
12M. A. Speidel, A. P. Lowell, J. A. Heanue, and M. S. Van Lysel, “Frame-
by-frame 3D catheter tracking methods for an inverse geometry cardiac
interventional system,” Proc. SPIE 6913, 691311 ?2008?.
13M. R. Wolff, J. G. Keevil, M. A. Speidel, M. Wilfley, B. P. Wilfley, J. M.
Star-Lack, J. A. Heanue, and M. S. Van Lysel, “Pilot study with a
scanning-beam digital x-ray system,” Am. J. Cardiol. 94, 95E ?2004?.
14R. Sedgewick, Algorithms in C, 3rd ed. ?Addison-Wesley, Reading,
15M. Vembar, M. J. Garcia, D. J. Heuscher, R. Haberl, D. Matthews, G. E.
Bohme, and N. L. Greenberg, “A dynamic approach to identifying desired
physiological phases for cardiac imaging using multislice spiral CT,”
Med. Phys. 30, 1683–1693 ?2003?.
16M. A. Speidel, “Performance characteristics of the scanning-beam digital
x-ray ?SBDX? cardiac imaging system,” Ph.D. dissertation, University of
Wisconsin, 2003 ?available from UMI Dissertation Services, University
of Wisconsin, Ann Arbor, MI?.
17J. M. Fitzpatrick, D. L. G. Hill, and C. R. Maurer, Jr., Handbook of
Medical Imaging Volume 2. Medical Image Processing and Analysis, ed-
ited by M. Sonka and J. M. Fitzpatrick ?SPIE, Bellingham, 2004?, pp.
18S. Bouix, K. Siddiqi, and A. Tannenbaum, “Flux driven fly throughs,” in
Proceedings of the 2003 IEEE Computer Society Conference on Com-
puter Vision and Pattern Recognition, Vol. 1, pp. I449–I454, 2003 ?un-
published?; available at http://ieeexplore.ieee.org using the doi 10.1109/
19M. Brunet-Benkhoucha, F. Verhaegen, S. Lassalle, D. Beliveau-Nadeau,
B. Reniers, D. Donath, D. Taussky, and J.-F. Carrier, “Clinical implemen-
tation of a digital tomosynthesis-based seed reconstruction algorithm for
intraoperative postimplant dose evaluation in low dose rate prostate
brachytherapy,” Med. Phys. 36, 5235–5244 ?2009?.
20L. Gepstein, G. Hayam, and S. A. Ben-Haim, “A novel method for non-
fluoroscopic catheter-based electroanatomical mapping of the heart. In
vitro and in vivo accuracy results,” Circulation 95, 1611–1622 ?1997?.
21A. P. Lowell and W. Sun, “Real-time x-ray tomosynthesis imaging using
an ATCA general-purpose data acquisition and analysis platform,” in
IEEE Nuclear Science Symposium Conference Record, pp. 27–31, 2008
?unpublished?; available at http://ieeexplore.ieee.org using the doi
6389 Speidel et al.: 3D catheter tracking using inverse geometry x-ray fluoroscopy6389
Medical Physics, Vol. 37, No. 12, December 2010