Intrinsic gating for small-animal computed tomography: a robust ECG-less paradigm for deriving cardiac phase information and functional imaging.

1
Intrinsic Gating for Small-animal CT: A Robust,
ECG-less Paradigm for Deriving Cardiac Phase
Information and Functional Imaging
(ALMOST FINAL VERSION AS PUBLISHED)

Julien Dinkel, Dr.1*, Soenke H. Bartling, Dr. 2,3*, Jan Kuntz2,3, Michael Grasruck, Dipl.-
phys4, Masayoshi Iwasaki5, Dr, Stefanie Dimmeler Prof. Dr.5, Rajiv Gupta, PhD, MD6,
Wolfhard Semmler, Prof. Dr. Dr.2, Hans-Ulrich Kauczor, Prof. Dr.1, Fabian Kiessling,
PD Dr.2,3

* JD and SB contributed equally.

1Department of Radiology, 2Department of Medical Physics in Radiology, 3 Junior
Group Molecular Imaging, German Cancer Research Center (DKFZ), Heidelberg,
Germany,
4Siemens Medical Solutions, Forchheim, Germany
5Molecular Cardiology, University of Frankfurt, Germany
6Department of Radiology, Massachusetts General Hospital, Boston, MA, USA

Communicating author:
Julien Dinkel
Dept. Radiology, German Cancer Research Center (DKFZ),
Im Neuenheimer Feld 280
69120 Heidelberg, Germany
Phone: +49 6221 422524 - Fax: +49 6221 422572
E-mail: j.dinkel@dkfz-heidelberg.de

Word count: 5282
Subject Codes: [124] Cardiovascular imaging agents/Techniques

2
[130] Animal models of human disease

3
Abstract
A novel method of intrinsic cardiac gating in small-animal CT imaging is presented.
In this electrode-less method that operates without external ECG monitoring, the
gating reference signal is derived from the raw data of the CT projections. After
filtering, the derived gating reference signal is used to rearrange the projection
images retrospectively into data sets representing different time points in the cardiac
cycle during expiration. These time-stamped projection images are then used for
tomographic reconstruction of different phases of the cardiac cycle. Intrinsic-gating
was evaluated in mice and rats and compared with extrinsic retrospective gating. An
excellent correlation was achieved between ECG-derived gating signal and self-
gating signal (Pearson correlation test r >0.97). Functional parameters (ventricular
volumes and ejection fraction) obtained from the intrinsic and the extrinsic data sets
were equally significant (p>0.95). The ease of use and reliability of intrinsic gating
were demonstrated via a chemical stress test, where the system performed flawlessly
despite increased heart rate. Because of intrinsic gating, the image quality was
improved to the extent that even the coronary arteries of mice could be visualized in
vivo despite a heart rate approaching 430 bpm. Feasibility of intrinsic gating for
functional imaging and assessment of cardiac wall motion abnormalities was
successfully tested in a mouse model of myocardial infarction. Our results
demonstrate that self-gating using advanced software post-processing of projection
data promises to be a valuable tool for rodent CT imaging and renders ECG gating
with external electrodes superfluous.


Keywords: CT, small-animal imaging, flat-panel detector, self-gating

4
Introduction
Rodent models, particularly transgenic mouse models of cardiac disease, play a
major role in cardiovascular research and have proven their value and clinical
relevance to human medicine (1). In this context, the assessment of cardiac
morphology and function is important. Currently, the myocardial function and
structure of rodents are predominantly studied using magnetic resonance imaging
(MRI) (2) or echocardiography. However, MRI studies are complex and time
consuming. Furthermore, access to MRI systems, particularly dedicated small-animal
MRI systems, is limited and expensive. Echocardiographic techniques are rarely
three-dimensional (3), and global functional parameters must be approximated from
2D cross-sectional data (4).
Small-animal computed tomography (CT) imaging, despite its manifold applications
as a sensitive and cost-effective diagnostic tool, does not yet play a significant role in
the assessment of cardiac morphology and function. Compared with most
commercial micro-focus CT (µCT) systems, the flat-panel detector based small-
animal CT scanners that have recently been developed provide a higher scanning
speed, which makes cardiac imaging feasible in principle. Here, gating permits
physiological heart and lung motion to be taken into consideration during scanning.
The gating signal can be utilized to suppress motion artifacts and acquire functional
information. Both prospective (5-7) and retrospective (8, 9) methods for respiratory
and cardiac gating have been described, and their benefit for image analysis
demonstrated (7-10).
Regardless of whether they are retrospective or prospective, all gating methods
currently in use in small-animal CT imaging depend on electrocardiography (ECG) to
derive a gating reference signal. In addition, heart movement due to respiration must
either be suppressed via intubation or detected by a pneumatic cushion and taken

5
into consideration. Self-gating (synonymous with image-based, intrinsic, or raw-data
based gating) in small-animal CT, where the gating information is taken directly from
the acquired projection data, has so far only been demonstrated for respiratory
motion imaging (10).
To the best of our knowledge, cardiac self-gating in small animal imaging that takes
into account the characteristics of flat-panel-based cone-beam CT scanners (e.g.,
wide z-coverage, relatively slow data acquisition rate, and continuous volumetric
sampling), is presented here for the first time.
In order to demonstrate that the proposed method was equivalent to established
ECG-based gating method, two different approaches were employed. In a
prospective approach prior to the reconstruction, we showed equivalence of the
ECG-derived gating signal and the intrinsic gating signal by demonstrating a high
correlation between the two. In a retrospective outcomes-based approach, we
demonstrated the equivalence of cardiac function parameters derived from the two
gating methods. To illustrate the robustness of the proposed self-gating method and
how it might be used in the future, we performed dobutamine stress CT scans in mice
and scans of a murine model of infarction.

6
Materials and Methods
Flat-Panel-Based Volume-CT Scanner and Scans
A prototype CT scanner was used for image acquisition and gating experiments. Its
main features were a flat-panel detector and a modified X-ray tube, both mounted on
a multislice CT gantry (9, 13). The flat-panel detector (PaxScan 4030CB, Varian
Medical Systems, Palo Alto, CA) employed by this prototype scanner consisted of
2048 x 1536 detector pixels over an active area of 40 x 30 cm2, resulting in a pixel
size of 1942 µm2. For this experiment, the active detector area was limited to 192
lines in the z direction and 1024 rows in the x-y direction to increase the frame rate.
Moreover, the detector was read out in a 2 x 2 binning mode, i.e., four neighboring
pixels were averaged. This resulted in a decreased scan field-of-view of 25 x 25 x 4.5
cm3, which was still big enough to cover the entire thorax and diaphragm of a rat. The
resulting frame rate was 100 frames per second (fps), corresponding to an exposure
time of 10 ms per projection.
The spatial resolution of the scanner, as computed by scanning a tungsten wire
phantom, was 24 lp/cm at 10% modulation transfer function. This isotropic spatial
resolution translated into a minimal detectable feature size of approximately 200 µm.
For retrospectively gated imaging, projection images were acquired over 16 complete
rotations with a rotation time of 5 s, for a total scan time of 80 s. A tube voltage of 80
kV and a tube current of 50 mA with continuous radiation were selected. Both,
extrinsic and intrinsic motion-gated reconstruction was performed for each animal.
The reconstruction field of view for mice and rats was 4.5 cm transaxially with a
reconstruction matrix of 512 x 512 pixels and an axial slice spacing of 0.2 mm,
resulting in a voxel size of 0.08 x 0.08 x 0.20 mm3. A sharp reconstruction kernel
(H80s) was used for image reconstruction.

7

Extrinsic Gating
Extrinsic respiratory gating was performed as described previously (6, 7). A
commercial small-animal monitoring unit (1025L and Signal Breakout Module, SA
Instruments, Stony Brook, NY, USA) was used to track the respiration movements
using a pneumatic cushion. ECG signals were received from electrodes affixed to the
paws of the animal. Its waveform was processed to detect R-waves. Commercially
available software was used for synchronization of the physiological waveforms and
the acquired projection images.

Intrinsic Cardiac Gating
A region of interest (ROI) covering the heart on all projections was defined. If the
heart was not in the geometric isocenter of the CT scanner, it was automatically
tracked by a custom-developed sinusoidal ROI tracking method in the projection
images throughout the 360 degree rotation. Within this ROI, the center of mass
(COM) in the craniocaudal direction (P) was calculated from the raw projection data.
To calculate the COM, each line sum of projection values (mz) was multiplied with a
weighting factor equal to the z position of that particular line. Weighted projection
values from all lines were summed and divided by the total sum of projection values
from the ROI (M) as shown below.
zmMP z z
1 , with 
z
zmM (1)
The variations due to the angular position of the gantry had a fixed periodicity of 500
projections, reflecting the number of projections in one rotation around the animal.
The influence of angular position should be minimized in order to derive a gating
reference signal. Therefore, the COM of each projection throughout one rotation was

8
normalized to the median of the COM values at this particular angular position for all
acquired rotations. This minimized the influence of angular position on the gating
signal and revealed the cardiac pulsation in the craniocaudal direction. Despite this
processing, however, the oscillations in COM were only an approximate marker of
cardiac-pulsation due to image noise; the acquired data had to be further processed
using MATLAB (The MathWorks, Natick, MA) for motion-gated reconstruction as
described below.
Figure 1 illustrates the intrinsic gating approach and its main processing steps. The
projections acquired during a rotation of the cone-beam CT X-ray scan (Fig. 1a) were
combined with a manually selected and automatically-tracked ROI (Fig. 1b). From
this ROI, the vertical coordinate of the COM was calculated and normalized to take
the influence of angular position into consideration (Fig. 1c).
After smoothing and filtering (Fig. 1d) with a bandpass filter (180-550 bpm window for
the mouse and 80-450 bpm window for the rat), the maxima were automatically
determined by the zero-crossings of the first derivative of the waveform (Fig. 1e).

Intrinsic Respiratory Gating
Respiratory gating was performed using a similar method, whereby the manually
selected ROI covered the diaphragm instead of the heart. Neither a sinusoidal tracing
function nor a frequency filter was necessary to extract a gating reference signal,
since diaphragm movement induced changes in the COM amplitude that were much
stronger than those induced by heart movements. Here too, local maxima of the
resulting curve were used as gating reference points.

Motion-Gated Reconstruction

9
Cardiac and respiratory gating reference points were derived for every cardiac and
respiratory cycle for both extrinsic and intrinsic gating. During extrinsic gating, ECG
electrodes and a compression cushion for respiratory movement were used. As a
prerequisite for cardiac gating, respiratory gating is necessary to avoid the influence
of diaphragmatic motion during breathing excursion resulting in displacement of the
heart. All other steps such as retrospective binning of projections from several
rotations according to their cardiac phase and volumetric reconstruction using these
projection sets were identical in both gating methods. The steps involved in
retrospective binning and reconstruction are described in detail in (14) and are briefly
summarized below.
The starting point of each respiratory cycle (0% point) was defined to correspond to
the gating reference signal of every motion cycle. In order to reconstruct a given
phase of the respiratory cycle, the projections acquired within that respiratory phase
were selected for image reconstruction. Respiratory phases were defined by start
and end points, given as the percentage of the cycle length.
The selected projections from the rebinning step, representing projections pertaining
to a given phase of the respiratory cycle, were then interpolated to yield a new 360°
projection data set consisting of 600 evenly distributed projections. If two or more
selected projections were found to be at identical positions --- remember that the
angular position of each projection was recorded during different rotations --- they
were averaged to improve the signal-to-noise ratio of the interpolated projection. If no
projections were found for a selected angular position, interpolation from the closest
neighboring projections was performed. Interpolation was weighted with respect to
angular distance.
Only projections in respiratory motion-compensated, expiration phase were selected
for cardiac gated reconstruction. Optimal gating intervals depended on the animal

10
being imaged and were empirically derived through experimentation. In rats and
mice, the interval from 20% to 80% of the respiration phase was found to be optimal
for reconstruction since the most intense motion occurred either before 20% or after
80% of the respiratory phase. For cardiac gated reconstruction, respiratory motion
gating was used first. Additionally, ten phases of equal length were evenly defined
over the cardiac cycle. Cardiac motion was visible between all of these phases. The
end-diastolic phase was defined by the cardiac phase that showed the largest
ventricular volume.

Animals and Contrast Media
All experiments using rats and mice were approved by the Governmental Review
Committee on Animal Care.
To compare the two gating methods, four C3H/HeN wild type mice (20 g) and 4
healthy Copenhagen rats (250 g) were scanned.
To demonstrate potential future applications of self-gating, scans of a murine model
of infarction and a pharmacological cardiac stress test were performed.
Myocardial infarction was induced by permanent ligation of the left coronary artery in
a 12-week-old BALB/c mouse. Left coronary artery ligation was performed as
described previously (15). The infarction was confirmed through catheterization
before CT scanning and via histology after the imaging was completed.
In the stress test model, two C3H/HeN wild type mice were scanned before and
during intravenous injection of dobutamine at a dose of 30 ng/g body weight min-1
(16).
For scanning and surgery, mice and rats were anesthetized by continuous inhalation
of 3% Sevoflurane (Sevorane, Abbot, Maidenhead, UK) in oxygen during
preparation, injection of contrast media, and scanning.

11
The pneumatic cushion was attached to the animals to record the respiratory
movements. ECG electrodes were affixed to the paws to monitor the ECG waveform.
The animals were not intubated and were allowed to breathe freely throughout the
experiment.
All rodents were scanned after the administration of the intravascular contrast agent
Fenestra-VC (ART Advanced Research Technologies, Saint-Laurent, CA). A dose of
2.5 ml Fenestra-VC (50 mg iodine/ml) was injected into the tail vein of the rats 5 min
prior to scanning (1). Mice received 0.5 ml of the same contrast agent (5).

Postprocessing
The reconstructed image data sets were supplemented by a DICOM3-header to
permit their import into standard post-processing software. The CT image evaluation
used multi-planar reformations (MPR) from 4D data sets together with a commercial
workstation (InSpace Siemens Medical Solutions, Forchheim, Germany). The
Medical Imaging Interaction Toolkit (German Cancer Research Center, Heidelberg,
Germany) (17) was used to semi-automatically segment heart volumes.

Data and Statistical Analysis
Physiological data such as cardiac ejection fraction were calculated from the self-
gated 4D time series. The differences in volumes from the two ventricles during the
end-diastolic and end-systolic phases were computed.
End-diastolic and end-systolic ventricular volumes as well as cardiac ejection
fractions from intrinsic gating were compared with corresponding values from
extrinsic methods.
The correlation of the ECG and the intrinsic gating signal was used as an objective
quantitative measure for validation. We compared the median heart rate based on

12
ECG signal and the median detected heart rate based on the intrinsic cardiac motion
over 4 s for the mice and rats, resulting in 20 measurements of the heart rate during
a total scan time of 80 s. The correlation between the two methods was tested for
statistical significance using a Pearson correlation test. Calculations were made with
Excel 97 (Microsoft, Redmond, WA, USA) and BiAS for Windows (Version 8.2 -
07/2006, epsilon, Frankfurt, Germany). To test the agreement between the gating
signals of the two methods, we conducted a Bland Altman plot for the mice and rats.

13
Results
Evaluation of the Method
Mice and rats exhibited a gasping type of respiration with minimal changes in
respiratory motion and ventilation frequency once a steady state of narcosis was
reached. The respiration rate varied from 20 to 35 min-1 in mice and 20 to 39 min-1 in
rats.
The heart rates varied within a range of 343 to 428 min-1 in mice and 237 to 300 min-1
in rats.
The measured intra-individual variation in the mean heart rate was marginal for both
methods. Nevertheless, the mean standard deviation of measured heart rate was
significantly (Wilcoxon-matched-pairs test; p<0.02) higher for the extrinsic method
than for the intrinsic method (70 min-1 vs. 35 min-1 in mice, respectively; 37 min-1 vs.
23 min-1 in rats, respectively). An excellent agreement between intrinsic and extrinsic
gating signals was observed with a Pearson correlation test of 0.99 in rats (95%
confidence interval [0.99, 0.99], p<0.001) and 0.97 in mice (95% confidence interval
[0.96 - 0.98], p<0.001).
This excellent correlation is further demonstrated by Bland-Altman analysis shown in
Fig 2. A clustered dispersion of the heart rate data is seen because the four mice
(Fig. 2a) and the four rats (Fig. 2b) have different base heart rates. These results
indicate that the intrinsic signals coincide well with the extrinsic gating signal. The
mean difference between the two signals, given as (mean  SD), is (0.004  2.4) bpm
in rats and (0.9  6.8) bpm in mice.
Given such high correlation between the intrinsic and extrinsic gating signal, no
significant difference in image quality between the reconstructions employing these
two different types of gating schemes was expected. This is borne out by the images
analysis given below.

14

Quantitative Analyses of Functional Cardiac Imaging
Image reconstruction rendered 4D data sets of high quality. The volumetric CT (VCT)
offered high contrast between the iodinated blood and the myocardium and after
intrinsic gating allowed for clear delineation of epicardial and endocardial borders.
Cardiac, mediastinal, and lung anatomy were clearly depicted, and even coronary
arteries of mice were visible (Fig 3).
Table 1 shows the mean end-diastolic and end-systolic ventricular volumes as well
as cardiac ejection fractions revealed by semi-automated segmentation. The results
of left ventricular (LV), right ventricular (RV) volume, and global functional
measurements are also summarized. No outlier was found. The ventricular volumes
and the functional parameters obtained from the intrinsic and extrinsic datasets were
within one standard deviation of each other and found to be significantly equal
(p>0.95). Only for end-diastolic left ventricular volumes in mice there was a trend to 4
 3 µl higher values using the intrinsic gating method.
A literature search for the approximate values of ventricular volumes and the cardiac
ejection fractions was conducted. Values determined by self-gated and extrinsically
gated CT matched closely with those quoted in the literature (5, 8). Minor differences
in the values may derive from different animal sizes, ages, anesthesia states, the
physiological conditions under which the animals were tested, and measurement
error (18-20).

15
Example of Potential Applications (Dobutamine-Stress Cardiac VCT in Mice and
Changes in Cardiac Geometry and Function in an Infarcted Mouse)
After dobutamine injection we observed a decrease in both LV end-diastolic volume
and end-systolic volume. Ejection fraction and LV wall thickness increased during the
dobutamine test (Table 2 and Fig. 4). Online movies can be viewed in the data
supplement.
Cardiac VCT of a mouse with infarcted myocardium after LAD ligation revealed gross
dilatation of the left ventricle with marked thinning of the LV anterior wall and
complete absence of systolic thickening. Dynamic analyses revealed clear akinesia
of the infarcted myocardium during systole (Fig. 5).