Home
Search
Collections
Journals
About
Contact us
My IOPscience
Investigation of four-dimensional computed tomography-based pulmonary ventilation imaging
in patients with emphysematous lung regions
This article has been downloaded from IOPscience. Please scroll down to see the full text article.
2011 Phys. Med. Biol. 56 2279
(http://iopscience.iop.org/0031-9155/56/7/023)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 129.78.32.23
The article was downloaded on 10/05/2012 at 11:03
Please note that terms and conditions apply.
IOP PUBLISHING
PHYSICS IN MEDICINE AND BIOLOGY
Phys. Med. Biol. 56 (2011) 2279–2298
doi:10.1088/0031-9155/56/7/023
Investigation of four-dimensional computed
tomography-based pulmonary ventilation imaging in
patients with emphysematous lung regions
Tokihiro Yamamoto1,4 , Sven Kabus2 , Tobias Klinder3 , Cristian Lorenz2 ,
Jens von Berg2 , Thomas Blaffert2 , Billy W Loo Jr1 and Paul J Keall1
1
Department of Radiation Oncology, Stanford University School of Medicine, 875 Blake Wilbur
Dr, Stanford, CA 94305-5847, USA
2 Department of Digital Imaging, Philips Research Europe, Roentgenstrasse 24-26, D-22335
Hamburg, Germany
3 Clinical Informatics, Interventional, and Translational Solutions, Philips Research North
America, Briarcliff Manor, NY 10510, USA
E-mail: [email protected]
Received 23 July 2010, in final form 12 January 2011
Published 16 March 2011
Online at stacks.iop.org/PMB/56/2279
Abstract
A pulmonary ventilation imaging technique based on four-dimensional (4D)
computed tomography (CT) has advantages over existing techniques. However,
physiologically accurate 4D-CT ventilation imaging has not been achieved in
patients. The purpose of this study was to evaluate 4D-CT ventilation imaging
by correlating ventilation with emphysema. Emphysematous lung regions
are less ventilated and can be used as surrogates for low ventilation. We
tested the hypothesis: 4D-CT ventilation in emphysematous lung regions is
significantly lower than in non-emphysematous regions. Four-dimensional
CT ventilation images were created for 12 patients with emphysematous lung
regions as observed on CT, using a total of four combinations of two deformable
image registration (DIR) algorithms: surface-based (DIRsur ) and volumetric
(DIRvol ), and two metrics: Hounsfield unit (HU) change (VHU ) and Jacobian
determinant of deformation (VJac ), yielding four ventilation image sets per
patient. Emphysematous lung regions were detected by density masking.
We tested our hypothesis using the one-tailed t-test. Visually, different DIR
algorithms and metrics yielded spatially variant 4D-CT ventilation images.
The mean ventilation values in emphysematous lung regions were consistently
lower than in non-emphysematous regions for all the combinations of DIR
algorithms and metrics. VHU resulted in statistically significant differences
for both DIRsur (0.14 ± 0.14 versus 0.29 ± 0.16, p = 0.01) and DIRvol
(0.13 ± 0.13 versus 0.27 ± 0.15, p < 0.01). However, VJac resulted in
non-significant differences for both DIRsur (0.15 ± 0.07 versus 0.17 ± 0.08,
4
Author to whom any correspondence should be addressed.
0031-9155/11/072279+20$33.00
© 2011 Institute of Physics and Engineering in Medicine
Printed in the UK
2279
2280
T Yamamoto et al
p = 0.20) and DIRvol (0.17 ± 0.08 versus 0.19 ± 0.09, p = 0.30). This study
demonstrated the strong correlation between the HU-based 4D-CT ventilation
and emphysema, which indicates the potential for HU-based 4D-CT ventilation
imaging to achieve high physiologic accuracy. A further study is needed to
confirm these results.
(Some figures in this article are in colour only in the electronic version)
1. Introduction
Pulmonary function tests (PFTs) provide a global measure of pulmonary physiologic
information which is relatively insensitive to regional changes. Pulmonary diseases present
regional involvement especially during the initial phases (de Jong et al 2006) before global
measures change. Imaging techniques of regional function would further our understanding of
pathophysiological characteristics of pulmonary diseases and could also be used for functional
avoidance in lung cancer radiotherapy (Marks et al 1995, Seppenwoolde et al 2002, Christian
et al 2005, McGuire et al 2006, Shioyama et al 2007, Yaremko et al 2007, Yamamoto et al
2011). There have been several techniques of pulmonary ventilation imaging, including
nuclear medicine, magnetic resonance (MR) and computed tomography (CT). Nuclear
medicine imaging has been the only technique of ventilation imaging for the past few decades
(Alderson and Line 1980, Suga 2002, Harris and Schuster 2007). Hyperpolarized gas MRI
(Albert et al 1994, Kauczor et al 1998) has been found to be useful for the assessment of
asthma (de Lange et al 2006), cystic fibrosis (McMahon et al 2006) and emphysema (Swift
et al 2005). Xe-CT has been pioneered by Gur et al (1979, 1981) and has been studied by
several investigators (Marcucci et al 2001, Tajik et al 2002). However, these techniques have
drawbacks such as low resolution, high cost, long scan time and/or low accessibility.
A ventilation image can be created by a four-dimensional (4D) CT-based technique
which has been applied to animal subjects (Guerrero et al 2007, Reinhardt et al 2008,
Ding et al 2010) and human subjects (Guerrero et al 2005, 2006, Christensen et al 2007,
Kabus et al 2008, Vik et al 2008, Castillo et al 2010, Yamamoto et al 2010). The 4D-CTderived ventilation can be considered ‘free’ information for lung cancer radiotherapy patients,
because 4D-CT scans are routinely acquired during treatment planning at many radiotherapy
centers and ventilation computation involves only image processing, i.e. deformable image
registration (DIR). Moreover, 4D-CT ventilation imaging is higher resolution, costs less, has
a shorter scan time and is more accessible from radiotherapy centers than existing techniques.
However, variations in DIR results between algorithms have been reported (Kashani et al
2008, Brock 2009, Kabus et al 2009). Recently, two multi-institution studies were conducted
to evaluate the accuracy of various DIR methods. These studies demonstrated large variability
in the maximum error ranging from 5.1 to 15.4 mm (vector) in a phantom study (Kashani
et al 2008) and from 2.0 to 7.8 mm (SI) in a patient study (Brock 2009). Such variability
in the DIR results may influence 4D-CT ventilation imaging. In addition, two classes of
ventilation metric have been used for 4D-CT ventilation imaging: Hounsfield unit (HU)
change (Guerrero et al 2005, 2006, Fuld et al 2008, Kabus et al 2008, Castillo et al 2010,
Ding et al 2010, Yamamoto et al 2010) and Jacobian determinant of deformation (Kabus
et al 2008, Reinhardt et al 2008, Castillo et al 2010, Ding et al 2010, Yamamoto et al 2010).
The variability in the resulting ventilation images has been reported by several investigators
(Castillo et al 2010, Ding et al 2010, Yamamoto et al 2010). The physiologic accuracy of
4D-CT ventilation imaging has been investigated using both animal subjects (Fuld et al 2008,
4D-CT lung ventilation imaging in emphysema patients
2281
Reinhardt et al 2008, Ding et al 2010) and human subjects (Castillo et al 2010, Yamamoto
et al 2010). The CT (4D-CT or gated-CT) ventilation was compared with the Xe-CT ventilation
using anesthetized sheep and was found to have reasonable correlations. However, these
studies suffered from limited axial coverage of Xe-CT scans (Fuld et al 2008, Reinhardt
et al 2008, Ding et al 2010). In addition, Guerrero et al (2007) demonstrated a reduction in
mass-specific pulmonary compliance following whole lung irradiation in mice using breathhold CT scans. More recently, Castillo et al (2010) and Yamamoto et al (2010) compared
the 4D-CT and single photon emission CT (SPECT) ventilation images in thoracic cancer
patients. Castillo et al (2010) reported relatively high Dice similarity coefficients between the
4D-CT and SPECT ventilation in low-functional regions, however low similarity overall. They
claimed that the low similarity was likely due to central airway depositions of the technetium99m-labeled diethylenetriamine pentaacetate (99mTc-DTPA) aerosols. Yamamoto et al (2010)
also reported low correlations between the 4D-CT and SPECT ventilation (r = 0.18). They
found much higher correlations between the 4D-CT ventilation and SPECT perfusion (r =
0.48), which is expected to highly correlate with ventilation. Given the lack of data showing
high correlations for the entire lungs, physiologically accurate 4D-CT ventilation imaging has
not been achieved and further studies are necessary.
The purpose of this study was to evaluate 4D-CT ventilation imaging in patients with
emphysematous lung regions by correlating ventilation with emphysema. In emphysema,
there is destruction of alveolar septa leading to decreased elastic recoil of the alveoli and
radial traction, which help hold small airways open (Levitzky 2007). Thus, emphysematous
lung regions are less ventilated (Zaporozhan et al 2004, Spector et al 2005), and can be
used as surrogates for low ventilation (Ley-Zaporozhan et al 2007). We tested the following
hypothesis: 4D-CT ventilation in emphysematous lung regions is significantly lower than
in non-emphysematous lung regions, which is a necessary condition for a physiologically
accurate 4D-CT ventilation imaging. Given that there are various DIR algorithms and two
classes of ventilation metrics that can be used for 4D-CT ventilation imaging, we investigated
a total of four combinations of two DIR algorithms and two metrics.
2. Methods and materials
2.1. Patients
This study was a retrospective analysis approved by Stanford University’s Institutional Review
Board. We studied 12 patients with emphysematous lung regions as observed on CT, who
underwent 4D-CT scanning as well as radiotherapy for the thoracic cancers. We confirmed
that all the patients had emphysematous lung regions based on the CT quantification described
below in section 2.4. The characteristics of the patients are described in table 1. Half of
patients had stage I non-small-cell lung cancer (NSCLC). Two of the 11 NSCLC patients had
two tumors.
2.2. 4D-CT pulmonary ventilation imaging
Figure 1 shows a schematic diagram for creating a ventilation image from 4D-CT and
comparing the ventilation in emphysematous and non-emphysematous lung regions. Fourdimensional CT ventilation imaging consists of three steps as follows. The first step is to
acquire 4D-CT images for radiotherapy planning purposes. At Stanford, we routinely acquire
4D-CT scans for thoracic and abdominal cancer. Four-dimensional CT images are created
by acquiring oversampled CT data simultaneously with a respiratory trace and reconstructing
2282
T Yamamoto et al
Table 1. Characteristics of the 12-patient cohort.
Parameter
Value
Age (y/o), median (range)
Smoking history (pack-year), median (range)
Gender, n (%)
Male
Female
Histology and stage, n (%)
NSCLCa, stage I
NSCLCa, stage II
NSCLCa, stage III
Follicular lymphoma, stage IV
Lung tumor location, n (%)
RULb
RLLc
LULd
LLLe
76 (62—90)
50 (10–165)
8/12 (66.7)
4/12 (33.3)
6/12 (50.0)
2/12 (16.7)
3/12 (25.0)
1/12 (8.3)
3/14 (21.4)
3/14 (21.4)
5/14 (35.7)
3/14 (21.4)
a
Non-small-cell lung cancer.
Right upper lobe.
c
Right lower lobe.
d
Left upper lobe.
e
Left lower lobe.
b
a number of 3D-CT data sets correlated with a given respiratory phase range (Rietzel et al
2005). We acquired 4D-CT scans on the GE Discovery ST multislice positron emission CT
(PET)/CT scanner (GE Medical Systems, Waukesha, WI) in cine mode with the Varian Realtime Position Management (RPM) system (Varian Medical Systems, Palo Alto, CA) to record
patient respiratory traces. The CT data were acquired at multiple couch positions covering
the entire lung for a cine duration that is a little longer than the estimated respiratory period
at each position, resulting in approximately 15 time-resolved images at each slice position.
Scan parameters were set as follows: 120 kVp, approximately 100 mAs per slice, 0.5 s gantry
rotation, 0.45 s cine interval and 2.5 mm slice thickness as used clinically in our radiation
oncology department. The GE Advantage 4D software was used to create a 4D-CT image set
by sorting raw 4D-CT slices correlated with the RPM data into 10 respiratory phase-based bins
(i.e. 0–90% at 10% intervals). We used paired 4D-CT images at the peak-exhale (typically
50%) and peak-inhale (typically 0%) phases for ventilation computation which were identified
based on the dome of diaphragm showing the most superior position and the most inferior
position, respectively. More details on the 4D-CT acquisition using the GE scanner with
Advantage 4D have been described by Rietzel et al (2005).
The second step of 4D-CT ventilation imaging is DIR for spatial mapping of the peakexhale CT image to the peak-inhale image, deriving a displacement vector field (DVF).
We investigated surface-based registration (DIRsur ) (von Berg et al 2007) and volumetric
registration (DIRvol ) (Kabus and Lorenz 2010). Kabus et al validated the geometric accuracy
of these algorithms by evaluating the distances between landmark positions at two different
phases with and without registration, which were reduced from 6.0 mm to 2.3–2.5 mm on
average for the four patients (Kabus et al 2008). We also performed validation using a publicly
available data set of five cases with 300 landmarks for each (Castillo et al 2009). On average,
4D-CT lung ventilation imaging in emphysema patients
2283
Acquire 4D-CT scan
DIR sur
Perform
DIR vol
Create DVF sur
Create
DVF vol
Perform
Detect
emphysema by
density masking
Create
sur
VHU
Create
sur
VJac
Create
vol
VHU
Create
vol
VJac
Correlate
Compare V in
emphysematous and
non-emphysematous
lung regions
Figure 1. A schematic diagram for creating a ventilation image from 4D-CT and comparing
the ventilation in emphysematous and non-emphysematous lung regions. The first step was the
acquisition of a 4D-CT image set. The second step was deformable image registration (DIR) for
spatial mapping of the peak-exhale CT image to the peak-inhale image, deriving a displacement
vector field (DVF). The third step was the creation of a 4D-CT ventilation image through the
computation of a ventilation metric. Given that there are various DIR algorithms and two classes
of ventilation metrics that can be used for 4D-CT ventilation imaging, we studied a total of four
combinations of two DIR algorithms: surface-based (DIRsur ) and volumetric (DIRvol ), and two
metrics: Hounsfield unit (HU) change (VHU ) and Jacobian determinant of deformation (VJac ),
sur , V sur , V vol and V vol ). The final step was to
yielding four ventilation image sets per patient (VHU
HU
Jac
Jac
correlate resultant 4D-CT ventilation images with emphysema.
the distances were reduced from 3.9–9.8 mm to 1.0–1.7 mm. Given the slice thickness of
2.5 mm and an average observer error of 0.7–1.1 mm, both DIRsur and DIRvol are expected to
achieve high physiologic accuracy. However, different DIR methods yielded varying DVFs in
regions apart from the landmarks despite having similar and small mean landmark registration
errors overall, which motivated further investigation of these two algorithms in this study.
DIRsur matches surface models at the peak-exhale phase (i.e. lung surface along with inner
structures such as vessel trees) to those at the peak-inhale phase, followed by thin plate spline
interpolation to create a dense vector field. Further details on DIRsur have been described
by von Berg et al (2007). DIRvol is volumetric by itself and tries to find a vector field that
minimizes both a similarity function (i.e. the sum of squared difference between the peakinhale and deformed peak-exhale images) and a regularizing term (i.e. elastic regularizer)
based on the Navier–Lamé equation. The Navier–Lamé equation has two parameters: λ and
µ, which can be converted into the parameters better known as Young’s modulus and Poisson’s
ratio, respectively. Further details on DIRvol has been described by Kabus and Lorenz (2010).
We used algorithm parameters identical to those previously investigated (Kabus et al 2008)
for both DIRsur and DIRvol . Given that emphysematous lungs have different elastic properties
2284
T Yamamoto et al
than healthy lungs, we studied the effect of elasticity parameter of DIRvol by using three
different µ values. The µ value of 0.0025 was used for the baseline elasticity setting (Kabus
et al 2008), which was decreased and increased by a factor of four to simulate a more elastic
and stiff material, respectively. The λ value was set to 0 to reflect a compressible material in
all settings. The elasticity settings were global and the entire lung was set to these parameters.
The accuracy of DIR may be improved with regionally different elasticity parameters. This
was outside the scope of this study. DIRvol denotes the baseline setting unless otherwise
noted.
The final step of 4D-CT ventilation imaging is the creation of a ventilation image at
the peak-exhale phase through quantitative analysis of the DVF. We investigated HU change
(Guerrero et al 2005, 2006, Fuld et al 2008, Kabus et al 2008, Castillo et al 2010, Ding
et al 2010, Yamamoto et al 2010) and Jacobian determinant of deformation (Kabus et al
2008, Reinhardt et al 2008, Castillo et al 2010, Ding et al 2010, Yamamoto et al 2010),
which were the only two classes of ventilation metrics proposed previously. Both metrics
are based on the assumption that regional ventilation is proportional to regional volume
change, which is supported by the literature, i.e. the HU metric (Fuld et al 2008) and Jacobian
metric (Reinhardt et al 2008) were found to have reasonable correlations with Xe-CT-measured
regional ventilation in sheep. For the HU metric, Simon (2000) originally derived a relationship
between the local change in fractional air content and local volume change, which was adapted
to the relationship between the local HU density change and local volume change by Guerrero
et al (2005). The exhale-to-inhale volume change ("Vol) normalized by the exhale air volume
(Volair
ex ) in the voxel at location (x, y, z) is given by
"Vol
Volair
ex (x, y, z)
HUin {x + ux (x, y, z), y + uy (x, y, z), z + uz (x, y, z)} − HUex (x, y, z)
= 1000
,
HUex (x, y, z)[HUin {x + ux (x, y, z), y + uy (x, y, z), z + uz (x, y, z)} + 1000]
(1)
where HU is the HU value and u is the displacement vector mapping the voxel at location
(x, y, z) of a peak-exhale image to the corresponding location of a peak-inhale image. Note
that the air and tissue densities were assumed to be −1000 and 0 HU, respectively. To date
equation (1) has been used as a ventilation metric by many investigators (Guerrero et al
2006, Fuld et al 2008, Kabus et al 2008, Castillo et al 2010, Ding et al 2010, Yamamoto et
al 2010). However, the value independent of the initial air volume was defined as the HU
ventilation metric (VHU ) in this study, considering the findings from a hyperpolarized 3He
study by Spector et al (2005). They demonstrated remarkably lower ventilation, which was
defined as the amount of 3He gas added to a region of interest normalized by the total lung
volume of that region (i.e. proportional to the absolute volume change), in emphysematous
rats than in healthy rats. The exhale air volume (Volair
ex ) in the voxel at location (x, y, z) can
be estimated by
Volair
ex (x, y, z) = −
HUex (x, y, z) voxel
Volex (x, y, z),
1000
(2)
where Volvoxel
is the exhale voxel volume (Hoffman and Ritman 1985). Substitution of
ex
equation (2) into equation (1) yields
"Vol =
HUex (x, y, z) − HUin {x + ux (x, y, z), y + uy (x, y, z), z + uz (x, y, z)}
HUin {x + ux (x, y, z), y + uy (x, y, z), z + uz (x, y, z)} + 1000
× Volvoxel
ex (x, y, z).
(3)
4D-CT lung ventilation imaging in emphysema patients
2285
Given that Volvoxel
is the same for all voxels, we defined the HU ventilation metric (VHU ) as
ex
VHU (x, y, z) =
HUex (x, y, z) − HUin {x + ux (x, y, z), y + uy (x, y, z), z + uz (x, y, z)}
.
HUin {x + ux (x, y, z), y + uy (x, y, z), z + uz (x, y, z)} + 1000
(4)
A mass correction was applied to HUin to account for the difference in CT-derived lung mass
which would be due to the changes in blood distribution between exhale and inhale in the
same manner as Guerrero et al (2006). HUex and HUin at the same location of the deformed
peal-exhale and peak-inhale images were used to compute VHU which was mapped back to the
original peak-exhale image domain to create a 4D-CT ventilation image. Given that VHU was
based on HU values and influenced by the statistical noise, the 4D-CT images were smoothed
using an isotropic Gaussian filter kernel before computing VHU . We studied the effect of
smoothing by using small (variance, σ 2 = 1.5 mm2) and large (σ 2 = 5 mm2) smoothing levels.
VHU denotes the HU metric based on the small smoothing level unless otherwise noted. For
the Jacobian metric, the Jacobian determinant (J ) of the displacement vector u is given by
!
!
!
∂ux (x, y, z)
∂ux (x, y, z) !!
!1 + ∂ux (x, y, z)
!
!
∂x
∂y
∂z
!
!
!
! ∂uy (x, y, z)
∂u
(x,
y,
z)
(x,
y,
z)
∂u
y
y
!.
!
1+
(5)
J (x, y, z) = !
!
∂x
∂y
∂z
!
!
! ∂uz (x, y, z)
∂uz (x, y, z)
∂uz (x, y, z) !!
!
1+
!
!
∂x
∂y
∂z
" voxel #
The volume of voxel deformed into the inhale phase Volin
can be estimated by
= Volvoxel
Volvoxel
in
ex J (x, y, z).
(6)
{J (x, y, z) − 1} .
− Volvoxel
= Volvoxel
"Vol = Volvoxel
in
ex
ex
(7)
VJac (x, y, z) = J (x, y, z) − 1.
(8)
The exhale-to-inhale volume change ("Vol) is expressed as
Given that
as
Volvoxel
ex
is the same for all voxels, we defined the Jacobian ventilation metric (VJac )
For both VHU and VJac , a value of zero corresponds to local volume preservation (i.e. zero
ventilation). A value smaller than zero indicates local contraction and a value larger than zero
indicates local expansion. The VHU or VJac values outside the segmented lung parenchyma
volumes have been zeroed. The lung volume was segmented by delineating lung voxels with
HU values less than a threshold of −600 (Shikata et al 2004) or −250 (Guerrero et al 2006,
Castillo et al 2010) within the lung outlines generated by the model-based segmentation of the
Pinnacle3 treatment planning system (Philips Radiation Oncology Systems, Fitchburg, WI).
Manual trimming of the central airways and great vessels was also performed where necessary.
Castillo et al (2010) used morphological growing to remove the central airways.
2.3. Comparison of the calculated and measured tidal volumes
The absolute tidal volumes calculated by 4D-CT ventilation were compared to those measured
by segmented lung parenchyma volumes (ground truth) to investigate the global accuracy of
4D-CT ventilation computation. For the calculated tidal volume, the exhale-to-inhale volume
changes were calculated for all lung parenchyma voxels at the peak-exhale phase based on
equations (3) for VHU and (7) for VJac , which were then integrated to determine a tidal volume.
For the measured tidal volume, the air volumes in the peak-exhale and peak-inhale lungs were
estimated based on equation (2) and were then subtracted to determine a tidal volume. The
2286
T Yamamoto et al
measured and calculated tidal volumes were determined for 12 emphysema patients. The
Pearson’s linear correlation coefficients between the calculated and measured tidal volumes
were determined for four ventilation image sets.
2.4. Emphysema quantification
Emphysematous lung regions were detected by density masking using MATLAB (The
MathWorks, Natick, MA) where CT voxels with HU values less than −910 within the lungs
were identified as emphysema. The density masking technique has been validated against
pathology (Hayhurst et al 1984, Muller et al 1988, Gevenois et al 1996, Bankier et al 1999)
and PFT (Kinsella et al 1990, Gould et al 1991, Gevenois et al 1996, Haraguchi et al 1998,
Park et al 1999, Arakawa et al 2001, Baldi et al 2001). The mean percentage low-attenuation
area below −910 HU (%LAA) for the study subjects was 21.9 ± 15.2%. Also, a metric
of global emphysema severity, i.e. the mean 15th percentile HU value below which 15% of
the lung voxels are distributed (Parr et al 2006, Newell 2008, Parr et al 2008), was −971 ±
21 HU. Peak-exhale CT images were used for emphysema quantification, given that exhale
high-resolution CT (HRCT) scans have been reported to show better correlations with PFTs
than inhale scans (Kauczor et al 2000, 2002, Arakawa et al 2001, Spiropoulos et al 2003,
Zaporozhan et al 2005).
2.5. Statistical analysis
Statistical analyses were performed to test whether the 4D-CT ventilation in emphysematous
lung regions is significantly lower than in non-emphysematous" lung regions (p < 0.05) #using
vol
vol
sur
sur
.
, VJac
, VHU
and VJac
the one-tailed t-test for the four 4D-CT ventilation image sets VHU
3. Results
3.1. Differences between two DVFs for the 12-patient cohort
Table 2 shows the mean length of 3D displacement vectors for DVFsur , DVFvol , and the
difference between these two DVFs for 12 patients. For the −600 HU threshold, the vector
difference was 2.0 ± 1.0 mm, i.e. both the mean and SD were smaller than the voxel
dimension of the image set. The differences were larger in non-emphysematous regions than
in emphysematous regions, however they were smaller than the voxel dimension. The larger
differences in non-emphysematous regions were likely due to larger displacement magnitudes
than in emphysematous regions. The −250 HU threshold demonstrated similar results.
3.2. Correlations between the calculated and measured tidal volumes
Figure 2 shows scatter plots for the absolute tidal volumes for the four 4D-CT
ventilation image sets of 12 patients. There were strong correlations between the VHU calculated and measured tidal volumes for both −600 HU and −250 HU thresholds
(figure 2(a)). The slopes of the least-squares regression lines of the −600 HU threshold
vol
sur
and VHU
. The
were closer to unity compared to the −250 HU threshold for both VHU
differences between the measured and calculated tidal volumes could be due to non-optimal
registration of high-contrast structures (e.g. small pulmonary vessels) and/or uncertainties in
lung segmentation. Even spatially small misalignments of high-contrast structures cause one
shadow region with a positive sign and the other region with a negative sign, resulting in
extreme erroneous VHU values. This effect can be reduced with accurate lung parenchyma
4D-CT lung ventilation imaging in emphysema patients
2287
2000
2000
sur
VJac (-600 HU threshold, r = 0.95)
surJac-600
1500
1000
sur
(-600 HU threshold, r = 0.98)
surHU-600
VHU
500
vol
(-600 HU threshold, r = 0.98)
volHU-600
VHU
sur
(-250 HU threshold, r = 0.98)
VHU
surHU-250
Calculated absolute tidal volume (ml)
Calculated absolute tidal volume (ml)
vol
volJac-600
VJac (-600 HU threshold, r = 0.98)
sur
surJac-250
(-250 HU threshold, r = 0.99)
VJac
1500
vol
volJac-250
(-250 HU threshold, r = 0.98)
VJac
1000
500
vol
VHU (-250 HU threshold, r = 0.96)
volHU-250
0
0
500
1000
1500
2000
0
0
Measured absolute tidal volume (ml)
500
1000
1500
2000
Measured absolute tidal volume (ml)
(a)
(b)
Figure 2. Correlations between the absolute tidal volumes measured by segmented lung
parenchyma volumes in the 4D-CT images and those calculated by 4D-CT ventilation derived
by (a) the HU metric (VHU ) and (b) the Jacobian metric (VJac ) for 12 emphysema patients. The
results are shown for the two thresholds of lung parenchyma segmentation (−600 and −250 HU).
The lines of best fit are also shown: solid (−600 HU) and dashed (−250 HU).
Table 2. Mean length of 3D displacement vectors (mean ± SD, mm) for DVFsur , DVFvol and the
difference between these two DVFs for 12 emphysema patients. The results are shown for the two
thresholds of lung parenchyma segmentation (−600 and −250 HU).
Threshold
−600 HU
−250 HU
DIR algorithm
DVFsur
DVFvol
Difference
DVFsur
DVFvol
Difference
Total lung
Emphysema
Non-emphysema
7.9 ± 2.3
8.5 ± 2.7
2.0 ± 1.0
8.2 ± 2.4
8.8 ± 2.8
2.0 ± 1.1
6.2 ± 2.0
6.4 ± 2.0
1.4 ± 0.7
6.2 ± 2.0
6.4 ± 2.0
1.4 ± 0.7
8.4 ± 2.3
9.0 ± 2.8
2.1 ± 1.1
8.6 ± 2.4
9.4 ± 2.9
2.2 ± 1.1
segmentation. Visually, the −600 HU threshold resulted in less high-contrast structures in
the segmented volumes than the −250 HU threshold, and hence the effects of non-optimal
registration were considered to be small and the slopes were closer to unity. There were
also strong correlations between the VJac -calculated and measured tidal volumes for both
−600 HU and −250 HU thresholds (figure 2(b)). The slopes of the regression lines of the
−600 HU threshold were much smaller than unity, while the slopes of the −250 HU threshold
vol
sur
and VJac
. This result was probably due to opposing effects
were close to unity for both VJac
of threshold values on the VJac -calculated and measured tidal volumes. For the calculated
tidal volumes, the −600 HU threshold masked out more voxels that normally have positive
VJac values and resulted in smaller tidal volumes overall compared to the −250 HU threshold.
For the measured tidal volumes, in contrast, the −600 HU threshold gave relatively smaller
lung volumes than the −250 HU threshold at peak exhale (average relative volume difference,
14 ± 5%) while relatively similar lung volumes at peak inhale (11 ± 3%), resulting in larger
tidal volumes. The optimal threshold for lung parenchyma segmentation remains an open
2288
T Yamamoto et al
Coronal
sur
VHU
CT
vol
VHU
sur
VJac
vol
VJac
Emphysema
not overlaid
V
Emphysema
overlaid
Axial
(a)
sur
VHU
CT
sur
VJac
vol
VHU
vol
VJac
Emphysema
not overlaid
Emphysema
overlaid
sur
VJac
0.5
1.6
3.5
1.4
3
1.2
2.5
1
Probability
Non-emphysema
Probability
(b)
Probability
Emphysema
1
vol
VHU
4
2
1.5
-0.5
0
Ventilation
0.5
1
0
-1
4
0.8
0.6
0.4
1
3
2
1
0.2
0.5
0
-1
vol
VJac
5
Probability
sur
VHU
1.5
-0.5
0
Ventilation
0.5
1
0
-1
-0.5
0
Ventilation
0.5
1
0
-1
-0.5
0
Ventilation
0.5
1
Figure 3. (a) Example coronal and axial images at the same level of peak-exhale CT and four
sur , V sur , V vol and V vol ) for patient 3 showing the largest difference between
4D-CT ventilation (VHU
HU
Jac
Jac
VHU values in emphysematous and non-emphysematous lung regions. (b) Probability density
functions of 4D-CT ventilation in emphysematous and non-emphysematous lung regions are also
shown. The lung parenchyma volumes were segmented with the −600 HU threshold.
question and is further discussed in section 4. The results for both −600 HU and −250 HU
thresholds are presented in this report.
3.3. 4D-CT ventilation in emphysematous versus in non-emphysematous lung regions for
example patients
Figure 3 shows example images of peak-exhale CT and four 4D-CT ventilation for patient 3
(90 year old male, %LAA = 11.6%) demonstrating the largest difference between VHU values
in emphysematous and non-emphysematous lung regions. Visually, different DIR algorithms
and metrics yielded spatially variant 4D-CT ventilation images. There were clearly less
highly ventilated lung voxels in emphysematous regions (or more highly ventilated lung
vol
sur
and VHU
. In contrast, there were mixtures
voxels in non-emphysematous regions) for VHU
vol
sur
of highly and poorly ventilated lung voxels in emphysematous regions for VJac
and VJac
as observed in the axial images. The peaks of the probability density functions of 4D-CT
ventilation in emphysematous and non-emphysematous lung regions were clearly separated
sur
(mean ventilation, 0.17 ± 1.32 in emphysema versus 0.47 ± 0.60 in
from each other for VHU
vol
(0.15 ± 1.12 versus 0.46 ± 0.50), while they almost overlapped
non-emphysema) and VHU
4D-CT lung ventilation imaging in emphysema patients
2289
Coronal
sur
VJac
sur
VHU
CT
vol
VJac
vol
VHU
Emphysema
not overlaid
V
Emphysema
overlaid
Axial
(a)
sur
VHU
CT
sur
VJac
vol
VJac
vol
VHU
Emphysema
not overlaid
Emphysema
overlaid
sur
VHU
sur
VJac
1
6
4
2
0.5
0
-1
2.5
8
Non-emphysema
1.5
-0.5
0
Ventilation
0.5
1
0
-1
10
8
2
Probability
Emphysema
vol
VJac
3
Probability
2
Probability
(b)
Probability
2.5
vol
VHU
10
3
1.5
1
0
Ventilation
0.5
1
0
-1
4
2
0.5
-0.5
6
-0.5
0
Ventilation
0.5
1
0
-1
-0.5
0
Ventilation
0.5
1
Figure 4. (a) Example coronal and axial images at the same level of peak-exhale CT and four 4Dsur , V sur , V vol and V vol ) for patient 7 showing the smallest difference between
CT ventilation (VHU
HU
Jac
Jac
VHU values in emphysematous and non-emphysematous lung regions. (b) Probability density
functions of 4D-CT ventilation in emphysematous and non-emphysematous lung regions are also
shown. The lung parenchyma volumes were segmented with the −600 HU threshold.
vol
sur
each other with no separations for VJac
(0.24 ± 0.23 versus 0.29 ± 0.30) and VJac
(0.26 ±
0.17 versus 0.30 ± 0.16) (figure 3(b)). Other example images for patient 7 (62 year old male,
%LAA = 7.0%) demonstrating the smallest difference between VHU values in emphysematous
and non-emphysematous lung regions are shown in figure 4. Compared to patient 3, there
were relatively more mixtures of highly and poorly ventilated lung voxels in emphysematous
sur
regions, though the peaks of the two probability density functions were still separated for VHU
vol
(0.23 ± 1.67 versus 0.24 ± 0.23) and VHU (0.19 ± 1.58 versus 0.23 ± 0.19). No separations
vol
sur
were observed for VJac
(0.16 ± 0.06 versus 0.16 ± 0.09) and VJac
(0.19 ± 0.07 versus
0.17 ± 0.09) as were observed in patient 3. For both VHU and VJac (especially VHU ), there were
considerable probabilities of negative ventilation values that would not, in principle, appear
during inhalation. Although the geometric accuracy of the DIR algorithms used in this study
has been validated as described above, we cannot rule out that the negative ventilation was
caused by residual errors in the DIR process. In a separate study, significant lung regions with
negative ventilation values were also observed by Christensen et al (2007). However, this still
leaves open the question of DIR algorithm errors.
2290
T Yamamoto et al
Table 3. Mean 4D-CT ventilation values in emphysematous and in non-emphysematous lung
regions of 12 patients for the four combinations of two deformable image registration (DIR)
algorithms and two ventilation metrics. The results are shown for the two thresholds of lung
parenchyma segmentation (−600 and −250 HU).
Ventilation, mean ± SD
Threshold DIR algorithm Metric Emphysema Non-emphysema
−600 HU
DIRsur
DIRvol
−250 HU
DIRsur
DIRvol
VHU
VJac
VHU
VJac
VHU
VJac
VHU
VJac
0.14 ± 0.14
0.15 ± 0.07
0.13 ± 0.13
0.17 ± 0.08
0.13 ± 0.08
0.15 ± 0.07
0.13 ± 0.07
0.17 ± 0.08
0.29 ± 0.16
0.17 ± 0.08
0.27 ± 0.15
0.19 ± 0.09
0.22 ± 0.12
0.18 ± 0.08
0.18 ± 0.08
0.18 ± 0.09
p-value
0.01
0.20
<0.01
0.30
0.03
0.19
0.06
0.36
3.4. 4D-CT ventilation in emphysematous versus in non-emphysematous lung regions for the
12-patient cohort
Table 3 shows a summary of the mean 4D-CT ventilation values in emphysematous and
non-emphysematous lung regions for the 12 patients. Different DIR algorithms and metrics
yielded variant ventilation values. The mean ventilation values in emphysematous regions
were consistently lower than in non-emphysematous regions for any combination of DIR
algorithms and ventilation metrics for both −600 HU and −250 HU thresholds. However,
the differences were statistically significant only for VHU and not for VJac . VHU resulted in
significantly lower ventilation in emphysematous regions than in non-emphysematous regions
for both DIRsur (p = 0.01) and DIRvol (p < 0.01) for the −600 HU threshold. However,
VJac resulted in non-significant differences for both DIRsur (p = 0.20) and DIRvol (p = 0.30).
The −250 HU threshold demonstrated similar results, except for VHU which resulted in
non-significant differences for DIRvol (p = 0.06) due to smaller mean VHU values in nonemphysematous regions than the −600 HU threshold. Aligned high-contrast structures give
small VHU values, and hence are considered to lower average values for the −250 HU threshold
compared to the −600 HU threshold. The elasticity parameter of DIRvol did not show a clear
impact on the statistical significance as shown in table 4, except for VHU for the −250 HU
threshold. The p-values decreased with decreasing elasticity, which was likely driven by nonoptimal registration of high-contrast structures. Misalignments of high-contrast structures
result in extremely high VHU values and contribute to increasing the average values. The
image registration with the elastic setting is expected to have more freedom to deform an
image than the stiff setting such that the sum of squared difference becomes smaller. The
stiff setting might result in more misalignments of high-contrast structures, i.e. high average
VHU values, than the elastic setting. The volumes segmented with the −600 HU threshold
contained less high-contrast structures than the −250 HU threshold, and hence the effect of
elasticity parameter was relatively small. For the −600 HU threshold, the mean VHU values in
non-emphysematous regions increased from 0.25 for the elastic setting to 0.27 for the baseline
setting and to 0.29 for the stiff setting. For the −250 HU threshold, the mean values were
0.15, 0.18 and 0.21, respectively. The image smoothing level of VHU had a clear impact on
the statistical significance as shown in table 5. The large smoothing level (σ 2 = 5 mm2) led
to statistically non-significant differences. The large smoothing level resulted in larger mean
4D-CT lung ventilation imaging in emphysema patients
2291
Table 4. Mean 4D-CT ventilation values in emphysematous and in non-emphysematous lung
regions of 12 patients for the four combinations of volumetric deformable image registration
(DIRvol ) with three elasticity settings and two ventilation metrics. The results are shown for the
two thresholds of lung parenchyma segmentation (−600 and −250 HU).
Threshold
−600 HU
Elasticity setting
Metric
Elastica
VHU
VJac
VHU
VJac
VHU
VJac
VHU
VJac
VHU
VJac
VHU
VJac
Baselineb
Stiffc
−250 HU
Elastica
Baselineb
Stiffc
Ventilation, mean ± SD
Emphysema Non-emphysema
0.12 ± 0.12
0.19 ± 0.10
0.13 ± 0.13
0.17 ± 0.08
0.12 ± 0.13
0.15 ± 0.07
0.11 ± 0.06
0.19 ± 0.10
0.13 ± 0.07
0.17 ± 0.08
0.13 ± 0.08
0.15 ± 0.07
0.25 ± 0.13
0.20 ± 0.10
0.27 ± 0.15
0.19 ± 0.09
0.29 ± 0.15
0.17 ± 0.07
0.15 ± 0.07
0.18 ± 0.09
0.18 ± 0.08
0.18 ± 0.09
0.21 ± 0.10
0.16 ± 0.07
p-value
<0.01
0.37
<0.01
0.30
<0.01
0.33
0.07
0.53
0.06
0.36
0.02
0.35
a
Elastic setting with µ = 0.0025/4 and λ = 0.
Baseline setting with µ = 0.0025 and λ = 0.
c
Stiff setting with µ = 0.0025 × 4 and λ = 0.
b
Table 5. Mean 4D-CT ventilation values in emphysematous and in non-emphysematous lung
regions of 12 patients for the four combinations of two deformable image registration (DIR)
algorithms and the HU metric with two smoothing levels. The results are shown for the two
thresholds of lung parenchyma segmentation (−600 and −250 HU).
Threshold
−600 HU
DIR algorithm
Smoothing
DIRsur
Smalla
Largeb
Smalla
Largeb
Smalla
Largeb
Smalla
Largeb
DIRvol
−250 HU
DIRsur
DIRvol
a
b
Ventilation, mean ± SD
Emphysema Non-emphysema
0.14 ± 0.14
0.19 ± 0.14
0.13 ± 0.13
0.18 ± 0.14
0.13 ± 0.08
0.16 ± 0.08
0.13 ± 0.07
0.15 ± 0.07
0.29 ± 0.16
0.25 ± 0.15
0.27 ± 0.15
0.24 ± 0.14
0.22 ± 0.12
0.20 ± 0.10
0.18 ± 0.08
0.16 ± 0.07
p-value
0.01
0.15
<0.01
0.16
0.03
0.17
0.06
0.34
Gaussian filter kernel with the smaller variance, σ 2 = 1.5 mm2.
Gaussian filter kernel with the larger variance, σ 2 = 5 mm2.
ventilation values than the small smoothing level (σ 2 = 1.5 mm2) in emphysematous regions,
however resulted in smaller mean ventilation values in non-emphysematous regions for both
DIRsur and DIRvol . Nevertheless, the mean ventilation values in emphysematous regions were
still lower than in non-emphysematous regions.
2292
T Yamamoto et al
4. Discussion
Four-dimensional CT ventilation imaging was evaluated by correlating ventilation with
emphysematous lung regions, which are less ventilated and were used as surrogates for
low ventilation, in a 12-patient study. A total of four combinations of two DIR algorithms
(DIRsur and DIRvol ) and two metrics (VHU and VJac ) were investigated. VHU was found to
be significantly lower in emphysematous regions than in non-emphysematous regions. VJac
was also found to be lower in emphysematous regions than in non-emphysematous regions;
however, the differences were not significant. Recently, Castillo et al (2010) demonstrated
significantly higher Dice similarity coefficients between the HU-based 4D-CT ventilation and
the SPECT ventilation than those between the Jacobian-based 4D-CT and SPECT ventilation
for seven patients (p < 10−4), which is consistent with our results. Reinhardt et al (2008)
found reasonably high correlations (average, R2 = 0.73) between the average ventilation values
in 4 mm thick sub-regions determined by the Jacobian-based 4D-CT and Xe-CT ventilation
for five anesthetized sheep. They might have obtained higher correlations in the HU-based
ventilation than the Jacobian-based ventilation if both metrics were evaluated, as with our
findings. More recently, Ding et al (2010) proposed a hybrid ventilation metric combining
the HU and Jacobian metrics and demonstrated consistently higher correlations with the XeCT ventilation than the HU-based ventilation for three anesthetized sheep. The hybrid metric
might result in higher correlations than VHU and/or VJac in our study as well. Both VHU and VJac
represent the ventilation values independent of the initial air volume and, in principle, should
achieve a similar performance to each other. Differences between the two metrics might
be due to (1) residual DIR errors, (2) 4D-CT image noise, (3) 4D-CT artifacts and/or (4)
inhomogeneous changes in the blood distribution during respiration. Given that an agreement
between the two metrics means that the volume change estimated from the HU density change
between the voxels at the peak-exhale and peak-inhale phases agrees with that estimated from
the displacement vectors in and around the corresponding voxel, residual DIR errors would
result in disagreements between the two metrics. Even spatially small DIR errors could result
in large disagreements. A general discussion point is that the accuracy of 4D-CT ventilation
imaging will always be limited by that of the DIR algorithms, which in turn are very difficult to
quantify for individual cases or even away from landmarks in validation studies. It is therefore
difficult to identify a specific cause of the disagreements between the two metrics. However,
improvements in DIR algorithm performance are likely to improve the quality of ventilation
images for future studies. For the 4D-CT image noise, the smoothing level for VHU had a
considerable impact on the statistical significance of the differences in ventilation between
emphysematous and non-emphysematous regions, though the ventilation in emphysematous
regions was consistently lower than in non-emphysematous regions for both smoothing levels.
The optimal smoothing level remains an open question and will be investigated in a future
study. The 4D-CT artifacts may also influence ventilation computation and the effect may be
different between the two metrics. Possible impact of 4D-CT artifacts are discussed in more
detail below. A mass correction was applied in the VHU calculation to account for the difference
in CT-derived lung mass due to respiration, where we assumed homogeneous changes in the
blood distribution. However, the distribution may be inhomogeneous and could influence the
VHU calculation.
The DIR algorithms have been found to have a small impact on the statistical significance
of the differences in ventilation between emphysematous and non-emphysematous regions,
which is most likely due to similar performance of DIRsur and DIRvol . The differences between
DVFsur and DVFvol were smaller than the voxel dimension of the image set on average. Kabus
et al (2009) demonstrated relatively remarkable differences in DVFs between DIRvol and three
4D-CT lung ventilation imaging in emphysema patients
2293
other volumetric DIR algorithms compared to the differences between DIRvol and DIRsur or the
other volumetric DIR algorithm for one case, despite having similar and small mean landmark
registration errors ranging from 1.0 to 1.4 mm for all of these algorithms. It should be noted
that the two algorithms were chosen in the current study independently of these results and
there were still considerable differences between DVFsur and DIRvol . Nevertheless, there may
be larger impact of DIR algorithms if another DIR algorithm was used in the current study.
The elasticity parameter of DIRvol has been found to have considerable impact for VHU only
when using the −250 HU threshold, which is likely attributed to non-optimal registration of
high-contrast structures.
In principle, only lung parenchyma volumes should be included in the analysis of
ventilation imaging. The lung parenchyma volumes were segmented by thresholding voxels at
either −600 or −250 HU within the lung outlines generated by the model-based segmentation
in this study. High-contrast structures (e.g. small vessels) would still remain in the segmented
volumes mostly because of partial volume effects. Such structures cause erroneous VHU values.
Visually, the −600 HU threshold resulted in less high-contrast structures (e.g. small vessels)
in the segmented volumes than the −250 HU threshold. Several discrepancies in the results
between −600 and −250 HU thresholds were most likely due to high-contrast structures as
described above in section 3.4. It is difficult to segment lung parenchyma accurately, given that
CT lung density is influenced by factors including subject tissue volume, air volume, physical
material properties of the lung parenchyma, transpulmonary pressure, and image acquisition
protocol. In particular, optimal thresholds vary by subject as demonstrated by Hu et al (2001).
Improvements in lung parenchyma segmentation and also HRCT imaging are likely to make
the analyses more accurate for future studies.
There are several limitations in 4D-CT ventilation imaging, including the artifacts in
vol
. Artifacts are observed in the 4D-CT images at an
4D-CT images and a potential bias in VHU
alarmingly high frequency (i.e. 90%) (Yamamoto et al 2008). Several artifacts were observed
in the images used in this study. As artifacts of non-lung structures (e.g. diaphragm) have
high-contrast, they are masked out from the lung parenchyma volumes after thresholding.
There should be artifacts in the lung volumes as well; however, it is difficult to segment
them because of low-contrast. Future studies will investigate the possible impact of 4D-CT
artifacts on regional ventilation computation, possibly by focusing on the CT data segments
where artifacts of non-lung structures occur. Given that DIRvol tries to minimize the sum
vol
of squared difference (see section 2.2) which is essentially the same metric as VHU , VHU
may underestimate the actual HU change, which is considered as a limitation. However, this
bias would be small because DIRvol is based on not only the sum of squared difference, but
also a regularizing term (i.e. elastic regularizer). Other similarity functions including mutual
information may avoid this issue. Furthermore, we used the 4D-CT images at the peak-exhale
and peak-inhale phases only in this study. Guerrero et al (2006) and Christensen et al (2007)
calculated ventilation images with different pairs of 4D-CT images, i.e. each phase paired
with the peak-exhale phase (Guerrero et al 2006) or adjacent phase (Christensen et al 2007).
Both studies demonstrated spatially varying distributions of regional ventilation, indicating
that regional ventilation is not necessarily constant during a breathing period and may vary
with phase. Considering the lung pressure–volume hysteresis, integrating the ventilation
calculated from all 4D-CT images over a full respiratory cycle would more appropriately
represent breathing process and may give more physiologically relevant ventilation images for
a future study.
In this study, the density masking technique with the threshold value of −910 HU was
used to detect emphysema in 4D-CT images acquired with the following scanning parameters:
120 kV, ∼100 mAs and 2.5 mm slice thickness as used clinically in our radiation oncology
2294
T Yamamoto et al
department. The justification for the use of these parameters was that, if ventilation on these
image sets proved physiologically accurate, the current practice would not need to be changed
to use the image sets for functional avoidance in radiotherapy treatment planning. However,
the density masking technique was investigated mostly based on HRCT images acquired
by a helical scan during a single breath-hold at peak exhale or peak inhale. Large clinical
COPD studies such as the evaluation of copd longitudinally to identify predictive surrogate
endpoints (ECLIPSE) study (Vestbo et al 2008) and feasibility of retinoids for the treatment of
emphysema (FORTE) study (Roth et al 2006) were also based on breath-hold HRCT images.
For example, HRCT scans were acquired with 120 kV, 40 mA and 1.00 or 1.25 mm slice
thickness during a breath-hold at full inhalation in the ECLIPSE study (Vestbo et al 2008).
There were several major differences in the CT acquisition methods between this study and
other HRCT studies, including the slice thickness and breathing during a scan. The slice
thickness used in this study (2.5 mm) was significantly larger than the slice thickness used in
large clinical COPD studies or recent HRCT studies (!1 mm) (Bankier et al 1999, Baldi et al
2001, Zaporozhan et al 2005, Camiciottoli et al 2006). Madani et al (2007) demonstrated that
density masking was significantly influenced by slice thickness. The criterion for emphysema
on the HU scale depends on the slice thickness (Friedman 2008, Lynch and Newell 2009).
Until recently, −910 HU used in this study was the most frequently used threshold value
for density masking (Muller et al 1988, Friedman 2008). More recently, lower threshold
values (e.g. −950 HU) have been used for thinner slice thicknesses !1 mm (Bankier et al
1999, Baldi et al 2001, Zaporozhan et al 2005, Camiciottoli et al 2006). Moreover, as yet
there is no clear consensus on the threshold value. We considered that using −910 HU for
2.5 mm slice thickness was reasonable for density masking. For breathing, 4D-CT scans
were acquired during free breathing without any patient respiratory control (breathing training
or breath-hold), which were subsequently sorted by correlating with the respiratory signal
into 10 respiratory phase-based bins, while HRCT scans are normally acquired during a
breath-hold. Only the peak-exhale CT images (typically 50%) were used for emphysema
quantification. We assumed that the respiratory status at peak exhale during a 4D-CT scan
was comparable to the respiratory status during a breath-hold HRCT scan, and hence there
was no significant impact on emphysema quantification.
In this study, we used emphysematous lung regions as surrogates for low ventilation
as suggested by Ley-Zaporozhan et al (2007). There is an uncertainty in the relationship
between ventilation and emphysema as reflected by several conflicting results in the literature.
There is destruction of alveolar septa in emphysema, leading to decreased elastic recoil of the
alveoli and less radial traction, and hence low ventilation (Zaporozhan et al 2004, Spector
et al 2005). Zaporozhan et al (2004) showed that emphysema was strongly associated with
a reduction of ventilation based on hyperpolarized 3He MR imaging. Morrell et al (1994)
showed increased collateral ventilation (i.e. ventilation of alveolar structures through passages
or channels that bypass the normal airways) by a factor of 10 in emphysema compared
to normal volunteers. However, ventilation was still four times less than that in lungs
showing normal ventilation. Furthermore, Spector et al (2005) demonstrated remarkably
lower ventilation in emphysematous rats than in healthy rats with clearly separated peaks of
the probability density functions of ventilation using hyperpolarized 3He MR imaging. While
they compared ventilation in the whole lungs of emphysematous and healthy rats rather than in
segmented emphysematous and non-emphysematous lung regions, their results were similar
to our results. They also observed large overlaps between the probability density functions
for emphysematous and healthy rats, similarly to our VHU results. In contrast, Johansson et al
(2004) found poor relationships between ventilation and emphysema in 6 of 20 patients, while
13 patients showed significant relationships. They compared the 2D ventilation scintigrams
4D-CT lung ventilation imaging in emphysema patients
2295
and 3D-CT scans (used for emphysema quantification), which could have led to inaccurate
alignment between the two images and could explain the poor relationships as discussed
by the authors. In addition to the uncertainty in the relationship between ventilation and
emphysema, we cannot rule out other possible causes of low ventilation in the study patients,
including airway narrowing (Nakano et al 2000), small-airway diseases (Gelb et al 1998),
chronic persistent asthma (Gelb and Zamel 2000) or airway obstruction due to a tumor.
The density masking-defined non-emphysematous lung regions might have been affected by
these causes and might have low ventilation. However, these possible effects may be small,
considering a larger volume of non-emphysematous lung (i.e. emphysematous regions: the
mean%LAA = 21.9 ± 15.2%). Emphysematous lung regions are still expected to have
lower ventilation on average compared with non-emphysematous regions. It is difficult to
evaluate the small airway narrowing in the 4D-CT images (2.5 mm thickness), given that
Lynch and Newell (2009) recommended a submillimeter slice thickness for evaluation of
airway narrowing. The hypothesis tested in this study was a necessary condition, but not
a sufficient condition, for physiologically accurate 4D-CT ventilation imaging. A further
study is necessary to investigate the physiologic accuracy. Our future studies will focus
on the comparison with the SPECT ventilation (assumed ground truth) and optimization of
the algorithm parameters to determine an appropriate class of DIR algorithm with optimal
parameters as well as appropriate ventilation metric. This comparison also has limitations
including central airway depositions of aerosol particles (Magnant et al 2006, Castillo et al
2010, Yamamoto et al 2010), temporal differences between 4D-CT and SPECT scans which
could lead to image registration uncertainties (i.e. misalignment) and temporal changes in
ventilation itself. This study was based on the same data set acquired at the same time,
and hence has advantages over the SPECT comparisons, i.e. no registration uncertainty and
temporal change.
5. Conclusion
This study has demonstrated the correlation between the 4D-CT ventilation and emphysema
for 12 patients. The correlation was sensitive to the ventilation metric, and the HU metric was
found to have a higher correlation than the Jacobian metric. Significantly lower ventilation in
emphysematous lung regions than in non-emphysematous regions indicates the potential for
HU-based 4D-CT ventilation imaging to achieve high physiologic accuracy. A further study
is needed to confirm these results.
Acknowledgments
Drs Ann Weinacker, Glenn Rosen and Ann Leung at Stanford provided advice and information
during the design of this study. Julie Baz from the University of Sydney carefully reviewed
and improved the clarity of this manuscript.
References
Albert M S, Cates G D, Driehuys B, Happer W, Saam B, Springer C S Jr and Wishnia A 1994 Biological magnetic
resonance imaging using laser-polarized 129Xe Nature 370 199–201
Alderson P O and Line B R 1980 Scintigraphic evaluation of regional pulmonary ventilation Semin. Nucl.
Med. 10 218–42
Arakawa A, Yamashita Y, Nakayama Y, Kadota M, Korogi H, Kawano O, Matsumoto M and Takahashi M 2001
Assessment of lung volumes in pulmonary emphysema using multidetector helical CT: comparison with
pulmonary function tests Comput. Med. Imaging Graph. 25 399–404
2296
T Yamamoto et al
Baldi S, Miniati M, Bellina C R, Battolla L, Catapano G, Begliomini E, Giustini D and Giuntini C 2001 Relationship
between extent of pulmonary emphysema by high-resolution computed tomography and lung elastic recoil in
patients with chronic obstructive pulmonary disease Am. J. Respir. Crit. Care Med. 164 585–9
Bankier A A, De Maertelaer V, Keyzer C and Gevenois P A 1999 Pulmonary emphysema: subjective visual grading
versus objective quantification with macroscopic morphometry and thin-section CT densitometry Radiology
211 851–8
Brock K K 2009 Results of a multi-institution deformable registration accuracy study (MIDRAS) Int. J. Radiat.
Oncol. Biol. Phys. 76 583–96
Camiciottoli G, Bartolucci M, Maluccio N M, Moroni C, Mascalchi M, Giuntini C and Pistolesi M 2006
Spirometrically gated high-resolution CT findings in COPD: lung attenuation vs lung function and dyspnea
severity Chest 129 558–64
Castillo R, Castillo E, Guerra R, Johnson V E, McPhail T, Garg A K and Guerrero T 2009 A framework for evaluation
of deformable image registration spatial accuracy using large landmark point sets Phys. Med. Biol. 54 1849–70
Castillo R, Castillo E, Martinez J and Guerrero T 2010 Ventilation from four-dimensional computed tomography:
density versus Jacobian methods Phys. Med. Biol. 55 4661–85
Christensen G E, Song J H, Lu W, El Naqa I and Low D A 2007 Tracking lung tissue motion and
expansion/compression with inverse consistent image registration and spirometry Med. Phys. 34 2155–63
Christian J A, Partridge M, Nioutsikou E, Cook G, McNair H A, Cronin B, Courbon F, Bedford J L and Brada M
2005 The incorporation of SPECT functional lung imaging into inverse radiotherapy planning for non-small
cell lung cancer Radiother. Oncol. 77 271–7
de Jong P A, Lindblad A, Rubin L, Hop W C, de Jongste J C, Brink M and Tiddens H A 2006 Progression of
lung disease on computed tomography and pulmonary function tests in children and adults with cystic fibrosis
Thorax 61 80–5
de Lange E E, Altes T A, Patrie J T, Gaare J D, Knake J J, Mugler J P 3rd and Platts-Mills T A 2006 Evaluation of
asthma with hyperpolarized helium-3 MRI: correlation with clinical severity and spirometry Chest 130 1055–62
Ding K, Cao K, Amelon R E, Christensen G E, Raghavan M L and Reinhardt J M 2010 Comparison of intensity- and
Jacobian-based estimates of lung regional ventilation Proc. Third Int. Workshop on Pulmonary Image Analysis,
MICCAI 2010 pp 49–60
Friedman P J 2008 Imaging studies in emphysema Proc. Am. Thorac. Soc. 5 494–500
Fuld M K, Easley R B, Saba O I, Chon D, Reinhardt J M, Hoffman E A and Simon B A 2008 CT-measured regional
specific volume change reflects regional ventilation in supine sheep J. Appl. Physiol. 104 1177–84
Gelb A F and Zamel N 2000 Unsuspected pseudophysiologic emphysema in chronic persistent asthma Am. J. Respir.
Crit. Care Med. 162 1778–82
Gelb A F, Zamel N, Hogg J C, Muller N L and Schein M J 1998 Pseudophysiologic emphysema resulting from severe
small-airways disease Am. J. Respir. Crit. Care Med. 158 815–9
Gevenois P A, De Vuyst P, de Maertelaer V, Zanen J, Jacobovitz D, Cosio M G and Yernault J C 1996 Comparison
of computed density and microscopic morphometry in pulmonary emphysema Am. J. Respir. Crit. Care Med.
154 187–92
Gould G A, Redpath A T, Ryan M, Warren P M, Best J J, Flenley D C and MacNee W 1991 Lung CT density
correlates with measurements of airflow limitation and the diffusing capacity Eur. Respir. J. 4 141–6
Guerrero T, Castillo R, Noyola-Martinez J, Torres M, Zhou X, Guerra R, Cody D, Komaki R and Travis E 2007
Reduction of pulmonary compliance found with high-resolution computed tomography in irradiated mice Int.
J. Radiat. Oncol. Biol. Phys. 67 879–87
Guerrero T, Sanders K, Castillo E, Zhang Y, Bidaut L, Pan T and Komaki R 2006 Dynamic ventilation imaging from
four-dimensional computed tomography Phys. Med. Biol. 51 777–91
Guerrero T, Sanders K, Noyola-Martinez J, Castillo E, Zhang Y, Tapia R, Guerra R, Borghero Y and Komaki R 2005
Quantification of regional ventilation from treatment planning CT Int. J. Radiat. Oncol. Biol. Phys. 62 630–4
Gur D, Drayer B P, Borovetz H S, Griffith B P, Hardesty R L and Wolfson S K 1979 Dynamic computed tomography
of the lung: regional ventilation measurements J. Comput. Assist. Tomogr. 3 749–53
Gur D, Shabason L, Borovetz H S, Herbert D L, Reece G J, Kennedy W H and Serago C 1981 Regional pulmonary
ventilation measurements by xenon enhanced dynamic computed tomography: an update J. Comput. Assist.
Tomogr. 5 678–83
Haraguchi M, Shimura S, Hida W and Shirato K 1998 Pulmonary function and regional distribution of emphysema
as determined by high-resolution computed tomography Respiration 65 125–9
Harris R S and Schuster D P 2007 Visualizing lung function with positron emission tomography J. Appl.
Physiol. 102 448–58
Hayhurst M D, MacNee W, Flenley D C, Wright D, McLean A, Lamb D, Wightman A J and Best J 1984 Diagnosis
of pulmonary emphysema by computerised tomography Lancet 2 320–2
4D-CT lung ventilation imaging in emphysema patients
2297
Hoffman E A and Ritman E L 1985 Effect of body orientation on regional lung expansion in dog and sloth J. Appl.
Physiol. 59 481–91
Hu S, Hoffman E A and Reinhardt J M 2001 Automatic lung segmentation for accurate quantitation of volumetric
x-ray CT images IEEE Trans. Med. Imaging 20 490–8
Johansson A, Vikgren J, Moonen M, Tylen U and Bake B 2004 Regional ventilation and distribution of emphysema—a
quantitative comparison Clin. Physiol. Funct. Imaging 24 58–64
Kabus S, Klinder T, Murphy K, van Ginneken B, Lorenz C and Pluim J P W 2009 Evaluation of 4D-CT lung
registration Proc. MICCAI 2009 ed G Z Yang, D J Hawkes, D Rueckert, J A Noble and C J Taylor (London:
Springer) pp 747–54
Kabus S and Lorenz C 2010 Fast elastic image registration Proc. Medical Image Analysis For The Clinic—A Grand
Challenge, MICCAI 2010 pp 81–9
Kabus S, von Berg J, Yamamoto T, Opfer R and Keall P J 2008 Lung ventilation estimation based on 4D-CT imaging
Proc. First International Workshop on Pulmonary Image Analysis, MICCAI 2008 pp 73–81
Kashani R et al 2008 Objective assessment of deformable image registration in radiotherapy: a multi-institution study
Med. Phys. 35 5944–53
Kauczor H, Surkau R and Roberts T 1998 MRI using hyperpolarized noble gases Eur. Radiol. 8 820–7
Kauczor H U, Hast J, Heussel C P, Schlegel J, Mildenberger P and Thelen M 2000 Focal airtrapping at expiratory
high-resolution CT: comparison with pulmonary function tests Eur. Radiol. 10 1539–46
Kauczor H U, Hast J, Heussel C P, Schlegel J, Mildenberger P and Thelen M 2002 CT attenuation of paired HRCT
scans obtained at full inspiratory/expiratory position: comparison with pulmonary function tests Eur. Radiol.
12 2757–63
Kinsella M, Muller N L, Abboud R T, Morrison N J and DyBuncio A 1990 Quantitation of emphysema by computed
tomography using a “density mask” program and correlation with pulmonary function tests Chest 97 315–21
Levitzky M G 2007 Pulmonary Physiology (New York: McGraw-Hill)
Ley-Zaporozhan J, Ley S, Eberhardt R, Weinheimer O, Fink C, Puderbach M, Eichinger M, Herth F and Kauczor H U
2007 Assessment of the relationship between lung parenchymal destruction and impaired pulmonary perfusion
on a lobar level in patients with emphysema Eur. J. Radiol. 63 76–83
Lynch D A and Newell J D 2009 Quantitative imaging of COPD J. Thorac. Imaging 24 189–94
Madani A, De Maertelaer V, Zanen J and Gevenois P A 2007 Pulmonary emphysema: radiation dose and section
thickness at multidetector CT quantification—comparison with macroscopic and microscopic morphometry
Radiology 243 250–7
Magnant J, Vecellio L, de Monte M, Grimbert D, Valat C, Boissinot E, Guilloteau D, Lemarie E and Diot P
2006 Comparative analysis of different scintigraphic approaches to assess pulmonary ventilation J. Aerosol
Med. 19 148–59
Marcucci C, Nyhan D and Simon B A 2001 Distribution of pulmonary ventilation using Xe-enhanced computed
tomography in prone and supine dogs J. Appl. Physiol. 90 421–30
Marks L B, Spencer D P, Sherouse G W, Bentel G, Clough R, Vann K, Jaszczak R, Coleman R E and
Prosnitz L R 1995 The role of three dimensional functional lung imaging in radiation treatment planning:
the functional dose–volume histogram Int. J. Radiat. Oncol. Biol. Phys. 33 65–75
McGuire S M, Zhou S, Marks L B, Dewhirst M, Yin F F and Das S K 2006 A methodology for using SPECT
to reduce intensity-modulated radiation therapy (IMRT) dose to functioning lung Int. J. Radiat. Oncol. Biol.
Phys. 66 1543–52
McMahon C J, Dodd J D, Hill C, Woodhouse N, Wild J M, Fichele S, Gallagher C G, Skehan S J, van Beek E J
and Masterson J B 2006 Hyperpolarized 3helium magnetic resonance ventilation imaging of the lung in cystic
fibrosis: comparison with high resolution CT and spirometry Eur. Radiol. 16 2483–90
Morrell N W, Wignall B K, Biggs T and Seed W A 1994 Collateral ventilation and gas exchange in emphysema Am.
J. Respir. Crit. Care Med. 150 635–41
Muller N L, Staples C A, Miller R R and Abboud R T 1988 ‘Density mask’. An objective method to quantitate
emphysema using computed tomography Chest 94 782–7
Nakano Y et al 2000 Computed tomographic measurements of airway dimensions and emphysema in smokers.
Correlation with lung function Am. J. Respir. Crit. Care Med. 162 1102–8
Newell J D Jr 2008 Quantitative computed tomography of lung parenchyma in chronic obstructive pulmonary disease:
an overview Proc. Am. Thorac. Soc. 5 915–8
Park K J, Bergin C J and Clausen J L 1999 Quantitation of emphysema with three-dimensional CT densitometry:
comparison with two-dimensional analysis, visual emphysema scores, and pulmonary function test results
Radiology 211 541–7
Parr D G, Sevenoaks M, Deng C, Stoel B C and Stockley R A 2008 Detection of emphysema progression in alpha
1-antitrypsin deficiency using CT densitometry; methodological advances Respir. Res. 9 21
2298
T Yamamoto et al
Parr D G, Stoel B C, Stolk J and Stockley R A 2006 Validation of computed tomographic lung densitometry for
monitoring emphysema in alpha1-antitrypsin deficiency Thorax 61 485–90
Reinhardt J M, Ding K, Cao K, Christensen G E, Hoffman E A and Bodas S V 2008 Registration-based estimates of
local lung tissue expansion compared to xenon CT measures of specific ventilation Med. Image Anal. 12 752–63
Rietzel E, Pan T and Chen G T 2005 Four-dimensional computed tomography: image formation and clinical protocol
Med. Phys. 32 874–89
Roth M D et al 2006 Feasibility of retinoids for the treatment of emphysema study Chest 130 1334–45
Seppenwoolde Y et al 2002 Optimizing radiation treatment plans for lung cancer using lung perfusion information
Radiother. Oncol. 63 165–77
Shikata H, Hoffman E A and Sonka M 2004 Automated segmentation of pulmonary vascular tree from 3D CT images
Proc. SPIE 5369 107–16
Shioyama Y et al 2007 Preserving functional lung using perfusion imaging and intensity-modulated radiation therapy
for advanced-stage non-small cell lung cancer Int. J. Radiat. Oncol. Biol. Phys. 68 1349–58
Simon B A 2000 Non-invasive imaging of regional lung function using x-ray computed tomography J. Clin. Monit.
Comput. 16 433–42
Spector Z Z et al 2005 Quantitative assessment of emphysema using hyperpolarized 3He magnetic resonance imaging
Magn. Reson. Med. 53 1341–6
Spiropoulos K, Trakada G, Kalamboka D, Kalogeropoulou C, Petsas T, Efremidis G, Tsiamita M, Trakada A
and Dimopoulos I 2003 Can high resolution computed tomography predict lung function in patients with
chronic obstructive pulmonary disease? Lung 181 169–81
Suga K 2002 Technical and analytical advances in pulmonary ventilation SPECT with xenon-133 gas and Tc-99mTechnegas Ann. Nucl. Med. 16 303–10
Swift A J, Wild J M, Fichele S, Woodhouse N, Fleming S, Waterhouse J, Lawson R A, Paley M N and
Van Beek E J 2005 Emphysematous changes and normal variation in smokers and COPD patients using
diffusion 3He MRI Eur. J. Radiol. 54 352–8
Tajik J K, Chon D, Won C, Tran B Q and Hoffman E A 2002 Subsecond multisection CT of regional pulmonary
ventilation Acad. Radiol. 9 130–46
Vestbo J et al 2008 Evaluation of COPD longitudinally to identify predictive surrogate end-points (ECLIPSE) Eur.
Respir. J. 31 869–73
Vik T, Kabus S, von Berg J, Ens K, Dries S, Klinder T and Lorenz C 2008 Validation and comparison of registration
methods for free-breathing 4D lung CT Proc. SPIE 69142 1–10
von Berg J, Barschdorf H, Blaffert T, Kabus S and Lorenz C 2007 Surface based cardiac and respiratory motion
extraction for pulmonary structures from multi-phase CT Proc. SPIE 6511 65110Y
Yamamoto T, Kabus S, von Berg J, Lorenz C, Goris M L, Loo B W Jr and Keall P J 2010 Evaluation of four-dimensional
(4D) computed tomography (CT) pulmonary ventilation imaging by comparison with single photon emission
computed tomography (SPECT) scans for a lung cancer patient Proc. 3rd Int. Workshop on Pulmonary Image
Analysis, MICCAI 2010 pp 117–28
Yamamoto T, Kabus S, von Berg J, Lorenz C and Keall P J 2011 Impact of four-dimensional computed tomography
pulmonary ventilation imaging-based functional avoidance for lung cancer radiotherapy Int. J. Radiat. Oncol.
Biol. Phys. 79 279–88
Yamamoto T, Langner U, Loo B W Jr, Shen J and Keall P J 2008 Retrospective analysis of artifacts in four-dimensional
CT images of 50 abdominal and thoracic radiotherapy patients Int. J. Radiat. Oncol. Biol. Phys. 72 1250–8
Yaremko B P, Guerrero T M, Noyola-Martinez J, Guerra R, Lege D G, Nguyen L T, Balter P A, Cox J D and Komaki
R 2007 Reduction of normal lung irradiation in locally advanced non-small-cell lung cancer patients, using
ventilation images for functional avoidance Int. J. Radiat. Oncol. Biol. Phys. 68 562–71
Zaporozhan J, Ley S, Eberhardt R, Weinheimer O, Iliyushenko S, Herth F and Kauczor H U 2005 Paired
inspiratory/expiratory volumetric thin-slice CT scan for emphysema analysis: comparison of different
quantitative evaluations and pulmonary function test Chest 128 3212–20
Zaporozhan J, Ley S, Gast K K, Schmiedeskamp J, Biedermann A, Eberle B and Kauczor H U 2004 Functional analysis
in single-lung transplant recipients: a comparative study of high-resolution CT, 3He-MRI, and pulmonary
function tests Chest 125 173–81