Tracking Lung Motion with Inverse

1
Tracking Lung Motion with Inverse Consistent
Image Registration
Joo Hyun Song1, Gary E. Christensen1, Issam El
Naqa2, Wei Lu2, and Daniel A. Low2 1Department of
Electrical and Computer Engineering, The
University of Iowa, Iowa City, IA
522422Department of Radiation Oncology,
Mallinckrodt Institute of Radiology, Washington
University School of Medicine, St. Louis, MO 63110

• Abstract
• Breathing motion is one of the major limit-ing
  factors for reducing dose and irradiati-on of
  normal tissue for conventional conf-ormal
  radiotherapy. This poster describes a
  relationship between tracking lung moti-on using
  spirometry data and image regi-stration of
  consecutive CT image volumes collected from a
  multislice CT scanner ov-er multiple breathing
  cycles. The temporal CT sequences from 5
  individuals were a-nalyzed in this study. The
  couch was mo-ved from 11 to 14 different
  positions to im-age the entire lung. At each
  couch positi-on, 15 image volumes were collected
  ov-er 2-3 breathing cycles. Small deformatio-n
  inverse consistent linear elastic (SICLE) image
  registration was used to register consecutive
  image scans. The log-Jacobian of these
  incremental transform-ations was computed to give
  a map of the local expansion and contraction of
  the lu-ng during the breathing cycle. In four out
  of five patients, the correlation between the
  average log-Jacobian value and the differential
  spirometry data correlated well (R2 0.858 on
  average for the entire lun-g). The correlation
  for the fifth patient was not as good (R2 0.377
  on average for the entire lung) and can be
  explained by the small variation in tidal volume
  for this patient. The correlation of the average
  log-Jacobian value and the differential
  sp-irometry data near the diaphragm correla-ted
  well in all five patients (R2 0.943 on
  average).
• Introduction
• Breathing motion is a significant source of error
  in radiotherapy treatment planning for the thorax
  and upper abdomen. Incre-asing the beam aperture
  is the method u-sed to account for lung motion in
  convent-ional conformal radiotherapy which
  incre-ases the risk of delivering radiation dose
  to healthy tissue and critical organs. Sev-eral
  other methods have been proposed to account for
  breathing motion induced error such as
  respiratory gating, couching and breath-hold.
  Image registration meth-ods can be used as a
  method to track lun-g motion and apply this
  information for ra-diotherapy treatment. In our
  previous pap-er, we have showed preliminary
  results fr-om correlating image registration
  results to spirometry volume measurements. In the
  paper, we have discussed how there was a close
  relationship between the cha-nge in lung volume
  and the amount of co-ntraction/expansion measured
  by image registration of CT images acquired at
  infe-rior lung positions near the diaphragm. In
  this poster, the relationship between the i-mage
  registration results and the spirom-etry volume
  measurements was found for all locations of the
  lung. Also, a notewort-hy contraction and
  expansion pattern fou-nd in the superior
  positions of the lung will be discussed in this
  poster.
• Material and methods
• CT acquisition
• 16-slice CT scanner operated in twelve-slice mode
  (Sensation 16, Siemens Medical Systems, Iselin,
  NJ).
• 15 scans were acquired at each couch position.
  This process was repeated until the entire thorax
  was scanned (2326 couch Positions).
• The selection of 15 scans was based on the
  expectation of 34 s per breathing period, and
  the desire to acquire scans over 3 breathing
  cycles.

• Spirometry
• Simultaneous spirometry measurements (VMM-400,
  Interface Associates, Laguna Niguel, CA) were
  acquired using an independent workstation.
• Spirometer data acquisition software (Labview,
  National Instruments, Austin, TX) was written to
  store the time and relative tidal volume at 10 ms
  intervals.
• SICLE Image Registration
• The small deformation inverse consistent linear
  elastic (SICLE) image registration was used to
  nonrigidly register adjacent 12 slice temporal
  image volumes.
• Inverse consistent image registration jointly
  estimates the forward and reverse transformations
  between two images while minimizing the inverse
  consistency error between the forward and reverse
  transformations.
• For a particular registration algorithm,
  minimizing the inverse consistency error provides
  more accurate correspondence between two images
  compared to independently estimating the forward
  and reverse transformations.
• The SICLE image registration algorithm jointly
  estimates the forward and reverse transformations
  h and g, respectively, that minimize the cost
  function.

Results and discussion Twelve to fifteen 12 slice
CT scans were collected over an 11 second period
to capture approximately 3 breathing cycles at a
single couch position for the five patients in
this study. Figure 5 shows the spirometry data
collected for the 15 scans collected at couch
position 10 for patient 3 which was near the RMB.
This graph shows that the patients breathing
with respect to the tidal volume was not constant
over time for this data acquisition. The
variation included the time per breathing cycle
(5 scans were required for the first two cycles
and 3 for the last cycle), the incremental
difference in tidal volume between scans, and the
maximum/minimum tidal volume.
Figure 6 shows the log-Jacobian values
color-coded and superimposed on transverse slice
7 from couch position 10. The color-coded
log-Jacobian images are superimposed on the
transverse section of the second of the two image
volumes, respectively. The contraction and
expansion indicated by the log-Jacobian images
agree with the change in volume indicated in the
spirometry plot. In fact, the two measurements
showed a correlation of R20.934.
Figure 7 illustrates the
distribution pattern of regions of expansion and
contraction for 5 different patients in the same
region of the lung at the same breathing phase.
The red lines on the spirometry plots indicate
the intervals in which the scans were made. The
spirometry plots also indicate that the volumes
of air exhaled and inhaled are different for each
patient. From visual observations, it is
difficult to find close correlation between the
log-Jacobian images. However, what is interesting
is that the log-Jacobian images of all patients
in the exhalation phase appear to be inverses of
their inhalation images. In other words,
replacing red with magenta and vice versa of the
exhalation images will result in a series of
images that will resemble the inhalation images.
This observation is most obvious with patient 2
and 4 images. The plot in Figure 8 shows the
summary of the correlation of the difference in
tidal volume between two consecutive image scans
plotted against the average log-Jacobian value
across the lung segmentation in slice 7 of the
volume of all 5 patients. As it can be observed
from the plot, the correlation of patients 1-4 is
good in the superior portion of the lung.
However, the correlation worsens as it gets
closer to the middle section of the lung and then
the correlation improves again as it approaches
the inferior section of the lung. Patient 5 did
not seem to produce a good correlation for the
most part, except for the inferior portion of the
lung. It was discovered that patient 5s lungs
were severely diseased, and that partly explains
why the correlation of patient 5s data was poor
compared to the other four patients. Conclusion Th
is poster presented preliminary results that show
a good overall correlation (R0.848 for all 5
patients and R0.927 with patient 5 data
excluded) between expansion/contraction
determined from inverse consistent image
registration of consecutive multislice CT image
scans and spirometry data. This result suggests
that the SICLE image registration algorithm is
able to retro-actively track breathing motion of
the lung across consecutive CT image
acquisitions. Future work is needed to verify
these findings and to use image registration to
improve real-time conformal radiation
treatment. Acknowledgements This work supported
in part by NIH R01 96679, NIH R24 HL64368, and a
corporate grant from Computerized Medical Systems.
Figure 8 Summary of R-values of the volume
difference vs. log-Jacobian correlation of
patients 1-5. The starting couch positions were
aligned to illustrate the relationship between
the R-value and the relative couch position.
Figure 1 The inverse consistency error is
defined as the distance between the starting and
ending points produced by mapping a point through
the forward and then reverse transformations.
Figure 5 Spirometry volume vs. time for patient
3, couch position 10. The red crosses in this
plot correspond to the air volume of the lung for
the A through O CT volumes collected at couch
position 10.
Figure 6 Color coded log-Jacobian images
superimposed on target CT volume for patient 3,
couch position 10. Magenta corresponds to a
contraction, red corresponds to an expansion, and
green to no change. Panels show the Log-Jacobian
superimposed on the CT for slice 7 for each of
the 14 image registrations. The subscripts for
each panel denote the direction of the
transformation displayed.
Figure 2Illustration of how two sets of lung
images are registered against each other. h(x)
denotes the forward transformation while g(x)
denotes the reverse transformation.
Figure 3Illustration of multiple sets of lung CT
images (labeled A thru D) are registered against
the adjacent images.
Figure 7 The distribution pattern of regions of
expansion and contraction for 5 different
patients in the same region of the lung (near the
RMB) at the same breathing phase. The top row are
images taken during an exhalation phase while the
bottom row are images taken during the inhalation
phase. The red lines on the spirometry plots
indicate the intervals in which the scans were
made. It is interesting that the log-Jacobian
images of all patients in the exhalation phase
appear to be inverses of their inhalation images.
Figure 4 Absolute difference images between the
5 temporally adjacent image volumes before and
after nonrigid SICLE image registration for
patient 3, couch position 10. Panels show the
difference for slice 7. The subscripts for each
panel denotes the volumes used in the difference.