484 / 1096
S469
ESTRO 36
_______________________________________________________________________________________________
this study was to develop and validate a 4D-MRI guided
mid-position (midP) correction strategy on an MR-Linac.
Material and Methods
Experiments were performed on an MR-Linac (ATL1, Elekta
AB, Sweden), using the CIRS MRI-LINAC Dynamic Phantom
(CIRS Inc., USA). The moving cylinder was filled with
anisotropic MRI contrasts and a Perspex spherical target.
Motion was performed in CC direction using a Lujan 4
motion pattern with a 20mm amplitude and 4s period.
First, a T2-weighted MRI scan was acquired in midP. The
cylinder and target were segmented and the target was
expanded with a non-uniform margin (LR, AP:10mm;
CC:20mm). A density overwrite of 1 was assigned to the
structures and a treatment plan consisting of a single
anterior beam shaped around the PTV was created in
Monaco (Version 5.19.01 Research). Then, baseline shifts
in CC direction of 5, 10, 15 and 20mm were applied to the
phantom motion. For every shift, a retrospective self-
sorted 4D-MRI was acquired (axial single-shot TSE,
2x2x5mm
3
, TE/TR=60/400ms, 30dyn) and each phase was
registered to the midP reference image to calculate the
time average displacement. The plan was adapted
accordingly, performing a virtual couch shift (simple dose
shift) using aperture morphing in Monaco. All plans were
delivered while electronic portal imaging device (EPID)
cine images were acquired. The time average
displacement of the target was calculated from the EPID
images and geometric accuracy of the workflow was
quantified as the distance of the average position of the
target to the field edges in the EPID images.
Results
In Figure1, MRI and EPID images of the midP and a shifted
inhale and exhale position are shown. Table1 shows the
time average displacement of the target in the 4D-MRI and
the EPID images with respect to the reference as well as
the distance of the average target position to the field
edges.
The geometric accuracy of the 4D-MRI guided workflow
was 0.3±0.4mm in CC, which includes the 4D-MRI
registration accuracy.
Conclusion
4D-MRI guidance on an MR-Linac was shown to be feasible
and had sub-millimeter accuracy. Such a correction
strategy has great potential for moving targets that are
difficult to visualize on alternative image guidance
modalities.
Acknowledgements: This research was partly sponsored by
Elekta AB, Stockholm, Sweden. The authors would like to
thank CIRS Inc., Robert Spaninks (Elekta) and Jochem
Kaas, Natasja Janssen, Ben Floot and Marco van den Berg
(NKI).
PO-0862 Correlation of Liver and Pancreas Tumor
motion with Normal Anatomical Stru ctures
R. Kaderka
1
, A. Paravati
1
, R. Sar kar
1
, J. Tran
1
, K. Fero
1
,
N. Panjwani
1
, D. Simpson
1
, J. Murphy
1
, T. Atwood
1
1
University of California San Diego, Department of
Radiation Medicine and Applied Sciences, San Diego, USA
Purpose or Objective
Target motion caused by respiration remains the central
challenge to delivering SBRT in the abdomen. For targets
in the pancreas and liver, SBRT oftentimes necessitates
placement of metal fiducials to determine tumor position
with fluoroscopy, due to difficulty in visualizing tumors on
non-contrast imaging. Metal fiducials have limitations in
that they represent an invasive procedure which can
introduce treatment delays. Furthermore, fiducials can
migrate from their intended position, and the metal can
introduce imaging artifacts which make tumor delineation
a challenge. We hypothesized that upper abdominal tumor
motion would correlate with the motion of nearby organs
and could thereby serve as a fiducial-less proxy for tumor
motion.
Material and Methods
Fifteen patients (12 with pancreas and 3 with liver tumors)
underwent a 4-dimensional (4D) CT simulation prior to
treatment with SBRT. 4D CT images were divided into 10
phases and normal tissues were contoured on a single 4D-
CT phase and propagated to the other phase s using
deformable image registration. As a means of quality
control for image registration and contour propagation the
liver was manually contoured on all phases for 5 patients
by physicians and compared to the automated contour
propagation using a Dice coefficient. Motion was defined
from the center-of-mass of each structure, and a patient-
specific linear tumor position prediction model based on
liver position was developed.
Results
We found a strong overlap of manually entered contours
and the automatically segmented contours with a mean
Dice-coefficient of 0.95 (standard deviation 0.01). The
linear models accurately predicted tumor motion with a
mean absolute error of 0.5 mm and no error greater than
3.0 mm. Mean absolute and maximum errors by direction
and tumor type are listed in the table below.
Left-right
direction
Anterior-
posterior
direction
Superior-
inferior
direction
Pancreas tumors
mean absolute
error (mm)
0.3
0.4
0.5
Pancreas tumors
maximum error
(mm)
1.0
1.7
2.6
Liver tumors mean
absolute error
(mm)
0.3
0.4
0.8
Liver tumors
maximum error
(mm)
1.4
2.7
3.0
Conclusion
This study demonstrates that normal organ motion could
serve as a fiducial-less proxy for tumor motion with SBRT
in the upper abdomen when on-site real-time 4D
volumetric imaging becomes available during treatment.
Deformable image registration has been demonstrated to
be a reliable and fast tool for segmentation of normal
organs. Moving this motion management approach into
clinic requires additional research to optimize 4D image
quality and understand inter-fraction reproducibility.
PO-0863 Suggestion of optimal planning target volume
margins for stereotactic body radiotherapy of the spine