S157
ESTRO 36
_______________________________________________________________________________________________
weeks, keeping the same positioning. The intra-fraction
reproducibility of the lung anatomy during breath hold was
investigated, by comparing the MRI of the first breath hold
with the three other MRIs of the same session. The inter-
fraction anatomical reproducibility was investigated by
comparing the first breath hold MRI of the first session
with the four MRIs during the second session. To avoid any
influence of setup variation, first a global rigid image
registration was performed. Then the lung volume was
semi-automatically segmented to define a region of
interest for the deformable image registration (DIR). DIR
was performed using Mirada RTx v1.2 (Mirada Medical,
Ltd.), with a DIR grid resolution of 3.5x2x3 mm
3
. The
deformation vector fields were analyzed using MATLAB
v2014b. Magnitudes of the deformation vectors were
calculated and combined for all five volunteers. The lung
volumes were divided into six segments, to analyze the
anatomical displacements on a local level. A boxplot
showing the intra- and inter-fraction displacements with a
schematic view of the six segments can be seen in figure
1.
Results
The lung volumes for all breath holds varied by 2% within
and 7% between fractions. Looking at all five volunteers,
up to 2 mm median intra- and inter-fraction displacements
were found for all lung segments. The anatomical
reproducibility decreased towards the caudal regions.
Inter-fraction displacements were larger than intra-
fractional displacements. Maximum displacements (99.3%
of the magnitude vectors) reached 6 mm intra-fractionally
and did not exceed 8 mm inter-fractionally.
Conclusion
While the lung volume differences were insignificant,
relevant anatomical displacements were found. Moreover,
a trend of increased displacements over time could be
seen. ABC mitigates motion to some extent. Nevertheless,
the remaining reproducibility uncertainties need to be
considered during scanned proton therapy treatments. As
next step, we aim to include this knowledge in a model to
estimate their dosimetric influence for scanning proton
therapy.
OC-0304 Real-time gamma evaluations of motion
induced dose errors as QA of liver SBRT tumour
tracking
T. Ravkilde
1
, S. Skouboe
1
, R. Hansen
1
, E.S. Worm
1
, P.R.
Poulse n
1
1
Aarhus University Hospital, Department of Oncology,
Aarhus N, Denmark
Purpose or Objective
Organ motion during radiotherapy can lead to serious
deterioration of the intended dose distribution. As modern
radiotherapy shifts increasingly towards escalated doses,
steeper dose gradients and hypofractionation, the
demands on accurate delivery increase concurrently. A
large body of studies show that tumour tracking can be
applied to mitigate the effects of motion and restore dose
fidelity, yet clinical introduction seems reluctant. In this
study we report on a method for continuous evaluation of
the tracking dose delivery that conforms to common dose
analysis practice and can be acted upon in real time.
Material and Methods
Experiments were performed on a TrueBeam linear
accelerator (Varian Medical Systems) with target motion
being recorded by an electromagnetic transponder system
(Calypso, Varian Medical Systems). A HexaMotion motion
stage (Scandidos) reproduced the liver motion traces for
five different liver SBRT patients as previously measured
using intrafraction kV imaging. VMAT SBRT treatment
plans were delivered to the moving phantom with MLC
tracking, without tracking (simulating the actual delivery)
as well as to a static phantom for reference (planned
delivery). Temporally resolved dose distributions were
measured at 72 Hz using a Delta4 dosimeter (Scandidos).
Accelerator parameters (monitor units, gantry angle, MLC
leaf positions, etc.) were streamed at 21 Hz to prototype
software that performed continuous reconstruction of the
dose in real time by a simplified non-voxel based 4D pencil
beam convolution algorithm. Also in real time, but on a
separate thread, 3%/3mm gamma evaluations were
calculated continuously throughout beam delivery to
quantify the deviation from the planned intent. After
experiments, the time-resolved gamma tests were
compared with the same quantities from the measured
data.
Results
The motion induced gamma errors were well
reconstructed both spatially (Figure 1) and temporally
(Figure 2). In 95% of the time both actual and planned
doses were reconstructed within 100 ms. The median time
for reconstruction was 65 ms, which translates into a
typical frequency of about 15 Hz. Asynchronously, but also
continuously, 95% of gamma evaluations were performed
within 1.5 s with the median being at 1.2 s. Over all
experiments the root-mean-square difference between
reconstructed and measured gamma failure rates was
2.9%.