S80
ESTRO 36
_______________________________________________________________________________________________
propagate urethra delineation to the test patients. The n-
most similar individuals were selected and final
segmentation was obtained by a weighted vote. Leave-one
out cross validation of the atlas for urethra segmentation
was first performed on the training data set. Mean
Centerline Dispersion (MCD) and Hausdorff Distance (HD)
were used for accuracy assessment. The method was then
applied to a second set of 95 patients having received 78
Gy by IMRT for prostate cancer. Target volume and organs
at risks (bladder, prostate) were delineated on computed
tomography (CT) slices according to the French GETUG
group recommendations. Then, the urethra was
segmented using the proposed approach and dose was
measured inside the resulting segmentation and compared
to the dose to the prostate.
Results
From the training data set, the number of most similar
atlases was optimized to 10 in the leave one out scheme.
Average MCD of 2.3 mm and HD of 3.5 mm were thereby
obtained. In the testing data base dose received by the
segmented urethra were significantly higher than the
whole prostate in a range of dose from 74 Gy to 79 Gy
(Wilcoxon test p<0.01).
Conclusion
An accurate atlas based segmentation method was
proposed allowing assessment of dose within prostatic
urethra. Dose in this organ was significantly higher than
the whole prostate, mainly in the highest dose range.
Results open the way to further NTCP studies relating
urinary toxicity such as obstructive symptoms to the
urethra dose.
OC-0158 a priori scatter correction of cone-beam CT
projections in photon vs. proton therapy gantries
A.G. Andersen
1
, Y. Park
2
, O. Casares-Magaz
1
, U. Elstrøm
1
,
J. Petersen
1
, B. Winey
2
, L. Dong
3
, L. Muren
1
1
Aarhus University Hospital, Department of Medical
Physics, Aarhus V, Denmark
2
Massachusetts General Hospital, Department of
Radiation Oncology, Boston- Massachusetts, USA
3
Scripps Proton Therapy Center, Department of Medical
Physics, San Diego- California, USA
Purpose or Objective
Cone-beam (CB) CT is becoming available on proton
therapy gantries, to allow image/dose-guidance and
adaptation for protons. To use these techniques clinically,
the challenges related to image quality and Hounsfield
Unit accuracy need to be solved. Algorithms for scatter
correction have been developed, and have been explored
for CBCT systems on photon therapy gantries but so far not
on gantries for proton therapy that have a different
geometry and did not use a bow-tie filter. The
performance of an
a priori
scatter correction algorithm
was in this study compared for the first time on CBCT
systems for photon vs. proton therapy gantries.
Material and Methods
The
a priori
scatter correction algorithm used a plan CT
(pCT) and raw CB projections. The projections were
acquired with On-Board Imagers of a Varian photon
therapy Clinac and of a Varian proton therapy ProBeam
system. The projections were initially corrected for beam
hardening followed by reconstruction using the RTK back
projection Feldkamp-Davis-Kress algorithm (rawCBCT).
Manual, rigid and deformable registrations were applied
using Plastimatch on the pCT to the rawCBCT. The
resulting images were forward projected onto the same
angles as the raw CB projections. The two projections sets
were then subtracted from each other, Gaussian and
median filtered, and then subtracted from the raw
projections and finally reconstructed to a scatter
corrected CBCT. To evaluate the algorithm, water
equivalent path length (WEPL) maps were calculated from
anterior to posterior on different reconstructions of the
data sets (CB projections and pCT). Initially we evaluated
CB projections of an Alderson phantom acquired on the
Clinac system before comparing CB projections of the
same CatPhan phantom acquired on both the Clinac and
the ProBeam systems.
Results
In the analysis of the Clinac projections of the Alderson
phantom, the scatter correction resulted in sub-mm mean
WEPL difference from the rigid registration of the pCT,
considerably smaller than what was achieved with the
regular Varian CBCT reconstruction algorithm (Figure 1).
The largest improvement was for the half-fan (below
neck) scans. With the Catphan phantom the rawCBCT was
very similar to the Varian reconstruction, due to a refitting
of beam hardening curve. When comparing
reconstructions of photon to proton gantry CB projections
(Figure 2) we found that the
a priori
scatter correction
improved the mean WEPL difference while preserving
image quality (the number of countable line pairs) for both
gantry types. The photon gantry showed less WEPL
difference, however used a higher pulse current per
acquisition ( 2.00 mAs), compared to the proton gantry
(1.4 mAs). The complete scatter correction is performed
within three minutes on a desktop with NVidia graphics.