454 / 1082
S439
ESTRO 36 2017
_______________________________________________________________________________________________
here several sampling strategies. We will also show that
current robust optimizers sample scenarios in a
statistically inconsistent way.
Material and Methods
Sampling must optimize the trade-off between clinical
optimality and robustness. Both were assessed by
computing the volume of the BSPTV and a confidence
interval (CI), respectively. The latter is defined as the
percentage of all possible ranges and beam positions that
the BSPTV encompasses. The findings can then be applied
later to robust optimizers.
We have designed a simulation phantom to model
uncertainties in lung tumors (Figure 1). Standard
deviations of the Gaussian distributions for (systematic)
setup errors, baseline shifts, and CT conversion errors
were 5 mm, 5 mm, and 2%, respectively. The errors were
sampled following three different methods:
1.
M1 (conventional approach): sampling of setup
errors and baseline shifts within conventional
lateral PTV margin for systematic errors
(encompassing 90% of possible beam positions).
The distal and proximal margins encompass 98%
of possible proton ranges scaled by a flat CT
conversion error (±3.3% to include 90% of
possible CT conversion errors).
2.
M2: same as M1 with random sampling of the CT
conversion error.
3.
M3: all errors are simulated within an iso-
likelihood hypersurface including 90% of all
possible scenarios.
A fixed breathing-induced motion amplitude of 1 cm has
been considered for every scenario.
Results
BSPTVs equaled 430, 420 and 564% of the CTV volume for
the three methods, respectively (see figure 2 that
illustrates the range margins). M1 does not ensure
statistical consistency because of the flat CT conversion
error, which overemphasizes unlikely scenarios (large
geometrical AND large CT conversion errors) and makes
non-trivial the computation of the CI. M3 guarantees at
least 90% CI, but with a 34% increase of the irradiated
volume. The latter is due to the non-prioritization of
errors and to blindness relative to their potential
degeneracies. The CI for M2 equals 88%, but 90% CI can be
achieved for M2 by extending slightly the lateral PTV
margin to encompass 92% of possible beam positions and
98% of possible ranges, leading to a 425% volume, thus still
better than M3.
Conclusion
The best tradeoff between robustness and optimality was
achieved through random sampling of all errors limited by
the lateral conventional PTV margin and a large margin for
the possible proton ranges.
PO-0826 Evaluation of the new InCise MLC for
Cyberknife stereotactic radiotherapy
C. Limoges
1
, J. Bellec
1
, N. Delaby
1
, M. Perdrieux
1
, F.
Jouyaux
1
, E. Nouhaud
2
, I. Lecouillard
2
, E. Chajon
2
, R. De
Crevoisier
2,3,4
, E. Le Prisé
2
, C. Lafond
1,3,4
1
Centre Eugène Marquis, Medical Physics Department,
Rennes, France
2
Centre Eugène Marquis, Radiation Oncology
Department, Rennes, France
3
INSERM, U 1099, Rennes, France
4
University of Rennes1, LTSI, Rennes, France
Purpose or Objective
The aim of this study was to evaluate treatment planning
performances of the new InCise multileaf collimator (MLC)
with reference to the Iris variable circular aperture
collimator for intracranial and extracranial Cyberknife
stereotactic radiotherapy.
Material and Methods
The study was performed on a Cyberknife M6 v10.6
(Accuray). A total of 50 cases including 10 brain
metastases, 10 acoustic neuromas, 10 liver targets, 10
spinal metastases and 10 prostate cases were
investigated. For each case, two treatment plans were
generated with TPS Multiplan v5.3 (Accuray): one plan
using the InCise MLC v2 associated with the Finite Size
Pencil Beam (FSPB) dose calculation algorithm and one
plan using the Iris collimator associated with RayTracing
(RT) or MonteCarlo (MC) dose calculation algorithm. Dose
was prescribed near the 80 % isodose and normalized to
obtain the same PTV coverage at ± 0.5 % for both plans.