S437
ESTRO 36
_______________________________________________________________________________________________
Conclusion
In this study, we measured spectra for external neutrons
and characterized neutron dose equivalents for a single
gantry proton system, whose use in the United States and
worldwide is increasing.
Poster: Physics track: Treatment plan optimisation:
algorithms
PO-0816 LRPM for fast automated high quality
treatment planning – towards a novel workflow for
clinicians
R. Van Haveren
1
, B.J.M. Heijmen
1
, W. Ogryczak
2
, S.
Breedveld
1
1
Erasmus Medical Center Rotterdam Daniel den Hoed
Cancer Center, Radiation Oncology, Rotterdam, The
Netherlands
2
Warsaw University of Technology, Control and
Computation Engineering, Warsaw, Poland
Purpose or Objective
The aim is to create a novel efficient workflow for
clinicians, where high quality treatment plans are ready
to be inspected minutes after the delineation is finished.
In the current clinical workflow, plans are automatically
generated using the in-house developed Erasmus-iCycle
optimiser, but planning times can be in the order of hours.
Therefore, we propose an extension of Erasmus-iCycle to
substantially reduce computation times, but maintain plan
quality.
Material and Methods
We developed the Lexicographic Reference Point Method
(LRPM), a fast algorithm to automatically generate multi-
criterial treatment plans in a single optimisation run. In
contrast, the currently implemented sequential method in
Erasmus-iCycle requires multiple optimisations to
generate a plan. We validate the LRPM by comparing
automatically generated VMAT plans (mimicked by 23
static beams) with the LRPM and the sequential method
for 30 prostate cancer patients and 15 head-and-neck
cancer patients. For these treatment sites (and others),
Erasmus-iCycle is in clinical use.
Results
For the 30 prostate cancer patients, plan differences
between the LRPM and the sequential method were found
neither clinically nor statistically significant. The LRPM
reduced the average planning time from 12.4 to 1.2
minutes, a speed-up factor of 10. For head-and-neck, the
LRPM reduced the planning times from 99.7 to 4.6
minutes, a speed-up factor of 22. Plan quality was mostly
similar on average. For individual cases however, the
LRPM plans showed clinically more balanced trade-offs
between OAR doses. In comparison with the plan resulting
from the sequential method, relatively large dose
reductions were possible for some OAR(s) at the cost of
relatively small increases of dose for other OAR(s).
Conclusion
The LRPM features very fast automatic multi-criterial
generation of high-quality treatment plans, reducing
Erasmus-iCycle planning times to the order of minutes.
Further research focuses on clinical implementation.
PO-0817 Anatomical robust optimization to deal with
variation in nasal cavity filling during IMPT
S. Van de Water
1
, F. Albertini
2
, D.C. Weber
2
, B.J.M.
Heijmen
1
, M.S. Hoogeman
1
, A.J. Lomax
2
1
Erasmus MC Cancer Institute, Radiation Oncology,
Rotterdam, The Netherlands
2
Paul Scherrer Institute, Center for Proton Therapy,
Villigen PSI, Switzerland
Purpose or Objective
Intensity-modulated proton therapy (IMPT) for tumors in
the sinonasal and skull-base regions can be seriously
affected by interfraction changes in nasal cavity filling,
resulting in underdosage of the tumor and/or overdosage
of organs-at-risk (OARs). The aim of this study was to
develop an anatomical robust optimization method that
accounts for variation in nasal cavity filling and to
compare it with the conventional single-field uniform dose
(SFUD) approach and with online plan adaptation.
Material and Methods
We included CT data of five patients with tumors in the
sinonasal region, for which the clinical target volume
(CTV) showed large overlap with the nasal cavity. Based
on the planning CT, we generated for each patient 25
‘artificial’ CTs with varying nasal cavity filling (Figure 1).
The minimax robust optimization method available in our
in-house developed treatment planning system was
extended to account for anatomical uncertainties by
including additional (artificial) CTs with varying patient
anatomy as error scenarios in the inverse optimization. For
each patient, we generated treatment plans using: 1) the
SFUD approach (with varying planning target volume (PTV)
margins of 0 mm, 3 mm or 5 mm), 2) anatomical robust
optimization (including two, three or four artificial CTs,
next to the planning CT), and 3) online plan adaptation
(generating a new treatment plan for each artificial CT).
We used the clinically applied 3- or 4-beam arrangements.
Treatment plans were evaluated by recalculating and
accumulating the dose for an entire fractionated 50-Gy
treatment, assuming each artificial CT to correspond to a
2-Gy
fraction. We assessed CTV and OAR dose parameters
for the accumulated dose and individual fractions. A
treatment planning strategy was considered adequate
when V
95%
≥99% and V
107%
≤2% in each fraction.