149 / 1082
S136
ESTRO 36 2017
_______________________________________________________________________________________________
consistent reporting of treatment planning for regional
nodal radiotherapy, particularly within clinical trials. It is
on this foundation that the ESTRO consensus guideline for
elective breast cancer radiotherapy has been developed.
We set out to evaluate variability of lymph node
contouring using the ESTRO consensus guideline across
multiple investigators and sites.
Material and Methods
As part of the UK FAST-Forward trial RTQA programme
each co-investigator is required to delineate a LN CTV
comprising of levels 1-4 as defined by the ESTRO guideline
on an outlining benchmark CT dataset. LN CTV’s were
defined by three clinicians (LN_CTV_CLIN_1/2/3) of the
Trial Management Group (TMG) with experience of the
ESTRO guidelines and a consensus LN CTV defined through
discussion and comparison.
LN CTV’s of 39 investigators from 32 radiotherapy centres
were analysed using in-house software based on the
Computational Environment for Radiotherapy Research
(CERR). Discordance Index (DI), Geographical Miss Index
(GMI), Jaccard Index (JI), and Mean Distance to
Conformity (MDC) indices were generated, comparing
LN_CTV_CLIN_1/2/3 and investigators LN CTV’s to the
consensus LN CTV in addition to standard volume
statistics.
Results
The interobserver variation (SD/Mean) in volume
contoured between the investigators was lower compared
to published literature(40.8%/55.9% axillary nodes and
60.5% SCF nodes- Li et al,2009).
The JI results indicate investigator volumes achieved
conformity in relation to the consensus comparable to
LN_CTV_CLIN_1 with the SD supporting low interobserver
variability across submissions. Larger mean and minimum
DI compared to GMI indicate a trend for over contouring
across investigator submissions, however associated range
and SD supports MDC analysis showing a larger degree of
variation was associated with under contouring. MDC
analysis on a slice by slice basis identifies defining the
caudal extent of level 1 as the region associated with the
largest degree of under contouring and the caudal aspect
of levels 3/4 with over contouring.
It is important to consider the variation in volume and
conformity in context of the ESTRO guidelines which
acknowledge that the anatomical boundaries are not
considered exact to mm, with a range in JI scores between
consensus clinicians despite the indices indicating good
spatial relationship to the consensus.
Conclusion
Contouring of the regional lymph nodes using the ESTRO
consensus reduces interobserver variability in volume
contoured. Comparing the consensus clinicians and
investigator results suggest that experience and/or
training is associated with less interobserver variability,
promoting the role of RTQA when adopting new outlining
guidelines as part of a multicentre trial.
OC-0266 Motion specific target delineation
significantly reduce treated volumes in liver SBRT
J. Wielaard
1,2
, C.H. Slump
2
, K. Muller
1
, A.W.H. Minken
1
,
H. Westendorp
1
1
Radiotherapiegroep Behandellocatie Deventer, Medical
Physics, Deventer, The Netherlands
2
University of Twente, MIRA: Institute for Biomedical
Technology and Technical Medicine, Enschede, The
Netherlands
Purpose or Objective
Target volumes based on patient’s individual respiratory
motion, so-called motion specific target volumes, can
possibly improve target coverage and reduce dose to
healthy tissue. We have derived motion specific target
volumes for stereotactic body radiation therapy (SBRT) of
liver tumors based on deformable image registration (DIR)
of 4DCT images and assessed the intrahepatic accuracy for
these algorithms.
Material and Methods
4DCT data sets of 15 patients in head-first supine position
were obtained with a Brilliance Big Bore 16-slice CT
scanner (Philips Healthcare, OH, US) and reconstructed
using phase binning. Patients were positioned using a
BlueBAG Vacuum Cushion (Elekta, Sweden) and
immobilized with a SBRT Body Pro-Lok system with
abdominal compression plate (CIVCO, IA, US). Consecutive
DIR of the 4DCT images was used to track patient’s
individual respiratory motion by mapping voxels over 10
respiratory phases. Two DIR algorithms were used: a
hybrid algorithm based on intensity and delineated
contours (ANACONDA), and a biomechanical algorithm
based on the finite element method (MORFEUS). The
hybrid algorithm was employed with and without
delineation of the liver contour. The motion specific
target volume was created by propagating a gross target
volume (GTV) contoured on a reference CT over the
obtained vector fields. The target volume encompasses
the GTV of 10 respiratory phases plus a 3 mm set-up
margin in order to account for variations in patient setup.
The motion specific target volumes were compared with
volumes generated using static margins (6x6x10 mm)
according to the clinical protocol. To evaluate the
intrahepatic performance, fiducial markers were used as
points-of-interest to calculate the residual error after
registration. The fiducial registration error (FRE, mean
absolute residual error) was chosen as the measure to
compare the algorithms.
Results
The motion specific target volumes resulted in the
following average target volumes: hybrid: 33 ± 36.1 cc,
hybrid with liver contour guidance: 37.0 ± 27.5 cc and
biomechanical: 39.7 ± 29.4 cc. The average volume using
static margins of 6x6x10 mm was 56.9 ± 38.8 cc. In 7 out
of 15 cases the static margins did not encompass the
motion specific situation, leaving target tissue uncovered.
The hybrid algorithm with and without additional liver
contouring resulted in a FRE of 0.8 ± 0.8 mm and of 0.8 ±
0.5 mm respectively. The biomechanical algorithm
showed a FRE of 1.9 ± 1.0 mm.
Conclusion
The motion specific target volumes show a volume
reduction compared to the target volumes generated using
static margins. Besides that, the motion specific target
volume extends the with static margins derived target
volume. Motion specific target volumes using deformable
image registration could decrease the dose to healthy
tissue and potentially improve target coverage. The hybrid
algorithm resulted in a lower FRE compared to the
biomechanical algorithm.