S500
ESTRO 36
_______________________________________________________________________________________________
were reconstructed to show the uptake of intravenous
contrast agent.
The sequences were tested on healthy volunteers and one
patient using a 3T MR system (Ingenia; Philips Healthcare,
The Netherlands) and reviewed by two MR-experts and one
radiologist. Pineapple juice was given orally to distend the
stomach and suppress signal from the stomach filling.
Gadolinium was used as intravenous contrast agent for the
patient only.
Results
Visual inspection showed that for TD and staging, T2w
exhale respiratory navigator triggered, rather than a
respiratory sensor, provides excellent contrast with
limited motion artifacts. For TP, mDixon with a large FoV,
a high signal to noise ratio (SNR) and HR in one BH is
feasible. For motion modeling, 4D T2w MRI resulted in a
good slice ordering, high SNR and HR. For MM, TSE Cine-
MRI gave a good SNR and HR without artifacts. For staging
and treatment response monitoring, FB DWI with an
increased number of averages gave the best result, only
limited motion and susceptibility artifacts were visible. FB
4D THRIVE DCE resulted in a good temporal resolution and
limited motion artifacts.
(Figure 1)
Conclusion
We developed a comprehensive imaging protocol for the
entire RT guidance treatment chain. The complex motion
artifacts were reduced by applying either navigator
triggering or BH techniques. The new gastric cancer
protocol looks therefore very promising and will be used
for MR-based delineation for RT.
[1] van de Lindt T, et al. ESTRO 35 2016 Abstract-book:PV-
0325; 171-172
PO-0905 (Semi-)Automatic contouring strategies for
rectal boost treatment on the MR-Linac
C.N. Nomden
1
, M.P.W. Intven
1
, A.N.T.J. Kotte
1
, I.H.
Kiekebosch
1
, S. Mook
1
, I.M. Jürgenliemk-Schulz
1
, G.G.
Sikkes
1
, L.T.C. Meijers
1
, E.N. De Groot
1
, G.H. Bol
1
, B. Van
Asselen
1
, L.G.W. Kerkmeijer
1
, B.W. Raaymakers
1
1
UMC Utrecht, Radiation Oncology, Utrecht, The
Netherlands
Purpose or Objective
The MR-Linac enables online treatment adaptations in
response to changes in anatomy. This stresses the need for
fast contouring strategies for target and OARs.
Unfortunately, manual delineation in an online workflow
is time consuming and therefore suboptimal. The purpose
of this study was to investigate whether automatic and
semi-automatic contouring strategies result in clinical
acceptable contours for an online workflow on the MR-
Linac.
Material and Methods
Fifteen patients with early staged rectal cancer were
scanned at an 1.5T MRI for five consecutive days. The scan
consisted of a T2 weighted MRI; voxelsize 0.63x0.63mm,
slice thickness 4 mm and a total number of 30 slices. For
each scan the following contours were delineated by an
experienced radiation oncologist (manual contours): GTV,
mesorectum, bladder, rectum, sphincter, gynecological
volume (in one contour: vagina, cervix and uterus), left
and right femur. The manual contours of the first day were
used as input for the automatic/semi-automatic
contouring strategies. Automatic contouring software
(ADMIRE research v1.13.5 Elekta AB, Stockholm, Sweden)
was used for MR based deformable registration and
contour propagation. For the automatic contouring
strategy the daily propagated contours were based on an
intra-patient atlas consisting of the manual contours of
the first day and propagated contours of other previous
days.The semi-automatic contouring strategy included
additional manual adjustments made by a technologist
after each daily automatic contour propagation serving as
input for the following days. All automatic and semi-
automatic contours were compared with the manual
contours of the corresponding day by calculating dice
coefficients, mean and Hausdorff distances. Timing
measurements were done for both strategies.
Results
Higher median dice coefficients with smaller ranges were
found for the semi-automatic strategy compared to the
automatic strategy (figure 1). However, large variations
after manual adjustments were still found for the GTV.
Outliers found in the mean and Hausdorff distances of the
automatic strategy were not seen in the semi-automatic
strategy (figure 2).
The contours were automatically propagated for day 2, 3,
4 and 5 in respectively 18, 38, 54 seconds and 1:13 minutes
on average. The propagated contours of the semi-
automatic strategy were manual adjusted with an average
time of 14:49 minutes (in comparison with approximately
45 minutes for full manual contouring). Manual
adjustments of the cranial and caudal slices of the
contours were most time consuming.