ESTRO 36 Abstract Book

943 / 1082

S927 ESTRO 36 2017
_______________________________________________________________________________________________
Conclusion
Registration remains accurate even for as little as 10% of
projection data, but with significant limitations in visual
image quality at 10 phase reconstructions and motion
detection at 3 phases. Simulating 25% dose over 10 phases
allows for accurate registration without significant loss of
image quality or motion detection and is therefore
acceptable for 4D CBCT. This result is particularly relevant
for patients with a good prognosis as it limits the radiation
exposure.
EP-1714 Automatic delineation of the gross-tumour
volume in prostate cancer using shape models
K. Cheng1, Y. Feng1 , D. Montgomery1, D.B. McLaren2, S.
McLaughlin3, W. Nailon1
1 Edinburgh Cancer Centre Western General Hospital,
Department of Oncology Physics, Edinburgh, United
Kingdom
2 Edinburgh Cancer Centre Western General Hospital,
Department of Clinical Oncology, Edinburgh, United
Kingdom
3 Heriot Watt University, School of Engineering and
Physical Sciences, Edinburgh, United Kingdom
Purpose or Objective
Digital models of anatomy have potential for assisting in
the segmentation of the prostate and organs at risk (OAR)
in radiotherapy planning of prostate cancer. However,
manual alteration of automatically generated contours is
often necessary to produce an accurate gross-tumour
volume (GTV). This is generally the case when the tumour
extends beyond the prostatic capsule into the bladder or
invades the seminal vesicles (>T3a tumours). The aim of
this study was to develop a digital model of the GTV in
prostate cancer that incorporates the range of shape
variability associated with different T-stages.
Material and Methods
Computerised tomography (CT) images from 42 prostate
cancer patients, which contained a range of T-stages and
had the prostate GTV and OARs outlined, were selected.
Three-dimensional (3D) isotropic volumes were created
for each data set and the correspondence between points
on the GTV surface of each case was established. To fit
the model to an unknown data set texture analysis
features were calculated in 5x5 volumes perpendicular to
the boundary between the interior and exterior of the GTV
surface. This estimates the location of the GTV boundary.
At each point a search for the GTV shape was conducted
by calculating the texture features and moving within the
established shape limits until reaching convergence.
Training was performed on 32 randomly selected cases and
testing on the remaining 10 cases. The Dice similarity
coefficient was used to compare the model results with
the clinically defined volumes.
Results
Figure 1 shows an example of a typical GTV produced by
the algorithm in which the seminal vesicles have been
included in the apical slice (left). Table 1 shows a
summary of the results obtained on the 10 test cases. The
mean clinical volume of the test cases was 64.5 cm3 and
calculated by the model was 60.3 cm3. The largest
difference was observed in the cases with the largest GTV.
Figure 1: Clinical contour in green, model result in red.
Left: apical slice taken form the inferior of the GTV
including the seminal vesicles. Middle: central slice of the
GTV shape. Right: basal slice from the superior of the GTV.
Table 1: Dice coefficients and volumes obtained on the 10
test cases.
Conclusion
The proposed model has potential for automatically
contouring the GTV when the tumour extends beyond the
prostatic capsule into the bladder or invades the seminal
vesicles. However, more cases must be included in the
model to ensure that the full range of shape variability is
represented.
EP-1715 Differences in delineation uncertainty using
MR images only vs CT-MR in recurrent gynaecological
GTV
D. Bernstein1, A. Taylor1, S. Nill 1, U. Oelfke1
1Royal Marsden Hospital Trust & Institute of Cancer
Research, Department of Medical Physics, London,
United Kingdom
Purpose or Objective
To build upon previous work [1] to utilise a new contouring
concept to quantify the differences in delineation
uncertainty when using co-registered CT-MR images vs MRI
only for recurrent gynaecological GTVs.
Material and Methods
A contouring concept was developed in which clinicians
draw up to two GTV boundaries per CT slice corresponding
to the inner (GTV_i) and outermost (GTV_o) possible
boundaries the GTV may have and therefore define a
boundary interval for the GTV.
Observers contoured centrally recurrent gynaecological
GTVs in accordance with this concept first on MRI images
and then on co-registered CT-MR images to allow for their
comparison. A rigid soft tissue registration localised