944 / 1082
S928
ESTRO 36 2017
_______________________________________________________________________________________________
around the GTV was used. Observers also contoured the
GTV in accordance with local clinical protocols using CT-
MR images (GTV_c) and scored their confidence in these
contours on a scale of 1 to 5, 5 being complete confidence.
Differences between CT-MR and MR volumes were assessed
using paired T-Test. Inter-observer variability was
assessed using an analysis of variance. The threshold for
significance was p<0.05.
Results
Four observers participated in the study, three of whom
had at least 3 years’ experience in treating this patient
group; two contoured 20 cases, one 19 and one 18. A
summary of the results are presented in Table 1.
MR only based volumes were statistically significantly
smaller for all observers for each GTV_i, GTV_o and
GTV_c, whilst the confidence scores in GTV_c were higher
for all observers, although for one this was not statistically
significant.
The magnitude of the offset between the centres of GTV_i
and GTV_o was calculated and the difference in these
offsets compared between CT-MR and MR only contours.
An overall difference was detectable for only one
observer, in which the distance between GTV_i and GTV_o
were smaller in the MR only images. For the remaining
observers, the agreement in the position of the volumes
was not affected by the imaging modality.
Boundary intervals were smaller on the MR only images for
all observers, although this difference was not statistically
significant for Observer 3.
The impact of multimodality imaging on observer variation
for the boundary interval size was tested using 17 cases
contoured by all observers, by comparing the ratio of the
boundary interval size from CT-MR to MR only between
observers. This did not show a statistically significant
difference between the observers, with a p-value of
0.507.
Conclusion
Contouring centrally recurrent GTVs using only MR images,
instead of the current practice of co-registered CT-MR
images, produces smaller volumes. When using MR images
alone, clinicians have a higher confidence in their clinical
GTV contours as well as having lower delineation
uncertainties overall. The differences between CT-MR and
MR only boundary intervals did not vary between
observers. This reduction in uncertainties supports an MR-
based workflow.
[1] Bernstein, D., et al.,
Measurement of GTV delineation
uncertainty for centrally recurrent gynaecological
cancers.
Radiotherapy and Oncology, 2015.
119
(1): p.
5615-5616.
EP-1716 Semantic PACS deployment enables research
in a radiation oncology research environment
M.S. Marshall
1
, H. Beemster
1
, M. Buiter
1
, T. Janssen
1
1
Netherlands Cancer Institute Antoni van Leeuwenhoek
Hospital, Clinical Physics, Amsterdam, The Netherlands
Purpose or Objective
Research involving imaging data requires intensive use of
costly DICOM operations in order to aggregate the data for
research. We installed software SeDI (Semantic DICOM,
developed by SOHARD) which creates a database
containing all radiotherapy-related DICOM metadata using
a Semantic PACS that extracts and stores header tags in a
special type of database. The resulting database can be
searched with the standardized query language SPARQL
using terms from DICOM and a radiation oncology
vocabulary. Once all relevant DICOM tags have been
extracted by SeDI, DICOM search is significantly
accelerated and can often be expressed as a single query.
The aim of this work is to show how SeDI helps to quickly
resolve typical DICOM queries.
Material and Methods
In order to quantify the potential speedup for a
representative task using SeDI we measured the time
spent to retrieve: the UIDs of the RTDose, RTStruct and
RTPlan for a set of patients where 1) a relevant organ at
risk is delineated and 2) the required DICOM objects exist
in the clinical database.
Such a question is relevant if, for example, one wants to
calculate DVHs for a large number of patients. Our
conventional approach to such a problem would be to use
a simple tool that browses through all patients on the
DICOM server, find the associated RTStruct, check
whether the organ at risk is delineated, and check if the
associated RTDose exists in a RTPlan that makes use of the
RTStruct. Using the logfiles of the tool that performs these
steps, we can estimate the efficiency by using a direct
indexing of the relevant DICOM objects using SeDI.
Results
Using the mentioned logfiles, we found that the above
question was asked for 7384 patients. Lookups for 2176
patients found the DICOM objects that met all criteria in
a total time of 14.24 hours with an average DICOM lookup
time of 23.6 seconds. Lookups for 5208 patients failed to
find matching DICOM objects at a total cost of 23.48 hours
and an average DICOM lookup time of 16.2 seconds. Since
failed lookups would not occur using SeDI, the use of SeDI
would save at least the 71% of the search time spent on
non-matching queries in our study question, amounting to
23.48 hours saved out of 37.72 total hours. By direct
indexing of the relevant objects further time decrease is
expected, but the magnitude of this remains open for
study.
Conclusion
We expect that SeDI will enable the researchers to rapidly
identify the right DICOM objects for calculations, as well
as accelerate ongoing research in outcome prediction,
toxicity modelling and radiomics.
EP-1717 Image Quality Comparison Between Two
Radiotherapy Simulators
N. Tomic
1
, P. Papaconstadopoulos
1
, J. Seuntjens
1
, F.
DeBlois
1
, S. Devic
1
1
McGill University, Oncology, Montreal, Canada
Purpose or Objective
In this work we compare image quality parameters derived
from phantom images taken on two CT simulators most
commonly used in radiotherapy departments. To make an
unbiased comparison, CT images were obtained with CT
scanning protocols leading to the same surface doses,
measured using XR-QA2 model GafChromic film reference
dosimetry protocol.
Material and Methods
Two radiotherapy CT simulators GE LS 16 (80 cm bore size)
and Philips Brilliance Big Bore (85 cm bore size) were
compared in terms of image quality using CATPHAN-504,
scanned with Head and Pelvis protocols. Dose was
measured at the phantom surface with CT scans taken
until doses during CT scans on both scanners were within
5%. Dose profiles were sampled using XR-QA2 model
GafChromic
TM
film strips placed at four sides of the
phantom (top, bottom, left, and right) and taped with a