148 / 1082
S135
ESTRO 36 2017
_______________________________________________________________________________________________
Using multi-atlas (MIR) compared to single-atlas (RS) for
prostate RT showed better in 19, similar in 2 and worse in
3 out of the 24 volumetric parameters. Even though the
results are promising, discordances could lead to large
DVH parameters differences.
OC-0264 Assessment of Atlas based auto-segmentation
in Breast Contouring
P.H. Fok
1
, W.K. Fung
1
, C.H.P. Ho
1
, G. Chiu
1
1
Hong Kong Sanatorium & Hospital, Department of
Radiotherapy, Happy Valley, Hong Kong SAR China
Purpose or Objective
To investigate the efficiency and accuracy of atlas based
auto-contour segmentation (ABAS) for radiotherapy of
breast cancer patients.
Material and Methods
Contour sets of 80 breast cancer patients were recruited
to build an atlas in MIM Software for ABAS purpose. They
consisted of structures including CTV, PTV, and OARs
including the esophagus, heart, liver, spinal cord and
contralateral breast. ABAS was evaluated by comparing
auto segmented contours with manually drawn contours
using Dice Similarity coefficients (DSC) and Hausdroff
distance (HD). CTV and contralateral breast were
dissected into superior, inferior, medial, lateral, anterior
and posterior portions (Fig.1) for individual evaluations of
their accuracy of ABAS. The study consisted of two parts.
Part A: All patients from the atlas were used as the test
subjects and the patient who was selected as the test
subject would be excluded from the atlas. Manual
contours of each subject were compared with those that
were generated by ABAS. Part B: Another 16 patients with
breast cancer were randomly recruited to evaluate the
efficacy of ABAS in clinical practice. The time required for
manual contouring and contouring by ABAS with or without
manual refinement were compared. Radiation Oncologists
(ROs) were asked for their satisfactory levels regarding the
ABAS generated contours.
Fig.1
Results
Part A
The comparison results of the ABAS generated contours
and the manual contours were listed in Table 1. High levels
of agreement (DSC=0.8 to 1) were shown in the right lung,
left lung, liver, heart, spinal cord and left breast PTV;
moderate levels of agreement (DSC=0.7 to 0.79) were
shown in the left breast (CTV, contralateral breast) and
the right breast (CTV, PTV, contralateral breast); poor
agreement (DSC<0.69) was shown in the esophagus.
The lateral portions of the CTV, PTV and contralateral
breast for both left and right sided breast cancer patients
had the largest volume differences among the other
portions. Thus, the lateral portion of the breast contour
required the most refinement.
Part B
Level of agreement and s atisfaction improvement were
shown after refinement on AB AS generated TVs, with the
mean DSC increased from 0.861 to 0.979 and the
satisfaction score (range:1-5 ) increased from 2.2 to 4.3.
The mean time for ROs to manually contour TVs and to
modify ABAS generated TVs with refinement was 26.8 and
15.3mins respectively. The DSC, mean time for manually
contouring OARs and that for ABAS generated contours
with refinement were as follows: heart (DSC=0.949; 8.2
and 7.3mins), liver (DSC=0.904; 24.8 and 20.5mins),
esophagus (DSC=0.556; 7.6 and 11.3mins), spinal cord
(DSC=0.855; 9.2 and 10.4mins) and contralateral breast
(DSC=0.904; 24 and 18mins).
Conclusion
ABAS could speed up the contouring process especially in
target delineation during breast patient treatment
planning. However, improvement in the a lgorithm is still
needed and auto post-ABAS refinement has to be
developed to further increase its efficiency for clinical
practice.
OC-0265 Evaluating variability of contouring using
ESTRO guidelines for elective breast cancer
radiotherapy
D. Megias
1
, M. Sydenham
2
, D. Wheatley
3
, M. Maclennan
4
,
E. Spezi
5
, A.M. Brunt
6
1
Mount Vernon Hospital, National Radiotherapy Trials
Quality Assurance Group RTTQA, Northwood, United
Kingdom
2
Institute of Cancer Research, ICR Clinical Trials and
Statistics Unit, Sutton, United Kingdom
3
Royal Cornwall Hospital, Clinical Oncology, Truro,
United Kingdom
4
Edinburgh Cancer Centre, Cancer Services, Edinburgh,
United Kingdom
5
Cardiff University, School of Engineering, Cardiff,
United Kingdom
6
Royal Stoke University Hospital & Keele University,
Cancer Centre, Stoke on Trent, United Kingdom
Purpose or Objective
Outlining of target and OAR volumes is an integral part of
the radiotherapy pathway but inherently subject to
variability. With an emerging body of evidence supporting
regional nodal irradiation in early breast cancer, there is
continued emphasis for optimal, standardised and