491 / 1082
S476
ESTRO 36 2017
_______________________________________________________________________________________________
Conclusion
The use of plan adaptations to correct patient setup errors
was experimentally validated on the MRL for the first
time. The reference dose distribution was reproduced at
the shifted location for rectum cancer patients. In
addition, high gamma pass rates were measured.
Acknowledgements:
The autors like to thank Robert Spaninks (Elekta).
PO-0879 Differences between planned and delivered
maximum spinal cord dose in Head &Neck cancer
patients
D. Noble
1,2
, P. Yeap
3
, K. Harrison
3
, S. Thomas
1,4
, M.
Parker
3
, N. Burnet
1,2
1
VoxTox Research Group - University of Cambridge.,
Oncology, Cambridge, United Kingdom
2
Addenbrooke's Hospital - Oncology Centre University of
Cambridge, Oncology, Cambridge, United Kingdom
3
VoxTox Research Group - University of Cambridge.,
Cavendish Laboratory- Department of High Energy
Physics, Cambridge, United Kingdo
4
Addenbrooke's Hospital - Oncology Centre University of
Cambridge, Medical Physics, Cambridge, United Kingdom
Purpose or Objective
Adaptive radiotherapy (ART) for head and neck cancer
remains resource intensive, and there is little consensus
on which patients will benefit most from having it done.
Concerns regarding maximum spinal cord dose sometimes
trigger re-planning at our centre, and we sought to
compute and model differences between planned and
delivered maximum dose to the spinal cord in patients
undergoing IMRT with daily image-guidance (IG) on the
TomoTherapy system.
Material and Methods
We drew planning kVCT, IG MVCT and planned dose
datasets from archive for 33 patients who were treated
for head & neck cancer (HNC) on TomoTherapy units at
our centre. All patients underwent daily IG, with matching
to high dose PTV (close to the spinal cord), or cervical
spine vertebrae. To automatically contour the spinal cord,
we developed an intensity based deformable image
registration (DIR) algorithm using the open source Elastix
toolkit to propagate manual contours from the planning
CT. Using ‘gold standard’ contours of an expert observer,
the algorithm was optimised on 30 MVCT datasets (567
slices) and validated on a further 90 (2203). Conformity
was measured with Jaccard conformity index (JCI) and
distance between centres (DBC), and compared with
results from intra- and inter-observer studies.
Using in-house dose recalculation software (CheckTomo),
TomoTherapy sinograms, MVCT datasets and algorithm
transformation parameters, we created ‘voxel histories’
for the spinal cord relative to the planning CT, and
calculated delivered dose. Maximum planned and
delivered spinal cord dose (D
2%
) were then compared.
Results
A summary of auto-contouring algorithm performance is
shown in Table 1. Auto-contouring performance appeared
comparable to manual segmentation, and we proceeded
to calculate delivered dose. These results are shown in
Figure 1 (A-C). Fig. 1A shows a waterfall plot of planned
D
2%
minus delivered D
2%
for each patient. Mean spinal cord
D
2%
was 35.96Gy (planned) and 36.01Gy (delivered), and
the mean absolute difference between planned and
delivered dose was 1.1Gy (3% of mean planned D
2%
).
Differences between planned and delivered dose were
plotted as a histogram, which appears to be normally
distributed around the mean difference (Fig 1B). The
mean difference (µ, -0.05) and standard deviation (σ,
1.448) were used to approximate a normal distribution to
this data – as shown in Fig.1C. Using this model, a z
statistic can be calculated for a chosen difference (e.g.
Prob. of delivered D
2%
being 4Gy higher than planned is
2.5%).
Conclusion
Differences between planned and delivered D
2%
to the
spinal cord in patients receiving daily IG are small in HNC
patients treated with daily IG on TomoTherapy. Our model