Download University of Windsor, Windsor, Ontario Sunnybrook Health Science

Automated Analysis of HDR Brachytherapy Treatment Plans
Samuel
1 University
1
Rusu ,
of Windsor, Windsor, Ontario
Matt
2Sunnybrook
2
Wronski
Health Science Centre, Toronto, Ontario.
RESULTS (continued)
INTRODUCTION
RESULTS
Cancer is the leading cause of premature death in Canada.
Based on 2015 estimates about 2 out of 5 Canadians are expected
to develop cancer during their lifetimes and the number of new
cancer cases in Canada is expected to rise about 40% in the next
15 years [1]. Analyzing the effectiveness of current treatment
methods, investigating ways to speed them up, and developing
new methods for cancer treatment is therefore very important.
n = 19
Part 1: Assessing the need for daily plan adaptation
To determine if daily re-planning is necessary we looked at the
D2CC (the 2cm3 that receives the most dose) of the rectum and the
D90 (dose delivered to 90% of the volume) of the target volume.
Gynecological (GYN) Interstitial High-Dose-Rate (HDR)
Brachytherapy is the newest Brachytherapy program at Odette
Cancer Centre. Patients are treated in either 3 or 4 fractions with a
single HDR implant or 2 implants 1 week apart. It is not currently
clear if daily plan adaptation is needed. We investigate if daily plan
adaptation is needed (i.e. if re-planning at every fraction would
provide a dosimetric benefit) for GYN Interstitial HDR patients . If
daily plan adaptation is needed/beneficial then we investigate
some methods we can use to reduce contouring time which is the
current key bottleneck.
METHOD
To do this study we developed a DICOM data mining framework
for the automated analysis of HDR brachytherapy treatment plans.
This framework has different DICOM data inputs, currently, the
Oncentra treatment planning system, and MIM Vista. The data is
organized in the framework and different analysis modules were
developed to analyze the data.
To see the dosimetric impact of contour uncertainties associated
with DIR-contours we looked at the D2cc ratio between the
manual D2cc and the DIR D2cc.
DISCUSSION
Part 2: Possible techniques to reduce contouring time
Investigation #1: Reduction of contour volumes based on
dosimetric sensitivity
In HDR, rectal planning metrics (D2cc) do not require full OAR
recontouring at each fraction. To identify an OAR region (“partial
OAR volume”) proximal to the target volume (HRCTV) that produces
accurate D2cc values for replanning purposes partial OAR volumes
were constructed by an OAR intersected with HRCTV expansion. The
HRCTV expansions studied were 0.5, 0.75, 1, 1.25, 1.5, 1.75 and
2cm.
1.5 cm HRCTV/GTV
Expansion
n = 40
This data suggests that for GYN Interstitial HDR, replanning at
every fraction would be beneficial for a number of patients .
In the first investigation we demonstrated the feasibility of using
partial OAR volumes (1.5 cm proximal to HRCTV) for contouring in
adaptive re-planning. In the second investigation we showed that
for rectum, if the DICE coefficients > 0.8 it produces a D2cc
accuracy within +/- 5%. Therefore if some difference is acceptable
in the plans using these two methods one can speed up the
contouring process.
In the future, we plan on re-running analysis modules for a larger
patient cohort, and we plan on developing interstitial implant
quality metrics and other analysis modules.
ACKNOWLEDGEMENTS
Special thanks to funding from Cancer Care Ontario.
REFERENCES
To assess the need of daily adaptation we looked at the dose
impact of the Day 1 plan made based on the Day 1 CT (P1_CT1),
the Day 1 plan projected onto Day 2 CT (P1_CT2), and the Day 2
plan based on Day 2 CT (P2_CT2). The possible techniques to
reduce contouring time investigated were the reduction of
contour volumes based on dosimetric sensitivity and contour
propagation using deformable image registration (DIR). The organ
at risk (OAR) we focused on in this study was the rectum.
Investigation #2: Contour Propagation using MIM Deformable
Image-Registration (DIR)
Using Deformable Image Registration (DIR), Day 1 OAR contours
can be propagated onto the Day 2 image set. To see how DIRgenerated contours compare with manual contours the DICE
correlation coefficient [2] was used.
[1] "New Cancer Cases Expected to Rise Dramatically within 15
Years -." Www.cancer.ca. Web. 5 Aug. 2015.
[2] The need for application-based adaptation of deformable
image registration. Kirby, Neil and Chuang, Cynthia and Ueda,
Utako and Pouliot, Jean, Medical Physics, 40, 011702 (2013),
DOI:http://dx.doi.org/10.1118/1.4769114
Samuel Rusu ([email protected])
Program: Medical Physics Co-op 2015
CO-OP Work Term: Summer 2015
Employer: Sunnybrook Health Sciences Centre
Supervisor: Dr. Matt Wronski