S233
ESTRO 36
_______________________________________________________________________________________________
response can be quantified and predicted using Tumor
Metabolic Ratio (TMR) matrix obtained during the early
treatment weeks from multiple FDG-PET imaging.
Material and Methods
FDG-PET/CT images of 15 HN cancer patients obtained
pre- and weekly during the treatment were used. TMR was
constructed following voxel-by-voxel deformable image
registration. TMR of each tumor voxel,
v
, was a function
of the pre-treatment SUV and the delivered dose,
d
, such
as TMR(
v
,
d
) = SUV(
v
,
d
)/SUV(
v
, 0). Utilizing all voxel
values of TMR in the controlled tumor group at the last
treatment week, a bounding function between the pre-
treatment SUV and TMR was formed, and applied in early
treatment days for all tumor voxels to model a tumor voxel
control probability (TVCP). At the treatment week
k
, TVCP
of each tumor voxel was constructed based on its pre-
treatment SUV and TMR obtained at the week
k
using the
maximum likelihood estimation on the Poisson TCP model
for all dose levels. The DPF at the week
k
was created
selecting the maximum TVCP at each level of the pre-
treatment SUV and TMR measured at the week
k
. In
addition, 150Gy was used as an upper limit for the target
dose.
Results
TVCP estimated in the early treatment week, i.e. week 2,
had their D
50
=13~65Gy; g
50
= 0.56~1.6 respectively with
respect to TMR = 0.4~1.2; Pre-treatment SUV = 3.5~16.
Figure 1 shows the TVCP estimated using the TMR
measured at the week 2 with different levels of pre-
treatment SUV, as well as TVCP at different weeks, the
week 2 ~ week 4. Large dose will be required to achieve
the same level of tumor control for the same level of TMR
appeared in the later week of treatment. Figure 2 shows
the corresponding DPF for the week 3 TMR, as well as the
prescribed tumor dose distribution for the 3 failures.
Figure
1
Figure 2
Conclusion
DPF can be estimated and constructed adaptively voxel-
by-voxel in human tumor using multiple FDG-PET imaging
obtained during the treatment course. DPF provides a
potential quantitative objective for adaptive DPbN to plan
the best clinical dose, escalate or de-escalate, in human
tumor based on its own radiosensitivity or radioresistance.
OC-0442 Intensity based synthetic CT generation from
standard T2-weighted MR images with three MR
scanners
L. Koivula
1
, L. Wee
2
, J. Dowling
3
, P. Greer
4
, T. Seppälä
1
,
J. Korhonen
1
1
Comprehensive Cancer Center- Helsinki University
Central Hospital, Department of radiation oncolocy,
Helsinki, Finland
2
Danish Colorectal Cancer Center South, Vejle Hospital,
Vejle, Denmark
3
Commonwealth Scientific and Industrial Research
Organisation CSIRO, CSIRO ICT Centre, Brisbane,
Australia
4
Calvary Mater Newcastle Hospital, Radiation Oncology,
Newcastle, Australia
Purpose or Objective
Recent studies have shown feasibility t o conduct the
entire radiotherapy treatment planning workflow relying
solely on magnetic resonance imaging (M RI). Yet, few
hospitals have implemented the MRI-only workflow into
clinical routine. One limiting issue is the requisite
construction of a synthetic computed tomography (sCT)
image. The majority of published sCT generation methods
necessitate inclusion of extra sequences into the
simulation imaging protocol. This study aims to develop an
intensity-based sCT generation method that relies only on
image data from standard T2-weighted sequence. The
work includes images derived from three different
manufacturers’ MR scanners. The primary target group
was prostate, for which T2-weighted images are already
used as standard target delineation images.
Material and Methods
The study utilized a total of 30 standard T2-weighted
images acquired for prostate target delineation in three
different clinics. The imaging was conducted with MR
scanners (GE Optima 1.5T, Philips Ingenia 1.5T, and
Siemens Skyra 3.0T) of each participating clinic by using
their typical clinical settings. Intensity value variations of
the obtained images were studied locally, and compared
to corresponding Hounsfield units (HUs) of a standard CT
image. The data of 21 of the 30 prostate patients was used
to generate conversion models for bony and soft tissues to
transform the MR image into sCT. The models were
optimized separately for the images obtained by each MR
platform. The sCT generation was tested for 9 of the 30