119 / 1082
S106
ESTRO 36 2017
_______________________________________________________________________________________________
comprehensive study regarding follow-up after CovP based
nodal boosting can be expected.
Symposium: Ultra fast online therapy adaptation
(replanning, dose accumulation QA)
SP-0212 Automatic image segmentation and structure
evaluation for on-line adaptive RT
S. Mutic
1
1
Washington University School of Medicine, Department
of Radiation Oncology, St. Louis, USA
The online adaptive radiotherapy (OART) has become a
practical reality in recent years. Through experience, we
have learned that the conventional radiotherapy planning
(RTP) practices are not directly translatable to
OART. With the OART, the entire planning process needs
to be performed very quickly and generally can take no
longer than 20-30 minutes for imaging, segmentation,
planning, and quality assurance. The efficiency
requirements of OART mean that each of the treatment
planning steps needs to be performed in minutes rather
than in hours or days, which can be afforded with the
conventional RTP processes. This rate of efficiency
demands a fundamental revisiting of the paradigms used
in conventional RTP. One of the major differences
between the OART and conventional RTP is image
segmentation and processing of the segmented
structures. For image segmentation, the OART efficiency
needs mean that 1) there will be a significantly higher
degree of reliance on auto-segmentation, 2) that few
structures may be used\delineated, 3) that the
conventional paradigms for structure creation will not be
followed and that some structure will be contoured only
to a limited extent, or 4) a combination of all three of
these approaches. Unfortunate reality is that even the
most sophisticated modern auto-contouring algorithms
still have an unacceptably high degree of failure and
inaccuracy and that these algorithms are almost always
guaranteed to need some degree of manual
editing. Additionally, many practical clinical cases are
proving themselves to be not very good candidates for
auto-contouring and that manual segmentation may be
the best\most practical approach. The suboptimal
performance of auto-contouring algorithms and use of a
significant amount of available time on manual contouring
or contour editing means that typically there is not much
time left for contour validation. This lack of time for
contour validation means that there is an increasing need
for automatic evaluation of the segmented structures. It
is well understood that suboptimal structure delineation
(target or normal organs) can lead to suboptimal or unsafe
treatment plans. Inaccuracies in structure delineation can
translate to errors in treatment planning with OART more
likely than with conventional RTP due the limited time
available for quality assurance of segmented structures
with OART. Many bodies have recommended that peer
review process should be used as the second check of
accuracy of organ delineation. With OART, prospective
peer review will likely never be practical and the accuracy
of auto-segmentation as well as the robustness of
automatic evaluation of segmented structures will have to
be able to compensate for limited ability for independent
human checks. This presentation includes discussion of
current practices in OART image segmentation with review
of disease sites which are leading in the in the initial use
of OART. Also discussed will be the image segmentation
approaches for OART. Finally, efforts for development of
automatic methods for verification of accuracy of
segmented structures will be discussed. Now that OART is
a practical reality, it is obvious that additional work is
needed in development of image segmentation algorithms
as well as development of automatic methods for
verification of segmented structures.
SP-0213 Ultra-fast treatment planning and dose
reconstruction
P. Ziegenhein
1
1
The Institute of Cancer Research and The Royal Marsden
NHS Foundation Trust, Joint Department of Physics,
Sutton, United Kingdom
A new generation of hybrid MRI and linear accelerator
machines, such as the MR-linac, is currently brought into
practice which allows monitoring the changing patient
anatomy during radiation delivery. The newly acquired
images cannot only be used for real-time position
verification but also to inform a re-planning strategy
which adapts the treatment to the latest geometries. This
new technique demands for a more interactive therapy
workflow than it is used today. Treatment planning and
dose verification steps need to be carried out more
frequently and faster in order to make use of the
continuously updated patient images.
In this talk we will address two vital aspects of realizing
an online therapy adaptation workflow: ultrafast
treatment planning and dose reconstruction. Nowadays,
with the help of modern computational hardware both
operations can be performed in real-time. We will present
state-of-the-art techniques designed especially for
multi/many-core CPUs and discuss opportunities and
challenges of their application. A proof of concept study
on realistic patient data will be presented while
alternative techniques and methods are analyzed and
critically evaluated.
SP-0214 Online tumour tracking – technology and
quality assurance
E. Colvill
1
1
Aarhus University Hospital, Radiation Oncology, Aarhus
C, Denmark
Intrafraction motion during radiotherapy delivery causes a
blurring of the delivered dose distribution. For treatment
sites affected by respiratory motion, including lung, liver
and pancreas this effect can result in substantial
deviations between planned and delivered dose,
potentially compromising clinical goals. Treatment
delivery accuracy may be increased through the
implementation of adaptive delivery techniques. Online
tumour tracking adapts the treatment to anatomical
changes which may occur on the scale of seconds or
minutes. It combines monitoring of the target motion and
adaptation to that motion in real-time. In this talk a
review of motion monitoring and treatment adaptation