Biomedical Engineering Reference
In-Depth Information
• Optimization of the steering parameters.
• Determination of the resulting treatment effects.
• Visualization and post-processing.
medical images obtained by MR, CT, etc. Segmentation is the
task of identifying the image regions belonging to a tissue and is
an important topic in the field of computer vision. A vast range
of literature on this subject exists (e.g., for overviews see [80,
144] and for more detailed reviews on vessel segmentation see
[47, 58, 89, 161, 162]). Compared to radiotherapy treatment plan-
ning, segmentation for hyperthermia treatment planning is far
more demanding as more structures have to be distinguished
with high accuracy (the dielectric and thermal parameters vary
strongly between tissues). Also, the exact shape of tissue inter-
faces can be highly relevant for hot spot prediction, especially in
regions with resonances such as the pelvis. The only alternative
to detailed segmentation would be to determine corresponding
tissue parameter maps for the patient using medical imaging,
which has not been achieved to date.
Segmentation methods can be classified according to many
criteria: (1) how automatic or interactive they are, (2) whether
they work on slices or volumes, (3) whether they identify regions
by using a homogeneity criterion (gray level/texture based) or
if instead they try to find borders, (4) whether some prior or
statistical knowledge is used, (5) whether they are based on
local or global image information, (6) whether they identify
single regions or use competitive approaches to identify multiple
regions at the same time, (7) whether they find a static solution
or a dynamic solution such as a curve evolution over time, etc.
Which method is best suited depends on the image type (e.g.,
MRI, CT) and quality, the regions that have to be identified, the
required accuracy, and the available time.
Requirements for segmentation tools in the context of HTP
are the robustness of the segmentation approach, the necessity of
generating detailed models (far more important than in radio-
therapy as the differences in tissue parameters are considerable),
the quality of the interface to the simulation tools, the ease of use
and no requirement for highly specialized staff, and that they
work with commonly available image data.
In the context of hyperthermia treatment planning, several
segmentation techniques have been used and suggested:
AMIRA HyperPlan offers a choice between manual segmen-
tation with a brush, a live-wire type delineation, manual thresh-
olding, a simple thresholded region growing with the possibility
of user-specified limits, and a type of evolving boundary method
(perhaps based on fuzzy connectedness or a fast marching
method). Interpolation can be used between slices, various filters
for pre-processing the image data are available, and surfaces can be
extracted and simplified based on the segmentation results. [191]
compared the speed and accuracy of manual, live-wire, region/
volume growing and watershed based segmentation. They con-
cluded that while region/volume growing and watershed based
segmentation are suitable, a well-interfaced manual segmentation
can sometimes be faster as well as more accurate. [191] further
concluded that 3D segmentation can be faster than 2D segmen-
tation if good interactions and correction routines are provided.
Contouring was also used in [86]. An automatic segmentation
technique for the thigh to be used in HTP was presented in [141].
[165] used a thresholding technique.
It has been established that high-resolution treatment planning
is required to correctly capture hot spots and reliably predict field
distributions [166]. Only 3D simulations can correctly describe the
temperature increase distribution generated by complex applicators
[158]. While temperature increase is likely to be the relevant factor
and the specific absorption rate (SAR) distribution does not corre-
late well with the temperature increase steady state [94] (due to ther-
mal conduction, tissue specific perfusion, and boundary effects),
temperature simulations carry additional uncertainties such as cor-
rect thermal model, large inter-patient variability of thermal tissue
properties, etc. It is therefore unclear, and disputed, whether SAR-
based or temperature-based planning is preferable [158].
7.2 Hyperthermia treatment
planning (Htp)
HTP [102, 103], while recommended by European Society for
Hyperthermic Oncology (ESHO) quality assurance guidelines
[104,180], is currently rarely performed clinically. Most of the
existing codes for HTP have been written and used for research
purposes [82,160,165,192,198]. They are usually optimized for
the variant of hyperthermia used at the specific institute and
cannot be easily generalized.
The commercially available treatment planning tool
HyperPlan is based on the AMIRA medical image analysis soft-
ware and has been developed mostly by the group of Professor
Peter Wust at the Charite Berlin and the Konrad Zuse Institute
Berlin (ZIB). It is designed for and sold with the BSD deep hyper-
thermia systems. A variant of it has been adapted for nanofluid-
based hyperthermia [69]. HyperPlan has also been adapted to
offer an interface to MRI thermometry.
An alternative HTP tool is HYCAT. This tool was developed
by the IT'IS Foundation [122] and has been introduced clini-
cally by the HT group of Dr. Van Rhoon (Erasmus MC-Daniel
den Hoed, Rotterdam). HYCAT is based on the EM simula-
tion engine of SEMCAD
x
(SPEAG, Switzerland) and is com-
plemented with the segmentation tool iSEG (Zurich Med Tech
AG, Switzerland), thermo simulation software, multiple field
optimizers, a Python scripting framework, and tools for effect
quantification and post-processing, as well as wizards to reduce
human error and simplify the setup of simulations. HYCAT has
been applied in treatment planning [44, 135], improving exist-
ing and developing novel applicators [136], investigating RF and
HIFU ablation, and helping to develop treatment guidelines
(e.g., water bolus shape and temperature) [44, 171].
7.3 Segmentation
The accurate patient model required for EM and thermal sim-
ulations is obtained by segmentation of three-dimensional
