TU/e

advertisement
9
Image Processing on Diagnostic Workstations
from simple PACS viewing into a really versatile
BART M. TER HAAR ROMENY
viewing environment. This chapter gives an overProfessor, Eindhoven University of Technology,
Department of Biomedical Engineering, Image
Analysis and Interpretation, PO Box 513, WH 2.106,
5600 MB Eindhoven, The Netherlands
view of these developments, aimed at radiologists’
readership. Many references and internet links
are given which discuss the topics in more depth
than is possible in this short paper. This paper is
CONTENTS
necessarily incomplete.
9.1
Introduction
9.2
Hardware
9.3
Software
9.4
3D Visualization
9.5
Computer Aided Detection (CAD)
9.6
Atlases
Syngo X, GE Advantage, etc.). There are dedicat-
9.7
CAD/CAM Design
ed companies for viewing software (a.o. Merge
9.8
Diffusion Tensor Imaging (DTI) - Tractography
eFilm) or OEM solutions (a.o. Mercury Visage,
9.9
Registration
Barco). The application domain of workstations is
9.10
RT Dose Planning
increasing. We now see them regularly employed
9.11
Quantitative Image Analysis
in PACS and teleradiology diagnostic review,
9.12
Workstations for Life Sciences
9.13
Computer-Aided Surgery (CAS)
9.14
New Developments
9.15
Outlook
Viewing stations are core business in a radiologist’s daily work. All big medical imaging
industries supply professional and integrated environments (such as Philips ViewForum, Siemens
3D/3D-time (4D) visualization, computer-aided
detection (CAD), quantitative image analysis,
computer-assisted surgery (CAS), radiotherapy
treatment planning, and pathology. Also the appli-
References
cations for medical image analysis in the lifesciences research are increasing, due to the inScientific terms marked with

are explained in
Wikipedia: www.wikipedia.org
creased attention to small-animal scanning systems for molecular imaging, and the many types
of advanced microscopes (such as confocal mi-
9.1
Introduction
Medical workstations have developed into the superassistants of radiologists. The overwhelming production of images, hardware that rapidly became cheaper and powerful 3D visualization and quantitative
analysis software have all pushed the developments
croscopy and two-photon laser scanning microscopes), all giving huge 3D datasets. The focus
of this chapter is on image processing (also termed
image analysis or computer vision) applications.
9.2
Hardware
plains why increasing the RAM of a slow computer can markedly upgrade its performance. In a
Early systems were based on expensive hardware
PACS system, the disk storage is typically done
platforms, called workstations, often based on the
on a ‘redundant array of inexpensive disks’


UNIX operating system . Today, most systems are
(RAID), where many disks in parallel prevent
based on readily available and affordable PC and
loss of data in case of failure of a component.
Mac hardware platforms (running MS-Windows or
Mac-OS respectively), which are still following
Moore’s law of increasing performance (a doubling
every 24 months) at a stable price level.
The central processor unit (CPU) is the core
of the system, running today at several Gigahertz,
and performance is expressed in Giga-FLOPS (109
floating point operations per second). Famous
CPUs are the Intel Pentium chip, and the AMD Athlon processor. Today, we see the current 32 bit processors being replaced by 64 bit processors, which
are capable of processing more instructions simultaneously and addressing a larger number of memory
elements (232 = 4.2 × 109, so a 32 bit system cannot
have more than 4.2 GB of memory(109 = Giga)).
There is also a trend to have more CPUs (‘dualcore’)
on the motherboard, paving the way to parallel
processing, which is currently still in its infancy.
The memory in the diagnostic workstation is
organized in a hierarchical fashion. From small to
large: the CPU has a so-called cache on its chip, as
a local memory scratchpad for super-fast access, and
communicates with the main RAM (random access
memory, today typically 1–4 GB) through the data
bus, a data highway in the computer. As the RAM
is fully electronic, access is fast (nanoseconds),
Fig. 9.1a–c. Brain aneurysm (a) and carotids (b):
much faster than access to a local hard disk (milli-
examples of volume renderings with a computer
seconds). When the RAM is fully occupied, the CPU
game graphics card (3Mensio Inc) (c)
starts communicating with the hard disk. This ex-
The speed of the network should be able to
They have finally become fully programmable
accommodate the network traffic. Typically the
(and can be instructed by languages as DirectX
workstation is part of a local area network (LAN).
and OpenGL) and are equipped with 1–
Today gigabit/second speeds are attained over wired
1.5 gigabytes of local memory. These ‘games’
networks, wireless is slower (30–100 Mbit/s) but
hardware boards are now increasingly used in 3D
convenient for laptops and ‘person digital assistants’
medical visualization applications (a.o. 3Mensio
(PDAs). Many PACS installations can be serviced
Medical Systems). There is also a community ex-
remotely through LAN connections to the supplier,
ploring the use of GPUs for general processing
anywhere.
(DICOM undated a).
Networks are so fast nowadays that 3D vol-
The viewing screens of diagnostic work-
ume rendering can be distributed from a central
stations have to be of special diagnostic quality.
powerful computer to simple (and thus low cost)
Excellent reviews of the important parameters
viewing stations, called ‘thin clients’ (a.o. Ter-
(resolution, contrast, brightness, 8, 10 or 12 bit
arecon Aquarius). A powerful dedicated graphics
intensity range, homogeneity, stability, viewing
board (in this case the VolumePro 1000) with dedi-
angle, speed, etc. are available in the so-called
cated hardware runs several 3D viewing applications
white papers by a variety of vendors (a.o. Barco –
simultaneously, and is remotely controlled by the
BARCO undated, Eizo – EIZO undated).
users of the thin clients. Advantage is the capability
to handle huge datasets (e.g. > 3000 slices) easily,
but scalability (to e.g. dozens of users) is limited.

cessing units’ (GPU , the processor on the video
card (or graphics accelerator card) in the system)
has increased even faster than CPU power, mainly
due to the fact that GPUs form the core of the computer game industry. The millions of systems needed
for this lucrative market and the high competition
between the market leaders NVIDIA and ATI have
created a huge performance/price ratio. A GPU has a
50 times faster communication speed of the data
internally between memory and processor, and has
dedicated hardware for rendering artificial environ
Software
The revolution in PACS (and teleradiology) view-
Interestingly, the power of ‘graphical pro
9.3
ing stations was fired by the standard “Digital
Imaging and Communications in Medicine”

(DICOM) standard (DICOM undated a), 4000
pages). In the 1990s the ACR (American College
of Radiology) and NEMA (National Electrical
Manufacturers Association) formed a joint committee to develop this standard. The standard is
developed in liaison with other standardization
organizations including CEN TC251 in Europe
and JIRA in Japan, with review also by other organizations including IEEE, HL7 and ANSI in the
USA. It is now widely accepted. Convenient short

ments, such as texture mapping , pixel shaders and
tutorials are available (BARCO undated). As the

an intrinsic parallel design with pixel pipelines .
scanners and viewing software continue to devel-
op, new features have to be added to the standard
into lower languages, like C, C++, Java. When
continuously. Vendors are required to make availa-
ultimate speed (and limited variability) is required,
ble their so-called conformity statements (see for
the code can be implemented in hardware (GPU,
example BURRONI et al. 2004), i.e. a specified list of
field programmable gate array’s (FPGA), dedi-
what conforms to the current version of the standard.
cated chips, etc.). Many packages offer software
The second revolution was the standardiza-
development kits for joint development (e.g.
tion of the internal procedural organization of medi-
MevisLab by MEVIS, ‘Insight Segmentation and
cal data handling in the ‘Health Level 7’ standard
Registration Toolkit’ (ITK) by NLM, etc.).
(HL7) (DICOM undated a).
The basic function of a viewing station is the
convenient viewing of the data, with a patient selection section. The functions are grouped in a so-called
‘graphical user interface’ (GUI). Versatile PC
based viewing packages are now widely available
(see RSNA 2006 for an extensive list), many also
offering ‘extended ASCI’

character sets for the
Chinese, Japanese and Korean markets.
3D Visualization
The first breakthrough in the use of workstations
has been by the invention of generating realistic
3D views from tomographic volume data in the
1980s. Now 3D volume rendering is fully interactive, at high resolution and real-time speed, and
with a wide variety of options, making it a nontrivial matter to use it.
Basic functions of the GUI include administrative functions as patient and study selection, report viewing and generation, and visualization functions as cine loop, ‘maximum intensity projection’

9.4
Many dedicated companies are now established (such as Vital Images with Vitrea, Mercury
Computer Systems with Amira, Barco with Voxar,
3Mensio with 3Vision, Terarecon with Aquarius,

(MIP), ‘multi-planar reformatting’ (MPR ) including oblique and curved reconstructions, cut planes,
etc.). Often a third party 3D viewing application is
integrated in the PACS viewing application, and
measurement tools for distances and angles, magni-
supplied as a complete system by such an ‘original
fying glass, annotations, etc.
equipment manufacturer’ (OEM).
The development of computer vision algorithms often follows a hierarchical pathway. The
design process (rapid prototyping) is done in high-
The principle of ray tracing (‘rendering’)
(NOWINSKI et al. 2005) is actually based on mimicking the physics of light reflection with the

level software (examples are Mathematica , Maple, Matlab), where very powerful statements and
algebraic functionality make up for very short code,
but his is difficult to extent to the huge multi-
computer: the value of a pixel in a 2D image of a
3D view (also called a 2.5D view) is calculated
from the reflected amount of light from a virtual
light source, either bouncing on the surface of the
dimensional medical images. When the formulas are
3D data (this process is called ‘surface render-
understood and stable, the implementation is made
ing’), or as the summation of all contributions
Fig. 9.3. Virtual colonoscopy with unfolding enables inspection of folds from all sides.
From VILANOVA ET AL. (2003)
from the inside of the 3D dataset along the line of
tings possible, users often get confused, and a
the ray in question, composed with a formula that
standard set of settings is supplied, e.g. for lung
takes into account the transparency (or the in-
vessels, skull, abdominal vascular, etc., or a set of
verse: the opacity) of the voxels (this process is
thumbnails is given with examples of presets,
called ‘volume rendering’).
from which the user can choose. Attempts are
underway to extract the optimal settings from the
statistics of the data itself (NOWINSKI et al. 2005).
In virtual endoscopy (e.g. colonoscopy)
the camera is positioned inside the 3D dataset.
Challenges for the computer vision application
are the automatic calculation of the optimal path
for the fly-through through the center of the
winding colon, bronchus or vessel. Clever new
visualizations have been designed to screen the
foldings in the colon for polyps at both the forward as backward pass simultaneously: unfolding
(VILANOVA 2003) (see Fig. 9.3) and viewing an
unfolded cube (VOS et al. 2003) (see Fig. 9.4).
Fig. 9.2. Volume rendering of the heart and coronaries (Terarecon Inc)
Segmentation is the process of dividing
the 3D dataset in meaningful entities, which are
The user can change the opacity settings
then visualized separately. It is essential for 3D
by manipulating the so-called ‘transfer func-
viewing by, e.g. cut-away views, and also, unfor-
tion’, this function giving the relation between
tunately, one of the most difficult issues in com-
the measured pixel value from the scanner and
puter vision. It is discussed in more detail in
the opacity. As there is an infinite number of set-
Sect. 9.5. When clear contrasts are available, such
as in contrast enhanced CT or MR angiography
and bone structures in CT, the simple techniques
of thresholding and region growing can be em-
9.5
Computer Aided Detec-
tion (CAD)
ployed, up to now the most often used segmenta-
One of the primary challenges of intelligent soft-
tion technique for 3D volume visualization.
ware in modern workstations is to assist the human expert in recognition and classification of
disease processes by clever computer vision algorithms. The often used term ‘computer-aided diagnosis’ may be an overstatement (better: ‘computer-aided detection’), as the final judgement
will remain with the radiologist. Typically, the
computer program marks a region on a medical
image with an annotation, as an attention sign to
inspect the location or area in further detail. The
task for the software developer is to translate the
Fig. 9.4. Unfolded cube projection in virtual colo-
detection strategy of the expert into an efficient,
noscopy. From VOS et al. (2003)
effective and robust computer vision algorithm.
Modern techniques are also based on (supervised
This also explains the popularity of maximum intensity projection, where pixels in the
2.5D view are determined from the maximum
along each ray from the viewing eye through the
dataset (such a diverging set of rays leads to a so-
and unsupervised) ‘data mining’ of huge imaging databases, to collect statistical appearances.
E.g. learning the shape and texture properties of a
lung nodule from 1500 or more patients in a
PACS database is now within reach. Excellent

called ‘perspective rendering’ ). As this may
easily lead to depth ambiguities, often the more
appealing ‘closest vessel projection’ (CVP) is
applied, where the local maximum values closest
to the viewer is taken. The sampled points of the
(oblique) rays through the dataset are mostly located in between the regular pixels, and are calculated by means of interpolation.
reviews exist on current CAD techniques and the
perspectives for CAD (DOI 2006; GILBERT and
LEMKE 2005). The field has just begun, and some
first successes have been achieved. However,
there is a huge amount of development still to be
done in years to come.
Some advances in CAD techniques that
have brought good progress are in the following
application areas.
Fig. 9.5a–c. Virtual colonoscopy with surface smoothing.
a Original dose (64 mAs); b 6.25 mAs; c 1.6 mAs. From PETERS (2006b)
Mammography: this has been the first
field where commercial applications found
ceeding some threshold are possible candidates
for further inspection.
ground, in particular due to the volume produc-
The location of the nipple is important as
tion of the associated screening, the high resolu-
a general coordinate origin for localization refer-
tion of the modality and the specific search tasks.
ences with, e.g. previous recordings. The general
Typical search tasks involve the automated detec-
statistical ‘flow’ of line structures points towards
tion of masses, micro-calcifications, stellate or
the nipple; the location can reasonably well be
spiculated tumors, and the location of the nipple.
found by modelling the apparent statistical line
How do such algorithms work? Let us
structure with physical flow models.
look in some detail to one example: stellate tumor
The role of MRI in breast screening is ris-
detection (HOFMAN et al. 2006). As breast tissue
ing. As in regular anatomical scans, too many
consists of tubular structures from the milk-
false negative detections are found, and current
glands to the nipple, tumor extensions may pref-
attention now focuses on dynamic contrast en-
erably follow these tubular pathways. In a projec-
hanced MRI. The rationale is the high vascularity
tion radiograph this leads to a spiculated or star-
of the neoplasm, leading to a faster uptake and
shaped pattern. The computer will inspect the
outwash over time of the contrast medium com-
contextual environment of each pixel (say 50 ×
pared to normal tissue. Current research focuses
50 pixels) on the presence of lines with an orien-
on the understanding of this vascular flow pattern
tation pointing towards the relevant pixel. In this
(e.g. by two-compartment modelling) and the
way a total of 2500 ‘votes’ are collected for each
optimal timing of the image sequence.
pixel. The pixels with a voting probability ex-
Polyp detection in virtual colonography:
9.6
Atlases
polyps are characterized by a mushroom-like extrusion of the colon wall, and can be detected by
The use of interactive 3D atlases on medical
their shape: they exhibit higher local 3D curva-
workstations is primarily focused on education
ture (‘Gaussian curvature’) properties. These
and surgery. As an example, K.-H. Höhne’s pio-
can be detected with methods from ‘differential
neering Voxel-Man series of atlases (HOFMAN et
geometry’ (the theory of shapes and how to
al. 2006) was initiated by the ‘visible human pro-
measure and characterize them), and highlighted
ject’. Atlases for brain surgery (e.g. the Cerefy
as, e.g. colored areas as attention foci for further
Brain atlas family; NOWINSKI et al. 2005) now
inspection.
become probabilistic, based on a large number of
Methods have been developed to carry out
patient studies.
an electronic cleansing of the colon wall when
contrast medium is still present. An interesting
current target is possible to reduce strongly the
radiation dose of the CT scan, and still be able to
detect the polyp structures, despite the deterioration of the detected colon wall structures. Clever
shape smoothing techniques and edge-preserving
smoothing of the colon surface have indeed enabled a substantial dose reduction.
Thorax CAD: here the focus is on the automated detection of nodules in the high resolution multi-slice CT (MSCT) data, on the detection
of pulmonary emboli, and of textural analysis by
Fig. 9.6. The famous Voxel-Man atlas explored
classification of pixels, e.g. for the quantification
many types of optimal educational visualization.
of the extent of sarcoidosis. See SLUIMER et al.
From HÖHNE (2004)
(2006) for a review.
Other CAD applications include calcium
scoring, used to detect and quantify calcified cor-
9.7
CAD/CAM Design
onary and aorta plaques, analysis of retinal fundus images for leaking blood vessels as an early
Workstations can also assist in the creation of
indicator for diabetes, and the inspection of skin
implants from the 3D scans of the patients. This
spots for melanoma (of particular attention in
is a highly active area in ENT, dental surgery,
Australia).
orthopedic surgery and cranio-maxillofacial sur-
gery. Many design techniques have been devel-
ment and register the DTI data with anatomical
oped to create the new shapes of the implants,
data, and find fiber crossings and endings auto-
e.g. by mirroring the healthy parts of the patient
matically. An interesting development is the pho-
of the opposite side of the body, 3D region grow-
torealistic rendering of the tiny bundle structures
ing of triangulated ‘finite element models’ in the
(with specularities and shadows), based on the
assigned space, etc. The ‘standard tesselation
physics of the rendering of hair.
language’ (STL) is a common format to describe surfaces for 3D milling equipment for rapid prototyping, such as stereolithography systems, plastic droplets ditherers, five-axes computerized milling machines, laser powder sintering systems, etc. Many dedicated rapid prototyping companies exist (e.g. Materialize Inc., see
also www.cc.utah.edu/~asn8200/rapid.html). In
the medical arena several large research institutes
are active in this area (Ceasar, Berlin; Co-Me,
Zürich).
Fig. 9.7. Muscle fibers tracked in a highresolution DTI MRI of a healthy mouse heart.
9.8 Diffusion Tensor Imaging


(DTI ) – Tractography
Lighting and shadowing of lines combined with
color coding of helix angle (h). From PEETERS et
al. (2006a)
Three-dimensional (3D) visualization of fiber
tracts in axonal bundles in the brain and muscle
fiber bundles in heart and skeletal muscles can
9.9
Registration
now be done interactively. The images are no
longer composed of scalar (single) values in the
Registration, or matching, is a classical technique
voxels, but a complete diffusion tensor (a 3 × 3
in image analysis (HAJNAL et al. 2001). It is em-
symmetric matrix) is measured in each voxel.
ployed to register anatomical to anatomical, or
The three so-called eigenvectors can be
anatomical to functional data, in any dimension.
calculated with methods from linear algebra;
Examples are MRI-CT, PET-CT, etc. The con-
they span the ellipsoid of the Brownian motion
struction of a PET and a CT gantry in a single
that the water molecules make at the location of
the voxels due to thermal diffusion. Complex
mathematical methods are being investigated to
group the fibers in meaningful bundles, to seg-
system effectively solves the registration problem
for this modality.
The matching can be global (only translation, orientation and zooming of the image as a
whole) or local (with local deformation, also
of the imaging beam) can be enhanced by such
called warping). Registration can be done by
techniques as (adaptive) histogram equalization.
finding correspondence between (automatically
detected) landmarks, or on the intensity landscape
9.11 Quantitative Image
itself (e.g. by correlation). There is always an
Analysis
entity (a so-called functional) that has to be
minimized for the best match: e.g. the mean
squared distance between the landmarks, a
Pierson correlation coefficient, or others. In particular, for multi-modality matching, the mutual
information (MI) has been found to be an effective minimizer. As an example, in MRI bone
voxels are black and in CT white; they show as a
This is the fastest growing application area of
medical workstations. The number of applications
is vast, every major vendor has research activities
in this area, and many research institutes are active. To quote from the scope of ‘Medical Image
Analysis’, one of the most influential scientific
journals in the field:
“The journal is interested in approaches
cluster in the joint probability histogram of the
MR vs CT intensities. The MI is a measure of
entropy (disorder) of this histogram.
that utilize biomedical image datasets at all spatial scales, ranging from molecular/cellular imaging to tissue / organ imaging. While not limited to
9.10
RT Dose Planning
these alone, the typical biomedical image datasets
of interest include those acquired from: magnetic
The accuracy of radiotherapy dose calculations,
based on the attenuation values of the CT scan of
the patient, needs to be very high to prevent un-
resonance, ultrasound, computed tomography,
nuclear medicine, X-ray, optical and confocal
microscopy, video and range data images.
derexposure of the tumor and overexposure of the
healthy tissue. Typically the isodose surfaces are
calculated and viewed in 3D in the context of the
local anatomy. Increasingly the images made in
the linear accellerator with the electronic portal
imaging device (EPID) are used for precise localization of the beam and repeat positioning of
the patient, by precise registration techniques.
The low contrast images (due to the high voltage
The types of papers accepted include
those that cover the development and implementation of algorithms and strategies based on the
use of various models (geometrical, statistical,
physical, functional, etc.) to solve the following
types of problems, using biomedical image datasets:
Fig. 9.8. Multimodality MRI of atherosclerotic plaque in the human carotid artery: (w1) T1-weighted 2D
TSE, (w2) ECG-gated proton density-weighted TSE, (w3) T1-weighted 3D TFE, (w4) ECG-gated partial
T2-weighted TSE, (w5) ECG-gated T2-weighted TSE. Middle: Feature space for cluster analysis. Right:
classification result. From HOFMAN et al. (2006)
Representation of pictorial data, visualiza-
are MICCAI, CARS, IPMI, ISBI and SPIE MI. In
tion, feature extraction, segmentation, inter-study
the following some often-used techniques are
and inter-subject registration, longitudinal / tem-
shortly discussed. There are excellent tutorial
poral studies, image-guided surgery and interven-
books (MOLECULAR
tion, texture, shape and motion measurements,
YOO 2004) and review papers for the field.
VISUALIZATIONS
undated;
spectral analysis, digital anatomical atlases, sta-
Segmentation is a basic necessity for
tistical shape analysis, computational anatomy
many subsequent viewing or analysis applica-
(modelling normal anatomy and its variations),
tions. Mostly thresholding and 2D/3D region
computational physiology (modelling organs and
growing are applied, but these often do not give
living systems for image analysis, simulation and
the required result. Proper segmentation is noto-
training), virtual and augmented reality for thera-
riously difficult. There are dozens of techniques,
py planning and guidance, telemedicine with
such as model-based segmentation, methods
medical images, tele-presence in medicine, tele-
based on statistical shape variations (‘active
surgery and image-guided medical robots, etc.”
shape models’), clustering methods in a high-
See also the huge amount of toolkits for
dimensional feature space (e.g. for textures), his-
computer vision: http://www.cs.cmu.edu/~cil/v-
togram-based methods, physical models of con-
source.html. Important conferences in the field
tours (‘snakes’, level sets), region-growing
methods, graph partitioning methods, and multi-
e.g. MUSICA (‘Multi-Scale Image Contrast Am-
scale segmentation.
plification’, by Agfa), and the Swedish Con-
The current feeling is that fully automated
textVision (http://www.contextvision.se/). En-
segmentation is a long way off, and a mix should
hancement is often used to cancel the noise-
be made between some (limited, initial) user-
increasing effects of substantially lowering the X-
interaction (semi-automatic segmentation).
ray dose, such as in fluoroscopy and CT screen-
Feature detection is the finding of specif-
ing for virtual colonoscopy.
ic landmarks in the image, such as edges, corners,
Quantitative MRI is possible by calculat-
T-junctions, highest curvature points, etc. The
ing the real T1 and T2 figures from the T1 and T2
most often used mathematical technique is multi-
weighted acquisitions, using the Bloch equation
scale differential geometry (TER HAAR ROMENY
of MRI physics. Multi-modal MRI scans can be
2004). It is interesting that the early stages of the
exploited for tissue classification: when different
human visual perception system seem to employ
MRI techniques are applied to the same volume,
this strategy.
each voxel is measured with a different physical
Image enhancement is done by calculat-
property, and a feature space can be constructed
ing specific properties which then stand out rela-
with the physical units along the dimensional
tive to the (often noisy) background. Examples
axes: e.g. in the characterization of tissue types in
are the likeness of voxels to a cylindrical struc-
atherosclerotic lesions with T1, T2 and proton
ture by curvature relations (‘vesselness’), edge
density weighted acquisitions, fat pixels tend to
preserving smoothing, coherence enhancing,
cluster, as do blood voxels, muscle voxels, calci-
tensor voting, etc. Commercial applications are,
fied voxels, etc.
Pattern recognition techniques like neural net-
set of variable shapes and performing a ‘principal
works and Bayesian statistics may find the
component analysis’, a well known mathemati-
proper cluster boundaries.
cal technique. The first eigenmode gives the main
Shape can be measured with differential
variation, the second the one but largest, etc. Fit-
geometric properties, such as curvature, saddle
ting an atlas or model-based shape on a patient’s
points, etc. It is often applied when, e.g. in the
organ or segmented structure becomes by this
automated search for (almost) occluded lung ves-
means much more computationally efficient.
sels in pulmonary emboli, polyps on the colon
Temporal analysis is used for bolus track-
vessel wall, measuring the stenotic index, spicu-
ing (time-density quantification), functional maps
lated lesions in mammography, etc. A popular
of local perfusion parameters (of heart and brain),
method is based on ‘active shape models’,
contrast-enhanced MRI of the breast, cardiac out-
where the shape variation is established as so-
put calculations by measuring the volume of the
called shape eigenmodes by analyzing a large
left ventricle over time, multiple sclerosis lesion
growth / shrinkage over time, regional cardiac
ity. The source images are from two-photon
wall thickness variations and local stress/strain
microscopy, where the collagen is specifically
calculations, and in fluoroscopy, e.g. the freezing
colored with a collagen specific molecular imag-
of the stent in the video by cancellation of the
ing marker.
motion of the coronary vessel.
Another example is the detailed study of
the micro-vascular structure in the goat heart
9.12
Workstations
for
Life
Sciences
from ultra-thin slices of a cryogenic microtome
(degree of branching, vessel diameter, diffusion
and perfusion distances, etc.). Typical resolution
In life sciences research a huge variety of (high
dimensional) images is generated, with many new

types of microscopy

is 25–50 micron in all directions, with datasets of
20003.

(confocal , two-photon ,
cryogenic transmission electron microscopy,
etc.) and dedicated (bio-) medical small animal
scanners (micro-CT, mini PET, mouse-MRI,
etc.). The research on molecular imaging and
molecular medicine is still primarily done in
small animal models.
Fig. 9.10. A 3D visualization of a microtome stack
(40×40×40 m) of the micro-vasculature of a goat
heart (VAN BAVEL et al. 2006) [BENNINK 2006]
This research arena will benefit greatly in
Fig. 9.9a,b. Two-photon florescence microscopy
of collagen fibers of tissue-engineered heartvalve tissue. a Result of structure preserving de-
the near future from the spectacular developments
in the diagnostic image analysis and visualization
workstations.
noising. From DANIELS et al. 2006
There is great need for quantitative image
analysis. An example is, e.g. the measurement of
quantitative
parameters
that
determine
9.13 Computer-Aided
Surgery (CAS)
the
strength of newly engineered heart valve tissue of
the patient’s own cell line, such as collagen fiber
thickness, local orientation variation and tortuos-
In the world of CAS some very advanced simulation and training systems (KISMET, Voxel-Man)
have been created. Especially in dental implants,
craniofacial surgery and laparoscopic surgery
there are many and highly advanced systems today. Surgical navigation workstations are routinely displaying the combination of the anatomy and
the position and orientation of the instruments in
the operating theatre.
Fig. 9.12a–c. Abdominal aorta aneurysm: a color
coding of displacement (mm); b Von Mises strain;
c Von Mises stress (kPa). From
DE
PUTTER et al.
(2005)
9.14
New Developments
The visual perception of depth (when viewing
3D) data is helped enormously if the viewer can
Fig. 9.11. Virtual laparoscopy trainer (Origin: For-
move the data himself. There are many depth
schungszentrum Karlsruhe KISMET)
cues (stereo, depth from motion, depth from per-
An interesting development is the use of
complex fluid dynamics modelling, which enables the prediction of rupture chances in abdominal aorta surgery, and selecting optimal
therapeutic procedures with bypass surgery in the
lower aorta.
In neurosurgery workstations can be employed in the calculation of an optimal (safest)
insert path for electrodes for deep brain stimulation (DBS), based on a minimal costs path avoiding blood vessels and ventricles, and starting in a
gyrus. Workstations assist in inter-operative visualization by warping the pre-operative imagery to
the real situation in the patient during the operation, by intra-operative MRI, or ultrasound.
spective), but depth from motion is the strongest.
That is why maximum intensity projections
(MIP) are preferably viewed dynamically. By
self-tracking also the muscle’s proprioceptors are
giving feedback to the brain, adding to the visual
sensation. The combination with human’s superb
eye-hand coordination has led to the concept of
the Dextroscope (www.dextroscope.com), where
a (computer generated) view or object can be
manipulated under a half-transparent mirror,
through which the viewer sees the display. Displays can also be equipped with haptic (tactile)
feedback systems, which are now commercially
available.
References
Fig. 9.13. Stereo viewing and manipulation with
haptic feedback
Super-large screens, and touch screens are
becoming popular; a new trend is the multi-touch
screen
(http://cs.nyu.edu/~jhan/ftirtouch/
with
movie), where multiple positions to interact simultaneously make more complex transformations
possible, such as zooming, multiple simultaneous
objects interactions, etc.
9.15
Outlook
We have actually just started with exploiting the
huge power these super assistants can add, in any
of the fields discussed above – hardware, software and integration. Image processing plays an
essential role, be it for visualization, segmentation, computer-aided detection, navigation, registration, or quantitative analysis. There will be an
ever greater need for clever and robust algorithms: it is the conviction of the author that the
study of human brain mechanism for the inspiration for such algorithms has a bright future to
come (TER HAAR ROMENY 2004). The radiologists will benefit from these super-assistants, and
finally: the patient has the best benefit of all.
Barco (2007) White Papers Barco (screens, DICOM):
http://www.barco.com/medical/en/downloads/whit
epapers.asp?dltype=15
Bovik AC (ed) (2000) Handbook of image and video processing (communications, networking and multimedia). Academic Press
Burroni M et al. (2004) Melanoma computer-aided diagnosis: reliability and feasibility study. Clin Cancer
Res 10:1881–1886
Daniels F, ter Haar Romeny BM, Rubbens MP, van Assen
HC (2006) Quantification of collagen orientation
in 3D engineered tissue. In: Ibrahim F (ed) Proc
Int Conf on Biomedical Engineering BioMed
2006, Kuala Lumpur, Malaysia, pp 344–348
de Putter S, Breeuwer M, Kosea U, Laffarguec F, Rouet J,
Hoogeveen R, van den Bosch H, Buthe J, van de
Vosse F, Gerritsen FA (2005) Automatic determination of the dynamic geometry of abdominal aortic aneurysm from MR with application to wall
stress simulations. Proc CARS, International Congress Series 1281, pp 339– 344
DICOM (2007a) standard: http://medical.nema.org/
DICOM (2007 b) conformance statement (example):
https://www.merge.com/RESOURCES/pdf/dcs/dc
s_efilm21.pdf
Doi K (2006) Diagnostic imaging over the last 50 years:
research and development in medical imaging science and technology. Phys Med Biol 51:R5–R27
Eizo
(undated)
White
Papers
Eizo
(screens):
http://www.eizo.com/support/wp/index.asp
Gilbert FJ, Lemke H (eds) (2005) Computer-aided diagnosis. Special issue of British Journal of Radiology,
vol. 78, British Institute of Radiology
GPGPU (undated) General-purpose computation using
graphics hardware, http://www.gpgpu.org/
Hajnal JV, Hill DLG, Hawkes DJ (eds) (2001) Medical
image registration. CRC
Health Level 7 (undated) Health Level 7,
http://www.hl7.org/. See also regional sites:
(Australia http://www.hl7.org.au/,
UK http://www.hl7.org.uk/,
Canada http://www.cihi.ca/hl7, etc.)
Hofman JMA, Branderhorst WJ, ten Eikelder HMM, Cappendijk VC, Heeneman S, Kooi ME, Hilbers PAJ,
ter Haar Romeny BM (2006) Quantification of
atherosclerotic plaque components using in-vivo
MRI and supervised classifiers”, Magn Reson Med
55:790–799
Höhne K-H (2004) VOXEL-MAN 3D-Navigator” (CDROM). Springer, Berlin Heidelberg New York
Karssemeijer N (1995) Detection of stellate distortions in
mammograms using scale-space operators. Proc.
Information Processing in Medical Imaging, pp
335–346
Kaufman, Müller K (2005) Overview of volume rendering.
The visualization handbook. Elsevier
Molecular visualizations (undated) Molecular visualizations: http://molvis.sdsc.edu/visres/index.html
Nowinski W, Thirunavuukarasuu S, Benabid G (2005) The
Cerefy clinical brain atlas. Enhanced edition with
surgical planning and intraoperative support (CDROM). Thieme
Peeters THJM, Vilanova A, Strijkers GJ, ter Haar Romeny
RBM (2006b) Visualization of the fibrous structure of the heart. Proc. 11th workshop on Vision,
Modelling and Visualization. Aachen, Germany
Peeters THJM, Vilanova A, ter Haar Romeny RBM
(2006a) Visualization of DTI fibers using hairrendering techniques. In: Lelieveldt BPF,
Haverkort B, de Laat CTAM, Heijnsdijk JWJ (eds)
Proc ASCI 2006. Lommel, Belgium, pp 66–73
Peters JF, Grigorescu SE, Truyen R, Gerritsen FA, de Vries
AH, van Gelder RE, Rogella P (2005) Robust
automasted polyp detection for low-dose and
normal-dise virtual colonoscopy. Proc CARS
2005, International Congress Series 1281, Berlin,
Germany, pp 1146–1150
RSNA 2006 Exhibitor list:
http://rsna2006.rsna.org/rsna2006/V2006/exhibitor
_list/home.cfm
Schaefer-Prokop CM, van Delden OM, Bouma H, Sonnemans JJ, Gerritsen FA, Lameris JS (2006) To assess the added value of a prototype computer-aided
detection (CAD) system for pulmonary embolism
(PE) in contrast-enhanced multi-detector computed
tomography (CT) images. Proc Eur Conf of Radiology, Vienna, Austria, EPOS Poster
Sereda P, Vilanova A, Gerritsen FA (2006) Automating
transfer function design for volume rendering using hierarchical clustering of material boundaries.
In: Sousa Santos B, Ertl T, Joy K (eds) Eurographics/IEEE VGTC Symposium on Visualization (EuroVis),Lisboa, Portugal, pp 243–250
Siggraph (undated) Ray tracing tutorial (SIGGRAPH):
http://www.siggraph.org/education/materials/Hype
rGraph/raytrace/rtrace0.htm
Sluimer IC, Schilham AMR, Prokop M, van Ginneken B
(2006) Computer analysis of computed tomography scans of the lung: a survey. IEEE Transactions
on Medical Imaging, vol. 25, pp. 385–405
Sonka M, Michael Fitzpatrick J (eds) (2000) Handbook of
medical imaging – vol. 2. Medical image processing and analysis. SPIE, Belligham, WA
ter Haar Romeny BM (2004) Front-end vision and multiscale image analysis. Springer, Berlin Heidelberg
New York
ter Haar Romeny BM (2004) Front-end vision and multiscale image analysis. Springer, Berlin Heidelberg
New York
van Bavel E, Bakker EN, Pistea A, Sorop O, Spaan JA
(2006) Mechanics of microvascular remodelling.
A review. Clin Hemorheol Microcirc 34(1/2):35–
41
Vilanova A, Gröller E (2003) Geometric modelling for
virtual colon unfolding. In: Brunnett, Harmann,
Müller, Lisen (eds) Geometric modeling for scientific visualization. Springer, Berlin Heidelberg
New York, pp 453–468
Vos FM, van Gelder RE, Serlie IWO, Florie J, Nio CY,
Glas AS, Post FH, Truyen R, Gerritsen FA, Stoker
J (2003) Three-dimensional display modes for CT
colonography: conventional 3D virtual colonoscopy versus unfolded cube projection. Radiology
228:878–885
Voxel Man surgical workstation: http://www.uke.unihamburg.de/medizinische-fakultaet/voxelman/index_ENG.php
Whitby J (2006) The DICOM Standard”, White paper Barco
Inc,
2006.
URL:
http://www.barco.com/barcoview/downloads/Whit
ePaper_DICOM.pdf
Yoo TS (2004) Insight into images: principles and practice
for segmentation, registration, and image analysis.
AK Peters
Download