Conference PaperPDF Available

A Visualization System for the Clinical Evaluation of Cerebral Aneurysms from MRA Data

EUROGRAPHICS 2001 / Jonathan C. Roberts Short Presentations
A Visualization System for the Clinical Evaluation of
Cerebral Aneurysms from MRA Data
J. S. Perrin
, A. Lacey
, R. Laitt
, A. Jackson
and Nigel W. John
Manchester Visualization Centre,
Imaging Science and Bio-medical Engineering, University of Manchester.
Dept. of Neuroradiology, Manchester Royal Infirmary
This paper details a work-in-progress application under development as part of a clinical visualization project.
The software has been designed to meet the specific needs of interventional neuro-radiologists evaluating the suit-
ability of intracranial aneurysms for endovascular coiling and also when planning the procedure. Providing rapid
(real-time) interaction with high resolution iso-surfaces derived from Time-of Flight (ToF) Magnetic Resonance
Angiography (MRA) data will enable the clinician to quickly assess the ability of the aneurysm to accept a coil,
with greater reliability than exisiting, 2D film techniques. Simulating the interface of the C-arm angiography sys-
tem, used during the procedure, allows the clinician to evaluate various surgical strategies, potentially reducing
procedure times and therefore patient radiation dosage. The first release of the software is currently under-going
clinical evaluation.
1. Introduction
Brain haemorrhage is one of the commonest causes of sud-
den death in adolescents and young adults. In most cases it
results from the rupture of a weak spot on one of the arter-
ies that feed the brain with blood. These areas of weakness
expand due to the high pressure of arterial blood to form
small balloon like protuberances known as aneurysms. The
wall of the aneurysm is very thin, weak and prone to spon-
taneous rupture. Rupture of the wall causes sudden loss of
consciousness and in 40% of people death occurs within an
hour. In the remaining 60% emergency treatment is required
to stop subsequent re-bleeds which are far more common
immediately following the initial episode.
Currently, one of the commonest forms of treatment in-
volves packing the aneurysm with a small platinum coil.
This is introduced into the body by a catheter inserted into
the femoral artery in the groin and fed up into the brain and
eventually into the aneurysm itself. The tiny flexible plat-
inum coil is pushed through this tube into the aneurysm, the
coil retains a memory of its original shape and expands to
fill the aneurysm, though it can take several coils to fill a
large aneurysm. The platinum promotes clotting and even-
tual healing of the aneurysm without the need for invasive
brain surgery.
One of the major technical difficulties in endovascular
coiling is the accurate identification of the shape, size and
origin of the aneurysm and in particular the relative propor-
tions of the aneurysm neck to the aneurysm itself, (figure
Figure 1: Aneurysms form as swelling on the artery wall
due to the high blood pressure in the vessel and a localised
weakness the wall. The aneurysm in a) has a well defined
neckthat is suitable for coiling. However in b) the aneurysms
has a wider neck and a coil may not stay in place
Imaging techniques have become a vital component in
this pre-operative stage, assisting in the assessment of the
shape, size and location of the aneurysm as well as its po-
sition relative to surrounding vascular structures. However,
much of the surgical planning relies on two-dimensional
° The Eurographics Association 2001.
Perrin, Lacey et al / CAMRAS
information from sources such as X-ray images, digital
subtraction angiography (DSA), multiplanar reformatting
(MPR) or maximum intensity projection (MIP) of contrast
enhanced Computer Tomography (CT) or magnetic reso-
nance angiography (MRA) data. The clinician is often left
with uncertainty in their assessment due to ambiguities re-
sulting from the projection of a complex 3D environment
onto 2D. Further, during the coiling procedure the clinician
relies on angiography to monitor the introduction and posi-
tioning of the coil. The clinician must identify at least two
view angles, one showing the neck of the aneurysm and a
perpendicular view to check alignment of the catheter. Ob-
taining these views can take many attempts which expose
patient to additional radiation doses.
This project has developed a visualization application that
aims to provide renderings which simplify and accelerate
the process of assessing the shape, size and position of the
aneurysm and for selecting the optimal aneurysm views for
the coiling procedure. Two Magnetic Resonance Imaging
(MRI) acquisition methods have been considered, Time of
Flight (ToF) and Black Blood (BB). ToF produces a sig-
nal proportional to the blood flow through the vessels, con-
versely, in BB sequences flow produces a signal void. Both
methods have their advantages and disadvantages however,
as many aneurysms appear near bone and bone also presents
a signal void in BB sequences, ToF is the preferred method
in this work
. Figures 2 and 3 show example images from
ToF and BB sequences respectively.
Figure 2: Time of Flight MRI image which shows the signal
fall off in the aneurysm cavity due to the turbulent flow of
blood and the jet entering the aneurysm
The work in this paper outlines the current capabilities of
the software and how we intend to complete development.
The software is currently undergoing the first stage of eval-
uation by our clinical partners using PC hardware with an
Figure 3: BB MRI image, both the flowing blood and bone
are shown as signal voids
Intel PIII 600Mhz, 256Mb RAM and an nVidia Geforce 256
graphics card.
2. Previous Work
Several studies have investigated methods for extracting vas-
cular structure from ToF sequences
2, 3, 4
. These techniques
attempt to segment the data using either a statistical or struc-
tural model of the data. Although attractive rendered images
often result from such techniques there is no attempt to as-
sess the validity of the results. In any model based method
it is important to know to what degree the prior expecta-
tion of the model is dictating the final result. What needs
to be known is how the different pre-processing techniques
effect the clinical choices and ultimate outcome of the pro-
cedure. Until the affects of the segmentation techniques are
understood we are reluctant to make use of algorithms which
utilise as muchprior information. The metricperformance of
these algorithms is also poor, taking many hours to segment
typical datasets on standard PC hardware.
An automatic optimal view selection method has been
, this used an adaptive thresholding method
segment the vascular structure from ToF. A skeletonization
method was then employed to create a path from the artery
into the aneurysm. Radial image maps were created at ap-
proximately sixty points along the path, each contains the
distance from the point to the vascular wall. These maps
were then used to determine the position and orientation
of the aneurysm neck. The results presented appeared good
although there was no evaluation of the performance of
the technique in clinical situations. Further, given the ready
availability of accelerated graphics hardware for PC plat-
forms it is now possible to present high quality 3D surface
° The Eurographics Association 2001.
Perrin, Lacey et al / CAMRAS
and volume renderings in real-time at minimal cost. Together
with the experience of the clinician whose knowledge of
the vascular structure and familiarity of the problem enables
them to interpret the data even though it may be incomplete
leads us to conclude that optimal view selection is best done
Currently in our partners clinical environment, MRA data
is analysed on a Philips EasyVision system which provides
good 2D image manipulation but has poor 3D performance.
Only a small volume of data can be rendered as a surface
which can take several minutes to prepare. The renderings
are of low resolution, with basic user interaction available at
low frame-rates (typically <1fps). Clinical staff often spend
in excess of half an hour studying a single case using this
system and then only in conjunction with angiography data.
We have set out tothe provide an application that will take
the user through from MRA acquisition to data preparation
and then visualization. Thesoftware will allow them to make
concise judgements on the structure of the the aneurysm and
then to plan the surgical procedure. Finally we provide them
with 3D visualsto aid inthe procedure itself.The application
makes use of the hardware currently available to provide an
interactive environment that givesto user as near to real-time
feed back as possible.
3. Endovascular Surgical Planning Tool
The development environment for the application needed to
fullfil the following requirements:
to prototype the application in a short space of time
to provide a professional look and feel
to be able to make rapid changes as feedback from the
clinicians was obtained
to be flexible and extendable
AVS/Express was chosen over alternatives such as the
open source Visualization Tool-Kit (VTK) for several rea-
sons. Although VTK has an excellent array of functional-
ity the Manchester Visualization Centre (MVC) has a lot
of experience with AVS/Express; it hosts the International
AVS Centre. The combination of the visual programming
environment and the V description language, which defines
AVS/Express applications, meant that rapid changes could
be made to the software. AVS/Express applications are also
trivially ported to other platforms. In addition Hardware vol-
ume rendering support is also being evaluated and it is ex-
pected that AVS/Express will provide support for these cards
as and when then appear.
Since the software is being developed for clinicians who
may not have a high degree of computer literacy and also
have limited time, the UI has been designed to be as intuitive
as possible. This means we have tried following established
practices and procedures. The software has been developed
with constant user input so that ideas can be quickly inte-
grated into the system and make them aware of what the
system is capable.
The system has been integrated with the MRI scanner
which exports data in the DICOM medical image format.
Data can be pushed directly onto the PC via the local net-
work within a few minutes of the scan being taken. The DI-
COM file is parsed for pertinent information so that a simple
database can be created that allows the clinician to be pre-
sented with a clear indication of the datasets present on the
Figure 4: Endovascular Surgical Planning tool showing
data that has just been selected from the systems database
and creation of an initial isosurface
The ESP tool presents the user with a large main view
to show the 3D visualization. Smaller 2D views provide the
standard perpendicular slices through the volume, (figure 4).
The user interface controls have been kept to a minimum
to make best use of the screen space and let the clinician
focus on the actual data. For each step the user selects the
appropriate UI from a menu. Less frequently used controls
have been moved to pop-up windows, though the number of
these is kept to a minimum.
Once a scan has been acquired it is pushed to the system.
The clinician can then select the volume to be read in from a
list of patients with the date and size of their scan.
The ESP will initialise and attempt to window the data
(clamping) using values that may be found in the DICOM
header and set other parameters to sensible defaults. The
clinician can then move back and forth through the slices,
and pan and zoom to the region of interest, which is a stan-
dard practice for analysing MRA. The slice views can be
redirected to the main view to make this task easier. Tools
are provided to take distance and area measurements from
these images.
Once the area containing the aneurysm has been identified
a crop volume can be selected by marking the area in the
orthographic views. Currently, only rectangular region can
be selected but more sophisticated methods will be added to
° The Eurographics Association 2001.
Perrin, Lacey et al / CAMRAS
aid in removing objects that may obscure the aneurysm. An
arbitrary number of crops can be generated.
Figure 5: The data is cropped to the region contaning the
aneurysm and isosurfaced using the histogram tool. An ob-
ject removal method has also been employed
To visualise the aneurysm in 3D an isosurface can be cre-
ated from each crop volume. An interactive histogram is
available and a visual indication of the isosurface value on
the orthographic views aids in selecting the correct value.
The colour and opacity of each isosurface can be selected by
the user.
The choice of isosurface is extremely important, low val-
ues are required to obtain as much vascular structure as pos-
sible but the noise in the ToF can cause the creation of a
large number of artifacts that obscure the aneurysm, see fig-
ure 9. Re-sampling the crop volume to a lower resolution is
an immediate help though there is the obvious loss of detail.
The ability to use lower resolution data also means larger
volumes can be rendered directly and adds some scalabil-
ity to the software. Additional noise reduction methods are
discussed in section 4.1.
Higher isosurface values can show the regions of fast
blood flow. Where these blood jets occur and how they en-
ter the aneurysm are of special importance as they can cause
compacting of the coils once in place. Visualization of the
blood flow is discussed further in section 4.3.
Though isosurfacing is used as the primary visualization
method, due to its speed and ease of use, volume rendering
of selected crop volumescan also be performed. The transfer
function is described using the colour-map editor, (figure 6).
The performance of the volume rendering can be adjusted to
suit the hardware, using fat rays to render at a lower resolu-
tion and a bounding box for interaction. The nVidia Geforce
chip-sets, however, have excellent texture map performance
that allows interactive frame-rates ( >10fps) to be obtained
using a back to front composite texture map volume render-
ing method. The data must however be converted from the
usual 12 or 16 bits to 8bits; the affect of this and the quality
of the volume rendering is to be assessed. In addition, the
MIP render method is provided, mainly as a comparison to
the composite method, as it has been a standard 3D visual-
ization method for MRI.
Figure 6: AVS/Express’ internal volume rendering methods
are used with a simple to use colour-map editor to describe
an appropriate transfer function
Once the clinician has identified the aneurysm and cre-
ated a visualization which clearly demonstrates the orienta-
tion and structure of the aneurysm neck and blood flow into
the region, planning of the working views can then proceed.
The angiography machine consists of an X-ray machine
mounted on a swivelling C-arm, (figure 7). The orientation
of the X-ray source can be positioned in single degree steps
set by dials on the machine’s control panel. The table may
also be moved in three directions.
There are three major arteries entering the cranium, how-
ever the contrast medium will only appear on part of the vas-
cular structure. Therefore, the clinician creates a crop vol-
ume containing the vascular structure that will be visible
during angiography. The ESP simulates the movement of the
C-arm about the patient’s head. To achieve this a common
centre of rotation must be decided upon that can be found
in the ESP and at the start of the surgical procedure. To co-
register the position of the machine with that of the scanned
cranium three intersecting planes are defined in the ESP us-
ing standard anatomical features as references. Oncea work-
ing centre of rotation has been defined the ESP presents a
° The Eurographics Association 2001.
Perrin, Lacey et al / CAMRAS
X-Ray Source
X-Ray Detector
Ceiling Mount
Figure 7: General arrangement of the C-arm, the table can
be moved in any direction
simulation of the C-arm controls to allow the clinician to
find the required views using the full rendering of the vascu-
lar structure to obtain clear and unobstructed views, (figure
8). The application visually warns the user, by changing the
background colour, if the view will position the C-arm in
an unattainable orientation i.e., coincides with the patient or
the table. The patient can then be positioned in the machine
according to the same anatomical references as used in the
ESP, and the C-arm orientated to the position obtained from
the simulation. An approximation to the projected angiogra-
phy image is performed using and inverted MIP. This pro-
vides a rough structural approximation to the image which
would be viewed by the clinician. A better simulation of the
angiogram images is to be added to the application in the
near future, using volume rendering.
Figure 8: With the reference planes defining the centre of
rotation the operator can find the working views using the
simulated C-arm controls
Images of the main view can be saved out and a VRML
model of the whole scene also generated. This allows remote
consultation with other clinicians and provides a 3D guide to
the clinician during the coiling procedure, this could be dis-
played on a laptop in the operating theatre when logistics
make viewing from the ESP system impractical. Each ses-
sion of work performed in the ESP may also be saved to file
so that it can be referred to at a later date.
4. Current and Further Work
4.1. Filter Methods
As was mentioned earlier, we have investigated segmenta-
tion algorithms and concluded that, as yet, the performance
of such techniques remains to be proven in a clinical context
(see section 4.2). However, given the poor signal to noise ra-
tio of ToF images we are investigating the performance of
image based noise filtering techniques in order to reduce ar-
tifact clutter in the final rendering. Without any attempt to
remove these artifacts the clinician is forced to reduce the
iso-surface value until only the strongest signals remain, of-
ten losing many important vessels.
We have considered several image and volume filtering
methods both 2D image based (Gaussian, linear sequential,
median and tangential smoothing) and 3D volume based im-
plementations of the median and tangential smoothing algo-
rithms. All of these techniques have some effect on the data
as well as the noise so we are currently evaluating these al-
gorithms in terms of their noise removal and data modifica-
tion behaviour. To achieve this we are measuring changes in
cross-sectional area and circumference of known vessels at
a given iso-surface value, as well as the number of noise ar-
tifacts removed. Preliminary indications are that the 3D ver-
sion of the tangentialsmoothing algorithm has least effect on
the vessel data, whilst reducing noise artifacts to an accept-
able level. The tangential smoothing algorithm is designed
to be edge preserving, smoothing along the edge (or plane
in 3D) and not isotropically as many smoothing algorithms.
Figure 9(a) and (b) show the before and after renderings at a
fixed iso-surface for the 3D tangential smoothing.
(a) (b)
Figure 9: (a) Isosurface of cranial vascular structure from
raw ToF, (b) Isosurface after application of 3D Tangential
The filtering techniques described above havenot yet been
° The Eurographics Association 2001.
Perrin, Lacey et al / CAMRAS
added to the ESP system. However a method for filtering out
isosurface objects based on size (number of triangles) has
been added and is available to the clinicians. This performs
a recursive search through the arrays of vertices and trian-
gles building up a list of topologically separate objects. The
N largest objects in the isosurface can then be displayed.
This feature has proved very helpful, though practically it
can only be run on small isosurfaces. A more sophisticated
method of selecting the objects to be displayed will be de-
4.2. Vascular Atlasing and Segmentation Algorithms
The best renderings will undoubtedly be achieved when only
the vessel information is used. This would require prior seg-
mentation and as we have stated the performance of such
algorithms is not known. However, we have devised a tech-
nique to assess one aspect of the performance of these tech-
niques in an automated atlasing system. This system is built
on the existing 3D wireframe model matching software,
which is part of our image analysis libraries called TINA
and is used to match a known wireframe model to data ex-
tracted from a stereo view (two camera) of an object
. By
buildinga generic 3D model ofthe vascularstructure we will
be able to compare segmentation algorithms in terms of their
ability to extract data suitable for the model matching pro-
cess. Further, by labelling the regions with suitable markers
it will be possible to automatically identify anatomical re-
gions within the vessel structure.
4.3. Flow Indicators
As mentioned above the flow direction and speed of the
blood is of interest as it has a direct affects on the place-
ment of the coils within the aneurysm cavity. Secondary
aneurysms can form on the back of the original aneurysm,
often at the region under the most pressure from the blood
inflow. Therefore it is of primary importance that this region
is protected by thecoil. Using ToF data the flow speed can be
visualised with a volume rendering, (figure 10). By reducing
the opacity of the slow flowing blood the observer can iden-
tify the faster flowing regions which likely act as the inflow
jet to the aneurysm, and thus aim to protect the opposing
wall. The current tool has this volume rendering capability.
Further, we intend to improve the visualisation of flow in-
let by computing the minimum path derivative. Using the
flow relationship in the ToF images we can compute the
local voxel-to-voxel derivatives, and construct streamlines
through the minimum energy paths, approximating the most
probable flow directions. These can then be presented to the
clinician as cues to the major flow directions.
5. Current Status
The software is currently undergoinginitial evaluationby the
end users at Manchester Royal Infirmary, Dept. of Neuro-
Figure 10: Volume Rendering can provide a direct visual-
ization of the flow intensities as they enter the aneurysm
radiology. This involves a retrospective case study focusing
on previous cases where the angiography and current MRA
techniques were unable to resolve the structure, size or shape
of the aneurysm. The study is also monitoring changes in ra-
diation dosage new patients are subjected to, so that a com-
parison may be made when the system becomes fully opera-
Early indications are very promising, the speed and easy
of use of the system allows the clinicians to focus on the
6. Conclusions
We have been able to provide a system that conforms to clin-
ician’s standard working practices but integrates new and
faster methods for examining and analysing MRA data than
is currently available. Additionally we have enabled them
to correlate the views and understanding of the aneurysm
structure obtained in the application with the angiography
equipment used during the coiling procedure. The clinician
will have a clearer idea of how the coils should be placed to
reduce the chance of aneurysm re-growth. Other benefits are
a reduction in time in assessing each case and for the proce-
dure. In turn this means a reduction exposure to radiation for
the patient.
AVS/Express has so far proved adequate to the task, the
first prototype being produced in four weeks (half of this be-
ing development of the DICOM image reader). The software
has been ported to Windows NT/2000 and SGI Irix. It has
also been adapted for use in the AVS/Express Multi-pipe edi-
tion which allows AVS/Express to run in virtual reality en-
vironments, such as caves, on multiple graphics pipes
, and
is provided as part of the demo suite provided by AVS. Stan-
dard, readily available PC hardware has been able to provide
° The Eurographics Association 2001.
Perrin, Lacey et al / CAMRAS
a system capable of providing the performance need for in-
teractive and real-time manipulation of the data.
Recently, hardware ray-cast volume rendering
has be-
come available, in particular the VolumePro
card from Real
Time Visualization
. The first cards were considered at the
start of the project but were found to have limitations. These
would however be overcome in the second generation cards
that will hopefully be available before the end of the project.
Real-time performance provides the possibility of volume
rendering becoming the primary visualization method rather
than isosurfacing
VolumePro and Real Time Visualization are registered
trademarks of TeraRecon.
This project is funded by the The Sir Jules Thorn Charitable
1. W. M. Adams, R. D. Laitt, J. Thorne and A. Jackson
“MRA Visualization of Cerebral Aneurysms”, Medica
Mundi 1 (vol. 43), pp. 2–9 (March 1999). 2
2. D. L. Wilson and J. A. Nobel An Adaptive Segmen-
tation Algorithm for Time-of-Flight MRA Data”, IEEE
Trans. on Medical Imaging 10 (vol. 18), pp. 938–945
(October 1999). 2
3. P. Hall, M. Ngan and P. Andreae “Reconstruction of
Vascular Networks Using Three-Dimensional Models”,
IEEE Trans. on Medical Imaging 6 (vol. 16), pp 919-
929 (December 1997). 2
4. L. M. Lorigo, O. Faugeras, W. E. L Grimson, R.
Keriven, R. Kikinis and C-F Westin “Co-dimension 2
Geodesic Active Contours for MRA Segmentation”,
Proceedings of IPMI ’99, Lecture Notes in Computer
Science 1613 pp 126-139, 1999. 2
5. D. L. Wilson, J. A. Nobel, D. Royston and J. V. Byrne
Automatically Finding Optimal Working Projections
for the Endovascular Coiling of Cranial Aneurysms”,
Proc. Medical Image Computing Computer-Assisted
Intervention, Lecture Notes in Computer Science 1496,
pp. 814–821 (1998). 2
6. P. G. Lever, G. W. Leaver, I Curington, J. S. Perrin,
A. Dodd, Nigel W. John, W. T. Hewitt “Design Issues
in the AVS/Express Multi-Pipe Edition”, IEEE Visu-
alization 2000 Conference Works In Progress pp. 1–9
(September 2000). 6
7. H. Pfister Architectures for Real Time Volume Render-
ing”, Journal of Future Generation Computer Systems
1 (vol. 15), pp. 1–9 (Febuary 1999). 7
8. 6
9. A. J. Lacey, N. A. Thacker, P. Courtney and S. B Pol-
lard, “TINA 2001: Closed Loop 3D Model Matching”,
BMVC 2001 submitted.
° The Eurographics Association 2001.
... According to Schreyer and Warfield, most clinical applications involving 3D visualization can take advantage of this technique [177]. As an example, Figure 2(a) shows such a surface generated by endovascular surgical planning software implemented at the Manchester Visualization Centre (MVC) [150]. Here the high contrast in the (a) (b) Figure 2: Visualization from MRA data of a patient suffering from a brain haemorrhage (from [89]). ...
Conference Paper
Full-text available
The medical domain offers many excellent opportunities for the application of computer graphics, visualization, and virtual environments, offering the potential to help improve healthcare and bring benefits to patients. This report provides a comprehensive overview of the state-of-the-art in this exciting field. It has been written from the perspective of both computer scientists and practicing clinicians and documents past and current successes together with the challenges that lie ahead. The report begins with a description of the commonly used imaging modalities and then details the software algorithms and hardware that allows visualization of and interaction with this data. Example applications from research projects and commercially available products are listed, including educational tools; diagnostic aids; virtual endoscopy; planning aids; guidance aids; skills training; computer augmented reality; and robotics. The final section of the report summarises the current issues and looks ahead to future developments.
Conference Paper
Full-text available
Automatic and semi-automatic magnetic resonance angiog- raphy (MRA) segmentation techniques can potentially save radiologists large amounts of time required for manual segmentation and can fa- cilitate further data analysis. Our proposed MRA segmentation method uses a mathematical modeling technique which is well-suited to the com- plicated curve-like structure of blood vessels. Specically , we dene the segmentation task as an energy minimization over all 3D curves then use a level set method to search for a solution. Our approach is an extension of previous level set segmentation techniques to higher co-dimension.
Over the last decade, volume rendering has become an invaluable visualization technique for a wide variety of applications. This paper reviews three special-purpose architectures for interactive volume rendering: texture mapping, VIRIM, and VolumePro. Commercial implementations of these architectures are available or underway. The discussion of each architecture will focus on the algorithm, system architecture, memory system, and volume rendering performance.
Conference Paper
The endovascular coil embolisation of intracranial saccular aneurysms requires a set of specific X-ray images with which to view the aneurysm during coiling. These two-dimensional images, known as working projections should be optimal for measuring the aneurysm sac diameter, inserting the first coil, and checking coil overhang into the surrounding vessels. At present the gantry angle that produces these images is found by the radiologist by trial and error. In this paper we present a method for automatically finding the angles that will produce the desired X-ray projections. Our method consists of four steps: (1) segmenting the vasculature from three-dimensional angiographic data; (2) locating the aneurysm neck; (3) labelling the aneurysm sac; and (4) determining the optimal angle for viewing the aneurysm during coiling. We discuss details of the algorithm steps and present the results of the algorithm applied to one synthetic and two pathological examples.
Reconstructing vasculature in three dimensions is a challenging problem. Early approaches concentrated on coronary vasculature in X-ray images, recent work uses magnetic resonance imagery of cerebral vasculature. In both cases a priori information has been used, and often the way this is represented has proven limiting to the scope of applications supported. For example, a particular representation may be useful only for X-ray images. This paper addresses two issues: 1) representing a collection of vasculature and 2) the reconstruction of individual vasculature from images. Our representation learns the variations in branching structures and vessel shapes that occur between individuals. It supports a vascular catalogue containing three-dimensional (3-D) anatomical models. The representation is task independent; here we use it to reconstruct vasculature from images. Our algorithm has four features to which we draw attention: 1) it is not premised wholly upon X-ray images (though that is our focus here); 2) it produces several feasible solutions rather than one; 3) it can generalize from the catalogue to reconstruct instances not yet learned; 4) it exhibits polynomial time complexity, reasonable memory consumption, and is reliable. Both our representation and reconstruction algorithm are new and useful approaches. In support of these claims, we present results gathered from X-rays of both simulated and real vasculature.
A three-dimensional (3-D) representation of cerebral vessel morphology is essential for neuroradiologists treating cerebral aneurysms. However, current imaging techniques cannot provide such a representation. Slices of MR angiography (MRA) data can only give two-dimensional (2-D) descriptions and ambiguities of aneurysm position and size arising in X-ray projection images can often be intractable. To overcome these problems, we have established a new automatic statistically based algorithm for extracting the 3-D vessel information from time-of-flight (TOF) MRA data. We introduce distributions for the data, motivated by a physical model of blood flow, that are used in a modified version of the expectation maximization (EM) algorithm. The estimated model parameters are then used to classify statistically the voxels into vessel or other brain tissue classes. The algorithm is adaptive because the model fitting is performed recursively so that classifications are made on local subvolumes of data. We present results from applying our algorithm to several real data sets that contain both artery and aneurysm structures of various sizes.
MRA Visualization of Cerebral Aneurysms
  • W M Adams
  • R D Laitt
  • J Thorne
  • A Jackson
W. M. Adams, R. D. Laitt, J. Thorne and A. Jackson "MRA Visualization of Cerebral Aneurysms", Medica Mundi 1 (vol. 43), pp. 2-9 (March 1999). 2
  • A J Lacey
  • N A Thacker
  • P Courtney
  • S Pollard
A. J. Lacey, N. A. Thacker, P. Courtney and S. B Pollard, " TINA 2001: Closed Loop 3D Model Matching ", BMVC 2001 submitted.
An Adaptive Segmentation Algorithm for Time-of-Flight MRA Data
  • D L Wilson
  • J A Nobel
D. L. Wilson and J. A. Nobel "An Adaptive Segmentation Algorithm for Time-of-Flight MRA Data", IEEE Trans. on Medical Imaging 10 (vol. 18), pp. 938-945 (October 1999). 2
Geodesic Active Contours for MRA Segmentation
Geodesic Active Contours for MRA Segmentation", Proceedings of IPMI '99, Lecture Notes in Computer Science 1613 pp 126-139, 1999. 2