Transcript No Slide Title

Interactive Visualization of Time-Resolved Contrast-Enhanced
Magnetic Resonance Angiography (CE-MRA)
Ethan
Departments of Electrical
Introduction
Magnetic Resonance Angiography (MRA) has been limited by long scan times. While x-ray
fluoroscopy and MRA both work by injecting a vascular contrast agent and imaging its
passage over time, they have very different properties. X-ray fluoroscopy is capable of 2D
imaging at very high frame rates, while the clinically accepted technique for MRA (3DFT
acquisition) produces a single 3D image at a time optimized for viewing arteries. The high
frame rate in x-ray fluoroscopy provides added diagnostic confidence as the radiologist can
watch injected contrast flow from the arteries to the veins. Having only a single image limits
the effectiveness of MRA for complex flow patterns such as dissections or retrograde filling.
It also makes it critical to time the scan to get good arterial signal without venous
contamination, which is difficult for patients with pathologies that cause varied contrast
arrival times, as is the case with aneurysm or stenosis.
A 3D undersampled projection-reconstruction acquisition, VIPR, allows for large speedup
factors and makes possible time-resolved 3D MRA[1]. VIPR exams have high isotropic
spatial resolution and good temporal resolution over a large field-of-view. A 30-60 second
scan generates a dataset with spatial resolution of 256x256x256 or greater and 15-30 time
frames. At each point, the scan characterizes the concentration of the contrast agent in blood
or tissue. Time-resolved exams eliminate scan timing concerns, ease diagnosis for complex
flow patterns, and can produce useful images at several stages of arterial filling.
Unfortunately, the visualization tools available from MRI scanner manufacturers and
radiology workstation vendors are ill suited to dealing with these large time-resolved
datasets. The commercial visualization tools offer radiologists the capability of doing MultiPlanar Volume Reformats (MPVR) to analyze 3D volumes, but are designed to work with a
single dataset from a single scan. Additionally, the commercial tools run on expensive,
specialized workstations. These workstations are centrally located in radiology reading
rooms and their availability is limited.
There is a need for fast, simple visualization tools that are adept at working with multiple
datasets and operate on inexpensive desktop workstations running Linux/X or Microsoft
Windows. They must perform maximum intensity projections (MIPs) through a volume,
with arbitrary cut-planes and viewpoints. They must provide the ability to rapidly switch
between time frames (maintaining the same viewpoint and cut-planes) for looking at timevarying properties and to switch between entire datasets for comparing various acquisition
and reconstruction techniques under development. They must have the ability to render
multiple datasets simultaneously in different colors, with the capability of doing simple pervoxel arithmetic operations. Finally, they must be able to generate still images and movies.
We have developed an OpenGL-based application to satisfy these requirements. It is written
in C and uses GLUT for all user-interface interactions, so it is portable between Linux/X and
Microsoft Windows without any source code changes.
1 Fig. 1: Commercial tools for
2
visualizing radiology datasets
offer a great deal of flexibility,
but deal poorly with multiple
time frames and have a low
frame rate.
Fig. 2: Our tool is specifically
designed
for
interactive
visualization of 3D timeresolved MRA.
Materials and Methods
CE-MRA exams are usually interpreted using MIPs through a volume of interest. The MIP
operation enhances the visibility of high-contrast vessels over the background tissue. The
tool uses a four-pane user interface, with three small windows showing orthogonal “orthonavigation” slices (axial, coronal, and sagittal) to guide the user, and a single large window
showing the 3D MIP.
1
Brodsky
1
Engineering ,
and Walter F.
2,3
Block
2
Engineering ,
Biomedical
University of Wisconsin-Madison
and Medical
The MIP is constructed in video hardware using the GL_MAX blending operation. The
volume is represented to the video card as a collection of 2D textures of slices on three
orthogonal planes. The entire 3D volume is rendered using the slice set nearest orthogonal
to the viewpoint.[2] The 3D volume can also be represented using a single 3D texture, but
we have observed significantly decreased rendering performance using this technique.
3
Physics
3
4
The volume to be rendered is bounded using a set of user-controlled cut-planes, which are
specified graphically on one of the ortho-navigation windows. The viewpoint can be
locked to remain perpendicular to the cut-planes, or can be unlocked to allow viewing from
arbitrary angles. Voxels are stored as 8-bit intensity values and window leveling is
performed in hardware using OpenGL texture palettes.
The software is capable of displaying either a single dataset or displaying multiple datasets
simultaneously in different colors. Multiple colors can be used to show arteries in red and
venous vasculature in blue, or to show the false lumen of a dissected aorta in a different
color to ease assessing whether vessels branch from the true or false lumen. Rendering in
stereo for viewing with 3D glasses is also supported, easing interpretation of complex
structures.
Hardware and Performance
Fig. 3: MRA is commonly used to check for renal
artery stenosis. Examining the renal system is
difficult because the interval between arterial and
venous enhancement is very small, and the renal
veins often obscure visibility of the renal arteries.
Time-resolved MRA allows the renal arteries and
veins to be clearly distinguished.
Fig. 4: Vessels feeding the brain have complicated
three-dimensional geometry, with veins enhancing
soon after arteries. We plan to use time-resolved
MRA to ease diagnosis of Arterio-Venous
Malformations (AVMs), a condition where oxygenrich blood “short-circuits” directly from artery to
vein, bypassing brain tissue.
5
6
The tool can run on a single x86-based workstation with a consumer-level video card. Early
work was performed on a dual-processor P3-1000 workstation with 1 GB of memory and an
ASUS V7700 GeForce2 GTS video card with 64 MB of video memory. More recent work
has used a PNY Quadro4 750XGL video card with 128 MB of video memory. The original
software was developed for Linux, but has been ported to Windows with minimal
modifications.
A single 256x256x256 volume requires 48MB of memory, as it must be stored three times,
with slices along each axis. However, only one third of this data is used to render a frame,
since only a single set of slices is used at any one time. Swapping new texture sets into
video memory (necessary when rotating the viewpoint through certain angles) is a relatively
fast operation that takes less than ¼ second. With adequate video memory to store two full
texture sets, the delay could potentially be eliminated in most cases by anticipating texture
switches and prefetching the necessary set. Using a single 3D texture reduces memory
requirements, allowing additional datasets to be simultaneously loaded into memory, but at
the expense of rendering performance. The tool also supports 256x256x1024 full-body
angiograms generated by whole body acquisition techniques.
Performance depends on the extent of the volume rendered (determined by the cut-planes)
and the size of the displayed image (determined by the effective viewing distance). Both of
these are pixel fill rate limitations. Vertex transform rate requirements are determined by the
number of datasets rendered simultaneously and thus remain relatively constant.
On the GeForce2 consumer-level video card, the frame rate varies with the rendering
parameters. Rendering a full volume over the full viewport gives a worst-case frame rate of
3 fps. With thinner slabs rendered over the full viewport, the frame rate can be as high as 60
fps. Rendering without magnification leads to significant improvements in frame rate, with
10 fps for full-volume MIPs and 70 fps for thin slabs. Frame rates during typical use
generally range from 6-20 fps. The Quadro4 based card maintains more constant frame
rates of 12 fps, regardless of parameters.
Future Work and Conclusions
The tool is designed to be flexible and easily extensible. It is well suited to acceleration
using parallel-execution methods. 4D cluster visualization methods being developed in
conjunction with this project support distribution of a time sequence across nodes in a
cluster. Interactive view manipulations are performed in parallel on all nodes and reading
back the rendered results sequentially from each node produces a high frame-rate
animation. This approach will be especially useful as higher resolution datasets and more
complex rendering algorithms lead to increased memory requirements and reduced frame
rates on desktop workstations.
Figs. 5 and 6: In patients with aortic dissections, the lining of the aorta has split and blood flows between
the inner and outer layers of the lining. The vessel is split into a “true lumen” that has good blood flow, and
a “false lumen” with poor flow. It is important clinically to asses whether vessels originate from the true or
false lumen. In these coronal MIPs made from two patient scans, we show the true lumen (enhances
quickly) in white and the false lumen (enhances slowly) in green. An understanding of the precise
geometry of the vessels is important for planning surgery, which involves reconstruction of the damaged
sections of the vessel or the placement of an artificial stent.
Time-resolved imaging offers additional information that is very useful clinically. However,
without a tool designed to take advantage of the time-resolved data, radiologists often
examine only a single time frame to make diagnoses. We plan to assess the use of our tool
in interpreting challenging cases such as aortic stent follow-up with endoleak
characterization and the analysis of dissections and anomalous portal or pulmonary flow
patterns.
We have developed a tool that allows for interactive 3D visualization on inexpensive
desktop workstations. It compares favorably with tools available on commercial radiology
workstations, achieving similar or higher frame rates with similar image quality at a far
lower cost. It has proved useful for our research applications, and its ability to work with
time-resolved information has great clinical potential.
Acknowledgements and References
This work was funded by the Whitaker Foundation and NIH R01 EB002075.
[1] Barger AV, et al. Time-resolved contrast-enhanced imaging with isotropic resolution
and broad coverage using an undersampled 3D projection trajectory. Magnetic Resonance
in Medicine 48(2):297-305 (2002).
[2] Tom McReynolds. Advanced Graphics Programming Techniques Using OpenGL. ACM
SIGGRAPH 98 Course. [Online course notes available at sgi.com]