EDUCATION EXHIBIT 637
CT Angiography of
Intracranial Aneu-
rysms: A Focus on
Postprocessing1
LEARNING
OBJECTIVES
FOR TEST 1
After reading this
article and taking
the test, the reader
will be able to:
� Describe the tech-
nique of CT angiog-
raphy performed for
detection of intracra-
nial aneurysms.
� List the technical
considerations in and
limitations of the
most common meth-
ods of image postpro-
cessing.
� Discuss when to
use CT angiography
as the sole diagnostic
test in a patient with
subarachnoid hemor-
rhage before therapy.
Bernd F. Tomandl, MD ● Niels C. Ko¨stner ● Miriam Schempershofe
Walter J. Huk, MD ● Christian Strauss, MD ● Lars Anker, MD ● Peter
Hastreiter, PhD
Computed tomographic (CT) angiography is a well-known tool for
detection of intracranial aneurysms and the planning of therapeutic
intervention. Despite a wealth of existing studies and an increase in
image quality due to use of multisection CT and increasingly sophisti-
cated postprocessing tools such as direct volume rendering, CT an-
giography has still not replaced digital subtraction angiography as the
standard of reference for detection of intracranial aneurysms. One rea-
son may be that CT angiography is still not a uniformly standardized
method, particularly with regard to image postprocessing. Several
methods for two- and three-dimensional visualization can be used:
multiplanar reformation, maximum intensity projection, shaded sur-
face display, and direct volume rendering. Pitfalls of CT angiography
include lack of visibility of small arteries, difficulty differentiating the
infundibular dilatation at the origin of an artery from an aneurysm, the
kissing vessel artifact, demonstration of venous structures that can
simulate aneurysms, inability to identify thrombosis and calcification
on three-dimensional images, and beam hardening artifacts produced
by aneurysm clips. Finally, an algorithm for the safe and useful applica-
tion of CT angiography in patients with subarachnoid hemorrhage has
been developed, which takes into account the varying quality of equip-
ment and software at different imaging centers.
©RSNA, 2004
Abbreviations: DSA� digital subtraction angiography, dVR� direct volume rendering, FOV� field of view, ICA� internal carotid artery,
MCA�middle cerebral artery, MIP�maximum intensity projection, MPR�multiplanar reformation, PICA� posterior inferior cerebellar artery,
SAH� subarachnoid hemorrhage, SSD� shaded surface display, 3D� three-dimensional, 2D� two-dimensional
Index terms: Aneurysm, CT, 17.12116, 17.73 ● Aneurysm, intracranial, 17.73 ● Computed tomography (CT), angiography, 17.12116 ● Computed
tomography (CT), image processing, 17.12117
RadioGraphics 2004; 24:637–655 ● Published online 10.1148/rg.243035126
1From the Department of Neurosurgery, University of Erlangen-Nuremberg, Schwabachanlage 6, D-91054 Erlangen, Germany. Presented as an edu-
cation exhibit at the 2002 RSNA scientific assembly. Received May 5, 2003; revision requested June 13 and received July 30; accepted July 31. All au-
thors have no financial relationships to disclose. Address correspondence to B.F.T. (e-mail: tomandl@neuroradiologie-erlangen.de).
©RSNA, 2004
R
a
d
io
G
ra
p
h
ic
s
CME FEATURE
See accompanying
test at http://
www.rsna.org
/education
/rg_cme.html
Introduction
Sudden onset of vigorous headache typically is
the leading symptom in patients with subarach-
noid hemorrhage (SAH) caused by the rupture of
an intracranial aneurysm. Computed tomography
(CT) is the first step in the examination of these
patients. Once SAH is confirmed, it is paramount
to detect the source of bleeding in order to initiate
therapy. Digital subtraction angiography (DSA) is
still the most sensitive tool for the detection of
intracranial aneurysms. The selective intraarterial
injection of contrast medium ensures optimal en-
hancement of the intracranial arteries with supe-
rior resolution compared with that of CT or mag-
netic resonance (MR) angiography. However,
DSA has the disadvantage of being an invasive
study. The risk of acquiring a permanent neuro-
logic deficit with cerebral angiography in patients
with SAH is below 0.1% (1). Despite this rela-
tively low risk, a noninvasive method yielding
three-dimensional (3D) information for the plan-
ning of therapeutic intervention is desirable.
The reported sensitivity of CT angiography lies
in the range of 80%–97% (2–8) depending on the
size and location of an aneurysm (3). In all of
these studies, some kind of 3D visualization was
used to analyze the CT angiography data. Little is
known about the influence of postprocessing
methods like maximum intensity projection
(MIP), shaded surface display (SSD), and direct
volume rendering (dVR) on the detection rate of
intracranial aneurysms. However, it can be as-
sumed that the same CT angiography data may
lead to varying detection rates when different vi-
sualization strategies, computer platforms, and
graphics hardware are used (9–12).
In the first section of this article, technical as-
pects of CT angiography with a focus on data ac-
quisition are discussed. In the second section,
different methods for the postprocessing of CT
data are presented, including the analysis of
source images and the methods currently avail-
able for two-dimensional (2D) and 3D postpro-
cessing, such as high-resolution dVR. Then, typi-
cal pitfalls encountered while working with CT
angiography data are demonstrated. Finally, we
propose a reasonable paradigm for the use of CT
angiography in patients with SAH, taking into
account that this method is still not a standard-
ized procedure.
Technique of Intra-
cranial CT Angiography
CT angiography can be defined as a fast thin-sec-
tion volumetric spiral (helical) CT examination
performed with a time-optimized bolus of con-
trast medium in order to enhance the cerebral
arteries (13). In order to visualize the intracranial
arteries, the examination includes the region from
the first vertebral body up to the vertex. It is im-
portant to include the atlas in the study to ensure
incorporation of the posterior inferior cerebellar
artery (PICA), which has an extracranial origin
from the vertebral arteries in about 18% of cases
(14).
On our four-row multisection scanner (Soma-
tom 4 Volume Zoom; Siemens Medical Solu-
tions, Erlangen, Germany), we used the following
parameters: 120 kVp, 200 mAs, collimation of
4� 1 mm, table feed of 2.7 mm per rotation, and
rotation time of 0.5 seconds. Image reconstruc-
tion parameters were as follows: section thickness
of 1.25 mm, overlapping steps of 0.5 mm, and
field of view (FOV) of 120 mm2. The applied
narrow FOV of 120 mm2 leads to an excellent
in-plane resolution (0.23� 0.23 mm2) and repro-
duces all relevant information (Fig 1). In addi-
tion, lateral parts of the skull are already elimi-
nated, which simplifies the postprocessing of
source data. It is possible to perform reconstruc-
tions in steps of 0.23 mm to produce isotropic
data (15), thus yielding voxels of equal extent in
all three dimensions. In our experience, this does
not noticeably increase image quality while dou-
bling the number of source images, thus leading
to an extension of time spent on postprocessing
source data.
For enhancement of intracranial arteries, 100
mL of contrast medium (Ultravist 300; Schering,
Berlin, Germany) was injected intravenously at a
flow rate of 4 mL/sec by using a power injector
(EnVision CT injector; Medrad, Indianola, Pa).
A bolus tracking method (16) was used routinely
to achieve optimal synchronization of contrast
medium flow and scanning. Once the injection is
started, the bolus tracking software measures at-
tenuation values within one internal carotid artery
(ICA), and the spiral scan is automatically started
as soon as a threshold of 100 HU is exceeded.
If bolus tracking is not available, the test bolus
method should be applied to calibrate timing of
the data acquisition (17): Ten seconds after bolus
injection of 20 mL of contrast medium, a dy-
namic single-axial-section study (one scan every
638 May-June 2004 RG f Volume 24 ● Number 3
R
a
d
io
G
ra
p
h
ic
s
2 seconds) at the level of the first cervical verte-
bral body is started until the contrast material
appears as hyperattenuating spots in the ICAs. By
using this technique, the time interval between
bolus administration and the beginning of data
acquisition can also be determined individually.
Analysis of CT Angiograms
The examples shown in this article were created
with the regular software of the workstation sup-
plied with a Somatom Volume Zoom CT scanner
(Syngo Wizard version VA 40C; Siemens Medical
Solutions). The dVR images were created on a
separate workstation (Syngo Leonardo 2002B;
Siemens Medical Solutions).
Prior to any kind of postprocessing, such as 3D
visualization, a detailed review of the source im-
ages that are the basis of CT angiography is man-
datory (18). These source images contain the en-
tire information that is available from the data.
Even the most sophisticated methods for 3D im-
aging will lead to a distinct loss of data and, thus,
potentially important information. Partial throm-
bosis or calcification of an aneurysm will be
missed if the source images are not reviewed in a
meticulous way. The interactive analysis of the
source images should be done on a workstation
rather than by looking at hard copies in order to
develop a better perception of the course and the
relationships of the intracranial arteries of inter-
est. A wide window setting is necessary to enable
differentiation between arteries filled with con-
trast medium, bone, and calcifications (Fig 2).
Figure 1. Effect of FOV on
image quality of 3D imaging
with SSD in a patient with two
aneurysms (arrows in b–d) at
the bifurcation of the left
middle cerebral artery (MCA).
(a) CT image shows the areas
covered by three different val-
ues for FOV: 200, 120, and 60
mm2. (b) SSD image obtained
by reconstructing the data with
an FOV of 200 mm2. Arteries
appear blurred. (c) SSD image
obtained by reconstructing the
data with an FOV of 120 mm2.
Vascular anatomy is shown
more clearly than in b. This
FOV contains all relevant intra-
cranial arteries from which an-
eurysms usually originate while
providing good in-plane resolu-
tion. Thus, we always use this
FOV for detection of intracra-
nial aneurysms. (d) SSD image
obtained by reconstructing the
data with an FOV of 60 mm2.
There is even better demonstra-
tion of vascular anatomy than
in c. It is sometimes useful to
perform a second reconstruc-
tion with a narrow FOV such as
this when CT angiography is
used for therapy planning and
very detailed information is re-
quired.
RG f Volume 24 ● Number 3 Tomandl et al 639
R
a
d
io
G
ra
p
h
ic
s
Figure 2. Importance of an adequate win-
dow setting to demonstrate the intracranial
arteries within the skull base. CT image ob-
tained with a window width of 500 HU and a
center of 150 HU. Both ICAs can be clearly
differentiated within the carotid canals (ar-
rows).
Figure 3. Preparation of the volume for
analysis. (a) Posterosuperior image shows
large veins (arrowheads), which are typically
also visible in CT angiography of intracranial
vessels and preclude an unobstructed view of
the circle of Willis and the basilar artery (ar-
row). (b) Left lateroposterior image shows
easy elimination of the most obscuring venous
structures by using a clip plane (dotted white
line) parallel to the clivus. (c) Posterosuperior
image obtained after application of the clip
plane (dotted white line) shows that the
basilar artery is demonstrated completely
(arrowheads).
640 May-June 2004 RG f Volume 24 ● Number 3
R
a
d
io
G
ra
p
h
ic
s
Many aneurysms can already be detected by
analyzing the source images. Smaller aneurysms
below 5 mm in diameter are often difficult to de-
tect on the basis of source images alone. There-
fore, several methods for 2D and 3D postprocess-
ing have been developed that allow more detailed
analysis and in addition an “angiographic” repre-
sentation of CT angiography data.
Postprocessing of
CT Angiography Data
The basic principle of 2D and 3D postprocessing
is to input cross-sectional images into a computer
and thereby to create a so-called volume. One can
imagine that this procedure is like putting back
together the pieces of a sliced potato. Once a vol-
ume is created, several methods for 2D and 3D
visualization exist (19). The easiest way to ana-
lyze a volumetric data set is multiplanar reforma-
tion (MPR), in which from a given angle of view a
plane is reconstructed in a defined depth of the
volume. This way it is possible to create coronal,
axial, sagittal, as well as any kind of oblique sec-
tions. The quality of the reconstructions depends
on the voxel size. With the use of isometric data
(ie, voxels have the same depth, length, and
height), all images are of the same quality as the
basic source images (20). In contrast to MIP and
the 3D methods discussed later, the recon-
structed planes contain all information that is
contained in the source images. Therefore, MPR
should always be the method of first choice for
the further examination of CT angiography data
(21).
To create useful “angiographic” representa-
tions from CT angiography data, it is always nec-
essary to eliminate disturbing structures from the
volume in order to ensure an unobstructed view
of the circle of Willis and its related arteries. For
this purpose, several graphical tools exist that vary
depending on the software of the workstation
used. To eliminate the straight sinus and other
veins that always prohibit an unobstructed view of
the basilar artery, a so-called clip plane can be
applied parallel to the clivus (Fig 3). This kind of
data manipulation is always necessary when MIP
or 3D visualization of CT angiography data is
performed by using one of the following methods.
Maximum Intensity Projection
The term maximum intensity projection (MIP)
means that from any given angle of view only the
brightest voxels of a volume are collected and
used to create an image (22). Therefore, MIP is
not a 3D method, as it creates 2D images in
which voxels from different locations within the
volume are collapsed into one plane. Thus, depth
information is lost and it is not possible to tell
whether a structure is located in the front or back
on the basis of a single MIP image. Because calci-
fications and bone are brighter than contrast ma-
terial–filled arteries, it is possible to differentiate
levels of attenuation (eg, to recognize a calcified
artery).
The use of MIP as a method to create CT an-
giograms is limited due to the fact that the skull
base has a much higher attenuation than the in-
tracranial arteries and therefore has to be elimi-
nated when MIP is used for image reconstruction
(23). When dealing with intracranial aneurysms,
it is often not possible to clearly depict the rela-
tions of the aneurysm to its adjacent arteries. Fur-
thermore, with the use of MIP, small aneurysms
will often be missed as they are eclipsed by the
signal of their parent vessels averaged into the
same 2D plane (Fig 4).
Figure 4. Influence of 3D visualization techniques on the detection of intracranial aneurysms. (a) MIP image (su-
perior view) shows the bifurcation of the left MCA (arrow). Owing to the lack of depth information, the image does
not allow visualization of two aneurysms at this site. (b, c) SSD (b) and dVR (c) images (superior views) of the bifur-
cation of the left MCA show the two aneurysms (arrows).
RG f Volume 24 ● Number 3 Tomandl et al 641
R
a
d
io
G
ra
p
h
ic
s
Figure 5. Analysis of CT angiography data with MPR and thin-section MIP. (a–c) Sagittal (a), coro-
nal (b), and axial (c) MPR images show a small aneurysm at the bifurcation of the right MCA (arrow).
Note the large intracerebral hematoma (arrowheads in a), which is usually not demonstrated on thresh-
old-based 3D images. (d–f) Sagittal (d), coronal (e), and axial (f) MIP images obtained with thin sec-
tions of 20 mm show the aneurysm more clearly (arrow) and show the intracerebral hematoma as well
(arrowheads in d).
642 May-June 2004 RG f Volume 24 ● Number 3
R
a
d
io
G
ra
p
h
ic
s
In contrast to the two methods for 3D visual-
ization described later, MIP is not threshold de-
pendent and therefore is relatively easy to use. It
has to be kept in mind that MIP uses only about
10% of the information contained in a given vol-
ume. In our experience, MIP is of minor use for
the creation of CT angiograms in order to search
for and analyze aneurysms but is often very help-
ful when used interactively on the workstation in
thin sections of about 10–20 mm in addition to
MPR (24) (Fig 5). In contrast to threshold-de-
pendent methods, smaller arteries are displayed
without user interaction (22).
Threshold-dependent
Methods for 3D Visualization
Shaded surface display (SSD) and direct volume
rendering (dVR) are more difficult to use than
MIP. They require the user to define thresholds
for the selection of voxels on the basis of their at-
tenuation (measured in Hounsfield units). For
SSD, typically upper and lower thresholds are
defined and from a chosen angle of view the first
layer of voxels with an attenuation within the de-
fined parameters is displayed. Therefore, the im-
ages show the surface of these structures and pro-
vide valuable information about the 3D shape of
an object (25). On the other hand, all structures
are shown in the same color and information
about the attenuation of a structure is lost com-
pletely. For example, it is not possible to see calci-
fications within an artery on SSD images. Since
MIP retains information about the attenuation of
objects yet does not allow the depth perception
provided by SSD, both methods may be used to
complement one another.
The definition of the thresholds is performed
interactively by the user and significantly influ-
ences the appearance of the vascular structures
(7) (Fig 6). Setting the lower threshold to a low
value (eg, 100 HU) will result in an image show-
ing many vascular structures, including the veins
and small arteries. When the lower threshold is
increased (eg, to 200 HU), structures of low at-
tenuation such as intracranial veins and small ar-
teries will disappear completely and the major
arteries will appear smaller. The “ideal” threshold
to depict intracranial arteries has to be found in-
teractively and depends on several parameters,
including the injection rate of the contrast me-
dium and cardiac output, both of which influence
the attenuation of the contrast material–filled vas-
culature (26).
Figure 6. Threshold-dependent 3D visualization with SSD. (a) Superoposterior view obtained
with a lower threshold of 100 HU shows smaller arteries like the left PICA (arrow) and venous
structures (arrowheads). (b) Superoposterior view obtained by increasing the lower threshold to
200 HU shows arteries that appear thinner compared with those in a and even demonstrate dis-
continuities (arrow). The venous structures are nearly eliminated (arrowheads), resulting in a less
complex image.
RG f Volume 24 ● Number 3 Tomandl et al 643
R
a
d
io
G
ra
p
h
ic
s
From the point of view of a computer scientist,
MIP and SSD are types of “volume rendering” in
which only one layer of voxels is used for visual-
ization. In the medical literature, the term volume
rendering is reserved for the technique described
next, in which all voxels of a volume are consid-
ered for visualization (27).
Direct volume rendering (dVR) is the most
sophisticated method for 3D visualization. The
basic principle is to select several groups of voxels
according to their attenuation in Hounsfield units
and to assign them a color and a so-called opacity
(28,29) (Fig 7). When dVR is used to create CT
angiograms, the voxels of high attenuation con-
taining information about bony structures are
selected separately from those voxels with an at-
tenuation bet