Original article

Accuracy of CT angiography in the assessment of the circle

of Willis: comparison of volume-rendered images and digital
subtraction angiography

Ari Han1, Dae Young Yoon1, Suk Ki Chang1, Kyoung Ja Lim1,2, Byung-Moon Cho3,
Yoon Cheol Shin4, Sam Soo Kim2 and Keon Ha Kim5
1
Department of Radiology, Kangdong Seong-Sim Hospital, Hallym University College of Medicine, Seoul; 2Department of Radiology,
Kangwon National University College of Medicine, Chuncheon, Kangwon-do; 3Department of Neurosurgery, Kangdong Seong-Sim
Hospital, Hallym University College of Medicine, Seoul; 4Department of Thoracic Surgery, Kangdong Seong-Sim Hospital, Hallym
University College of Medicine, Seoul; 5Department of Radiology, Samsung Medical Center, Sungkyunkwan University College of
Medicine, Seoul, Republic of Korea
Correspondence to: Dae Young Yoon. Email: evee0914@chollian.net

Abstract
Background: Computed tomography angiography (CTA) is increasingly used for non-invasive imaging of the
cerebrovascular diseases.
Purpose: To evaluate the accuracy of CTA in the assessment of the variation of the segment calibers of the
circle of Willis.
Material and Methods: One hundred and 17 patients with acute SAH (51 men and 66 women, mean age
50.9 years) who underwent CTA using a 16 detector-row CT scanner and DSA were evaluated retrospectively.
The CTA and DSA studies were performed within 24 h after the onset of symptoms and within 24 h of each
other. A total of 819 arterial segments (A-comA, right and left A1 segment, right and left P-com A, and right and
left P1 segment) of the circle of Willis were determined to be aplastic (grade 1), hypoplastic (grade 2), or
normal-sized (grade 3) by blinded observers evaluating CTA volume-rendered images. The CTA results were
then compared with findings on the corresponding DSA images (reference standard).
Results: The overall agreement between CTA and DSA was 92.4%. We had 62 (7.6%) cases of disagreement
(58 cases of under-estimation and four cases of over-estimation by CTA) between tow modalities. The
sensitivity and specificity of CTA in the detection of aplastic and normal-sized segments were more than 90%.
In contrast, subgroup analysis of the hypoplastic segments showed a sensitivity of 52.6% and a specificity
of 98.2%.
Conclusion: CTA is highly accurate in the assessment of anatomical variations of the circle of Willis; however,
its sensitivity is limited in depicting hypoplastic segments.

Keywords: Vascular, brain, CT angiography, digital subtraction angiography, circle of Willis, anatomic variation

Submitted February 21, 2011; accepted for publication May 28, 2011

The circle of Willis is a polygonal structure composed of
the A1 segments of anterior cerebral arteries (ACAs),
anterior communicating artery (A-comA), P1 segments of
posterior cerebral arteries (PCAs), posterior communicating
arteries (P-comAs) (1). The circle is an important potential
pathway that maintains adequate blood flow in case of
various steno-occlusive disease (2), but variations (hypopla-
sia or aplasia) of the arterial segments are common (3, 4).
In addition, the circle of Willis is the site of the majority
of intracerebral aneurysms. The patency and size of the
arteries in the circle of Willis is important in the planning
of surgical clipping or endovascular coiling in certain
cases of aneurysm.
Digital subtraction angiography (DSA) is considered the
standard of reference for assessment of the anatomy of
the circle of Willis (5). This procedure, however, is invasive
and is not always available in critically-ill patients and
not without its inherent risk (6, 7). With advance of

Acta Radiologica 2011; 52: 889– 893. DOI: 10.1258/ar.2011.110223

Downloaded from acr.sagepub.com at UNIV OF PENNSYLVANIA on June 20, 2015

890 A Han et al.
.................................................................................................................................................
multidetector-row computed tomography (CT) scanners,
computed tomography angiography (CTA) is increasingly
used for non-invasive imaging of the cerebrovascular
diseases, including aneurysms (8), vasospasm (9), and
steno-occlusive lesions (10).
However, to our knowledge, only one report (11) reported
the accuracy of CTA in the analysis of the circle of Willis. The
aim of the study was to retrospectively evaluate the accuracy
of CTA in assessing the variation of the segment calibers of
the circle of Willis compared with DSA in patients with
acute subarachnoid hemorrhage (SAH).
Material and Methods
Patients
Our institutional review board approved this study, and
informed consent was waived as the patients’ data were
evaluated retrospectively and anonymously.
From March 2004 through August 2010, 124 consecutive
patients, with acute SAH confirmed by unenhanced CT or
lumbar puncture, who underwent both CTA and DSA
within 24 h after the onset of symptoms and within 24 h
of each other, were included in this study. Of them, five
patients with arterial occlusions or severe stenoses in the
internal carotid artery (ICA) or proximal segment of
ACA/middle cerebral artery (MCA)/PCA were excluded
because intracranial hemodynamics might be altered in
these patients. One patient who had undergone prior surgi-
cal clipping or endovascular coiling for treatment of aneur-
ysm was excluded as well because it could not completely
be ruled out that cerebral blood flow dynamics might
be affected in such a patient. One patient was excluded
because of pronounced motion artifacts interfering with
diagnostic decision-making.
Thus, our final study population consisted of 117 patients;
this group included 51 men and 66 women (mean age 50.9
years + 12.4, range 23 – 84 years).
All patients underwent CTA before DSA, with the mean
interval between the two examinations being 6.2 hours + 6.3.
CTA protocol
All CTA examinations were performed with a 16 detector-
row CT scanner (MX8000 Infinite Detector Technology;
Philips, Haifa, Israel) with the following parameters:
1-mm section thickness; 0.5-second gantry rotation time;
pitch of 0.35; 512  512 matrix; 20 – 22-cm field of view;
and 120 kV and 200 –280 mA. The effective dose of CTA
was calculated to be 14.1 mSv. The scanning range was
planned in a caudocranial direction from the level of the
foramen magnum through a point 1 cm above the level
of the lateral ventricles, including whole circle of the
Willis (mean coverage 80 mm, range 75– 85 mm). For
optimal intraluminal contrast enhancement, the delay time
between start of contrast material administration and start
of scanning was determined for each patient individually
by using a bolus-tracking technique. A total of 60– 100 mL
of iohexol (Omnipaque 300; GE Healthcare, Princeton, NJ,
USA) was administered through an 18- or 20-gauge needle
Downloaded from acr.sagepub.com at UNIV OF PENNSYLVANIA on June 20, 2015
positioned in a peripheral vein. The contrast material was
administered with a power injector at a rate of 3 – 4 mL/s.
The volumetric data so obtained were transferred to
a workstation with commercially available software
(RAPIDIA 3D; Infinitt, Seoul, Korea) for further processing.
Transverse sections were reconstructed with a section width
of 0.5 mm. CTA images were processed from the obtained
source images by using two different methods: (a)
volume-rendered technique (VRT) algorithm; and (b) VRT
images after automatic segmentation of a precontrast scan
data-set (i.e. any overlapping bony structures, calcification,
and surgical materials). All acquired images were trans-
ferred to a picture archiving and communication system
workstation (Pi-ViewStar, Infinitt) for analysis.
DSA
All DSA was performed transfemorally with a DSA unit
(Integris Allura; Philips Medical Systems, Best, The
Netherlands) with an image intensifier matrix of 1024 
1024 pixels. DSA was performed with bilateral selective
ICA injections and either unilateral or bilateral vertebral
artery injections. Anteroposterior, lateral, and if necessary
oblique view(s) of each vessel were obtained by the injection
of 6– 9 mL of iodixanol (Visipaque 320; GE Healthcare,
Princeton, NJ, USA). Additional ICA angiograms were
also obtained with compression of the contralateral carotid
artery in cases in which the A-comA was not visible on
routine DSA images.
Image analysis
All CTA and DSA images were independently evaluated on
the workstations by two radiologists, who had 14 (DYY)
and 7 (SKC) years of experience in cerebral CTA and DSA.
Each examination was allocated a study number that was
known only to the study coordinator (ARH). Both readers
were blinded to the assessments of the other modality or
of the other investigator. The CTA and DSA images were
reviewed separately: DSA images were given in random
order to readers, 8 weeks after each reader completed the
analysis of CTA images. After independent assessment,
the CTA and DSA studies where there was interpretation
disagreement between two readers were reviewed jointly
for a final consensus.
For analysis purposes, the arteries of the circle of Willis
were separately evaluated as seven anatomic segments on
CTA and DSA images: A-comA, right and left A1 segments
of ACAs, right and left P-comAs, and right and left P1
segments of PCAs. All measures were performed on the
workstation with an electronic caliper after appropriate
magnification. Each segment was assigned one of three cat-
egories: grade 1, aplasia (no flow); grade 2, hypoplasia (
50% luminal narrowing); and grade 3, normal size.
Reference diameter of hypoplasia was the diameter of non-
diseased ipsilateral proximal P2 segment for P-comA and
P1 and mean diameter of A2 segments for A-comA and
A1. There was no case with arterial segments duplication in
the circle of Willis.
 Accuracy of CT angiography in the assessment of the circle of Willis 891
.................................................................................................................................................

The CTA and DSA examinations were compared for each hypoplasias are common and most commonly involve the
arterial segment using Spearman correlation coefficient. P-comAs (12, 13).
Sensitivity, specificity, positive and negative predictive In the present study, with the use of VRT algorithm, the
values, and accuracy of CTA for determination of aplasia, overall agreement between CTA and DSA was 92.4% in the
hypoplasia or normal size were calculated using DSA as assessment of anatomical variations of the circle of Willis.
standard of reference. However, the agreement rate was lower for the P-com As
(84.2%); 37 (59.7%) of the 62 segments of disagreement
between the two modalities were the P-com As. Our results
Results showed that CTA against DSA resulted in 58 segments of
under-estimation and four segments of over-estimation in
In 117 patients, a total of 819 arterial segments were ana-
the assessment of the circle of Willis. The under-estimation
lyzed on both CTA and DSA images. Overall, DSA depicted
of arterial segments with CTA may be explained by the
551 normal-sized, 78 hypoplastic, and 190 aplastic arterial
lower image resolution of CTA than DSA. In addition,
segments. The most frequently seen variant was aplasia of
some cases of under-estimation were likely to be due to col-
the P-comA, which was present in 124 (53.0%) of 234 seg-
lateral vessels opacified from the systemic injection of con-
ments. At DSA, a total of 106 aneurysms was present in
trast at CTA. Although DSA is the gold standard for the
98 of the 117 patients involved in the study; six patients
assessment of collateral flow, the forced arterial injection
had two aneurysms and one patient had three aneurysms.
of contrast and local increases in arterial pressure may
Of these aneurysms, 62 (58.5%) were located at the circle
cause changes in arterial diameter (14). Four segments (2
of Willis (35 A-comA and 27 P-comA aneurysms).
P-comAs and 2 P1 segments) with over-estimation at CTA
The number of arterial segments determined to be aplas-
in this series may be explained by the superimposition of
tic, hypoplastic, or normal-sized on each of the imaging
basal veins or adjacent arteries (the anterior choroidal
techniques is summarized in Table 1. The overall agreement
artery or the superior cerebellar artery), especially in the pos-
between CTA and DSA for all locations and types was
terior circulation.
92.4%, with a Spearman correlation coefficient of 0.9551.
Although our study demonstrated acceptable (. 90%)
In examining the data for P-comAs, we found poorer
agreement, sensitivity, and specificity values for CTA as
degree of agreement and correlation than for the other
compared with DSA, the sensitivity of CTA for depicting
segments. We had 62 (7.6%) of 819 segments of disagree-
hypoplastic segments was poor (52.6%).
ment between CTA and DSA. CTA showed a clear trend
Several previous studies (5, 11) have demonstrated that
to underestimate the arterial segments compared to DSA.
CTA is an accurate technique to assess a fetal origin of
The arterial segments were underestimated by one grade
the PCA (P-comA larger than P1 segment). However,
in 50 segments and by two grades in eight segments
we found in the literature only one article assessing the diag-
(Fig. 1). There were only four segments with over-estimation
nostic accuracy of CTA in the evaluation of the circle of
at CTA (all by one grade).
Willis. Skutta et al. (11) compared the combined both
Table 2 summarizes the sensitivity, specificity, positive
maximum-intensity projection (MIP) and source images of
and negative predictive values, and accuracy of CTA for
double-detector CTA with DSA in 112 patients with sus-
the detection of aplastic, hypoplastic, and normal-sized seg-
pected cerebrovascular disease. The authors focused on the
ments. The sensitivity and specificity of CTA in the detec-
steno-occlusive lesions, therefore, no comprehensive ana-
tion of normal-sized and aplastic segments were more
lyses of anatomic variants were made. In their study, of 703
than 90%. In contrast, subgroup analysis of the hypoplastic
arterial segments revealed by DSA, 527 were seen with
segments showed a sensitivity of 52.6% and a specificity
MIP (sensitivity 75%) and 584 were seen on source images
of 98.2%.
(sensitivity 83%); the specificity was, however, not reported.
Magnetic resonance angiography (MRA) is another
non-invasive technique that can demonstrate the configur-
Discussion ation of the circle of Willis (15 – 17). Previous studies (15,
It is well established that considerable variations exists in 18 –20) have demonstrated that the overall sensitivity of
the circle of Willis (3, 4). Segmental aplasias and MRA was acceptable (higher than 80%) in depicting the
Table 1 Arterial segments of the circle of Willis: assessment by CT angiography as compared with digital subtraction angiography in 117 patients
Aplasia Hypoplasia Normal size Arteries correctly Arteries Arteries
classified by underestimated overestimated
DSA CTA DSA CTA DSA CTA CTA (n) by CTA (n) by CTA (n) r
A-comA (n ¼ 117 ) 28 36 10 4 79 77 109 (93.2) 8 (6.8) 0 (0) 0.9503
A1 (n ¼ 234) 16 20 17 18 201 196 225 (96.2) 9 (3.8) 0 (0) 0.9329
P-comA (n ¼ 234) 124 154 34 13 76 67 197 (84.2) 35 (15.0) 2 (0.8) 0.9127
P1 (n ¼ 234) 22 23 17 19 195 192 226 (96.6) 6 (2.6) 2 (0.8) 0.9393
All segments (n¼ 819) 190 233 78 54 551 532 757 (92.4) 58 (7.1) 4 (0.5) 0.9551

Data are number of segments; data in parentheses are percentages

DSA ¼ digital subtraction angiography; CTA ¼ CT angiography; A-comA ¼ anterior communicating artery; A1 ¼ first segment of anterior cerebral artery (between
origin and the A-comA); P-comA ¼ posterior communicating artery; P1 ¼ first segment of posterior cerebral artery (between origin and the P-comA)

r ¼ Spearman correlation coefficient

Downloaded from acr.sagepub.com at UNIV OF PENNSYLVANIA on June 20, 2015

892 A Han et al.
.................................................................................................................................................

Fig. 1 A 39-year-old woman presented with SAH due to rupture of a left P-comA aneurysm (arrowhead in b and d). Anteroposterior view DSA images of the right
(a) and left (b) ICA and the left vertebral artery (c) reveal the complete circle of Willis. Note normal-sized right A1 segment (arrows in a and b). (d) Superior pro-
jection VRT CTA image shows the same configuration of the circle of Willis except A1 segment of the right ACA. The right A1 segment (arrows) was graded as
hypoplastic (underestimated by CTA)

presence or absence of arterial segments of the circle of

Willis. However, their specificity of MRA (63– 100%)
Table 2 Diagnostic performance of CT angiography compared with varied according to the MRA techniques and analysis
digital subtraction angiography in assessment of arterial segments of methods used. A recently published study (20) comparing
the circle of Willis
MRA (3D time-of-flight [TOF] and 2D phase contrast
Arterial Sensitivity Specificity PPV NPV Accuracy images) and DSA found that collateral flow measurements
segments (%) (%) (%) (%) (%)
via the anterior part of the circle of Willis yielded sensitivity
Aplasia 97.9 92.5 79.8 99.3 93.8 and specificity of 83% and 77%, respectively. However, the
Hypoplasia 52.6 98.2 75.9 95.1 93.9 sensitivity of MRA was very low (33%) for the P-comA.
Normal size 96.6 100 100 93.4 97.7
Although 3D TOF MRA is widely used in the diagnosis
PPV ¼ positive predictive value; NPV ¼ negative predictive value of intracranial aneurysms and cerebral steno-occlusive

Downloaded from acr.sagepub.com at UNIV OF PENNSYLVANIA on June 20, 2015

 Accuracy of CT angiography in the assessment of the circle of Willis
.................................................................................................................................................
diseases, the technique is not without limitations. Because
the sensitivity of MRA is dependent on the blood flow vel-
ocity, the technique may have difficulties in visualizing
small vessels in the circle of Willis with slow or turbulent
flows (3, 21). In addition, laminar flow-related spin dephas-
ing and partial volume averaging at the vessel wall may
result in a reduction of the vessel diameter, which lead to
under-estimation of the vessel on the source images and
MIP algorithm (19, 22).
In contrast, CTA is not dependent on the flow velocity,
thereby allowing accurate documentation of vessel diam-
eters. The disadvantages of CTA compared to MRA
include the lack of information regarding flow direction,
possible venous contamination, radiation exposure, and
administration of iodinated contrast material.
We acknowledge several limitations of our study. First,
there is still uncertainty about the definition of hypoplastic
vessels forming the circle of Willis. Several previous
workers have used different sets of criteria to define hypo-
plasia. They have stated that a vessel with a diameter of
smaller than 0.8 mm (3, 23) or 1 mm (24, 25) would be con-
sidered hypoplastic. In this study, however, accurate
measurement of vessel diameter was not possible on DSA
images because the magnification factor was not calculated.
Therefore, we compared diameters of the segments with
those of the distal arteries on both CTA and DSA, instead
of absolute values of the measurements. Second, another
possible problem in our study relates to the fact that we
used only VRT algorithms to assess the circle of Willis.
For accurate assessment of vascular status with CTA, it is
important to review 3D images as well as the axial source
images. Results of a previous CTA study (11) showed that
combining 3D (MIP) images and axial source data improved
the visibility of small vessels. Third, this study had only
patients with SAH. Therefore, vasospasm following SAH
may have hampered correct evaluation of the circle of
Willis in our patients. However, cerebral vasospasm rarely
occurs before the fourth day and reaches a peak around
the seventh day after onset of the SAH (26). In all our
patients, CTA and DSA were performed within 48 h after
onset of SAH and with the mean interval between the two
examinations being 6.2 h; therefore, we expect only a
minor effect on the accuracy.
In conclusion, the results of this study show that CTA is
highly accurate in the assessment of anatomical variations
of the circle of Willis. However, CTA has limited sensitivity
in depicting hypoplastic segments, although it remains
quite specific.
Conflict of interest: None.
REFERENCES
1 Osborn A. Diagnostic cerebral angiography. 2nd edn. Philadelphia, PA:
Lippincott, 1999
2 Schomer DF, Marks MP, Steinberg GK, et al. The anatomy of the
posterior communicating artery as a risk factor for ischemic cerebral
infarction. N Engl J Med 1994;330:1565 – 70
Downloaded from acr.sagepub.com at UNIV OF PENNSYLVANIA on June 20, 2015
893
3 Krabbe-Hartkamp MJ, van der Grond J, de Leeuw FE, et al. Circle of
Willis: morphologic variation on three-dimensional time-of-flight MR
angiograms. Radiology 1998;207:103 –11
4 Waaijer A, van Leeuwen MS, van der Worp HB, et al. Anatomic
variations in the circle of Willis in patients with symptomatic carotid
artery stenosis assessed with multidetector row CT angiography.
Cerebrovasc Dis 2007;23:267 – 74
5 Van der Lugt A, Buter TC, Govaere F, et al. Accuracy of CT angiography
in the assessment of a fetal origin of the posterior cerebral artery. Eur
Radiol 2004;14:1627 – 33
6 Pryor JC, Setton A, Nelson PK, et al. Complications of diagnostic cerebral
angiography and tips on avoidance. Neuroimaging Clin N Am
1996;6:751 –8
7 Cloft HJ, Joseph GJ, Dion JE. Risk of cerebral angiography in patients
with subarachnoid hemorrhage, cerebral aneurysm,, arteriovenous
malformation: a meta-analysis. Stroke 1999;30:317 –20
8 Yoon DY, Lim KJ, Choi CS, et al. Detection and characterization of
intracranial aneurysms with 16-channel multidetector row CT
angiography: a prospective comparison of volume-rendered images and
digital subtraction angiography. Am J Neuroradiol 2007;28:60 – 7
9 Yoon DY, Choi CS, Kim KH, et al. Multidetector-row CT angiography of
cerebral vasospasm after aneurysmal subarachnoid hemorrhage:
comparison of volume-rendered images and digital subtraction
angiography. Am J Neuroradiol 2006;27:370– 7
10 Hirai T, Korogi Y, Ono K, et al. Prospective evaluation for suspected
stenoocclusive disease of the intracranial artery: combined MR
angiography and CT angiography compared with digital subtraction
angiography. Am J Neuroradiol 2002;23:93– 101
11 Skutta B, Fürst G, Eilers J, et al. Intracranial stenoocclusive disease:
double-detector helical CT angiography versus digital subtraction
angiography. Am J Neuroradiol 1999;20:791– 9
12 Fawcett E, Blachford JV. The circle of willis: an examination of 700
specimens. J Anat Physiol 1905;40:63 –70
13 Riggs HE, Rupp C. Variation in form of circle of Willis. The relation of
the variations to collateral circulation: anatomic analysis. Arch Neurol
1963;8:24 –30
14 Nornes H, Sorteberg W, Nakstad P, et al. Haemodynamic aspects of
clinical cerebral angiography: concurrent two vessel monitoring
using transcranial Doppler ultrasound. Acta Neurochir (Wien)
1990;105:89 –97
15 Patrux B, Laissy JP, Jouini S, et al. Magnetic resonance angiography
(MRA) of the circle of Willis: a prospective comparison with
conventional angiography in 54 subjects. Neuroradiology 1994;36:193 –7
16 Katz DA, Marks MP, Napel SA, et al. Circle of Willis: evaluation with
spiral CT angiography, MR angiography, and conventional angiography.
Radiology 1995;195:445 –9
17 Edelman RR, Mattle HP, O’Reilly GV, et al. Magnetic resonance imaging
of flow dynamics in the circle of Willis. Stroke 1990;21:56 –65
18 Fürst G, Steinmetz H, Fischer H, et al. Selective MR angiography and
intracranial collateral blood flow. J Comput Assist Tomogr 1993;17:178 – 83
19 Stock KW, Wetzel S, Kirsch E, et al. Anatomic evaluation of the circle of
Willis: MR angiography versus intraarterial digital subtraction
angiography. Am J Neuroradiol 1996;17:1495 –9
20 Hendrikse J, Klijn CJ, van Huffelen AC, et al. Diagnosing cerebral
collateral flow patterns: accuracy of non-invasive testing. Cerebrovasc Dis
2008;25:430 –7
21 Fürst G, Hofer M, Sitzer M, et al. Factors influencing flow induced signal
loss in MR angiography: an in vitro study. J Comput Assist Tomogr
1995;19:692 –9
22 Urchuk SN, Plewes DB. Mechanisms of flow-induced signal loss in MR
angiography. J Magn Reson Imaging 1992;2:453 –62
23 El-Barhoun EN, Gledhill SR, Pitman AG. Circle of Willis artery
diameters on MR angiography: an Australian reference database. J Med
Imaging Radiat Oncol 2009;53:248 – 60
24 Eftekhar B, Dadmehr M, Ansari S, et al. Are the distributions of
variations of circle of Willis different in different populations? Results of
an anatomical study and review of literature. BMC Neurol 2006;6:22
25 Rhoton AL. The supratentorial arteries. Neurosurgery 2002;51:S53– S120
26 Kessell NF, Sasaki T, Colohan ART, et al. Cerebral vasospasm following
aneurysmal subarachnoid hemorrhage. Stroke 1985;16:562 –81