From: AAAI Technical Report SS-94-05. Compilation copyright © 1994, AAAI (www.aaai.org). All rights reserved.
3D Diagnostic Imaging of Blood Vessels using an X-Ray Rotational Angiographic System
Noboru Niki, Yoshiki Kawata, and Tatsuo Kumazald*
Department of Information Science, University of Tokushima, JAPAN
*Department of Radiology, Nippon Medical School, JAPAN
ABSTRACT:
In this paper, we describe a prototype
system that. reconstructs three-dimensional (3D) blood vessels
image from cone-beam projection images and analyzes the 3D
reconstructed image. Several 3D reconstruction methods utilize
biplane or stereo angiographic system and assume the cross
section shapes of blood vessels (e.g. ellipse). However,the
information provided by a few projections is insufficient for
reconstructing a precise 3Dimageof blood vessels. Furthermore,
the assumptionof cross sectional shapes is not usually valid for
abnormal blood vessels. In the present method using an X-ray
rotational angiographic system, digital angiogramsare collected
in a short period time and a 3D image of blood vessels is
approximately reconstructed using a short scan cone-beam
reconstruction algorithm. The analysis methodis based on the
structure description of 3D blood vessels image. The procedure
approximatesthe centerline of 3Dblood vessels imageusing a 3D
thinning algorithm and extracts the geometrical features of the 3D
line pattern. Final graph data-structure is realized by the list
representation with attributions representing geometrical features
like edge points, branchings, and loops. From results of the
application to patient’s cerebral blood vessels, we present the
~.ffectiveness of the system.
X-ray rotational
angiographic
system[ll]-[15].
3D
reconstruction of blood vessels images are reconstructed from
two-dimensional(21)) projection imagesby the short scan conebeamreconstruction algorithm [8],[13]. The analysis of the 3D
reconstructed blood vessels image includes the following
functions: (l)visualization of exterior structure and interior
morphology
of blood vessels, (2) extraction of the orientation
blood vessels, (3)me~urementof the cross sectional area, (4)
measurementof the distance between two points: along a 3D
blood vessel path, and (5) evaluation of the volumeof a vascular
disease like aneurysm.The effectiveness of the prototype system
is described with experimental results obtained by applying to
patient’s cerebral blood vessels. Section H presents the X-ray
rotational angiographic system and the 3D imagereconstruction
procedures. In Section m, the software tools to realize analysis
of the 3Dreconstructed imageare described. In Section IV, the
experimental results of the system application to patient’s
cerebral blood vessels are shown. A summaryand conclusion
are given in Section V.
H. 3D IMAGE RECONSTRUCTION
We used the X-ray rotational angiographic system
developed by Kumazaki et al.[15] to collect cone-beam
projection images of blood vessels. This measurementsystem
uses a cone.beamX-ray source and an ImageIntensifier ([I), and
is able to measure66 digital angiograms(512x512image) over
200 " within about 2 seconds. As preprocessing for the 3D
reconstruction, correction of image distortion generated by II
[16] and estimation of the measurementgeometryof the X-ray
rotational angiographic system[17] are performed, because a3D
image reconstruction methodneeds the accurate relationship
betweenthe coordinates in the 3Dimage and its corresponding
coordinates in the projection plane.
Various cone-beamreconstruction algorithms have been
proposed for medical application [4]-[10]. In practical
applications, the cone-beamreconstruction algorithm developed
by Feldkamp[5] obtained an efficient 3D reconstruction from
insufficient data. For sufficient data, the vertex of cone-beam
must be included on every plane that intersects the region of
interest [6]. However,the X-ray rotational angiographic system
scans so that the X-raysource is movedin a circular planar orbit,
thus the projection system does not meet the condition. Our
application has the additional restriction of not collecting
angiograms over a full 360° , because of the short period of
injecting contrast medium.Weused the short scan cone-beam
reconstruction algorithm for a circular orbit[8],[13] whichis
extended the short scan fan-beamreconstruction algorithm[ 18]
to a cone-beamreconstruction algorithm using the technique
presented by Feldkamp[5].
[. INTRODUCTION
Surgical planning or interventional radiology procedures
)f vascular disease require three-dimensional (3D) blood vessels
nformation such as precise 3D lesions structure
and
tuantification or 3D blood vessels path. X-ray angiogran~ has
:ertainly aided doctors in inspection of blood vessels because of
heir high spatial resolution, but two-dimensional(2D) projection
mages have limitation for providing 3D qualitative
and
tuantitative information of abnormalblood vessels. For that it is
lesirable to developthe followingtwo majorparts:
i1) 3Dreconstruction of blood vessels images, and
2) analysis of 3Dreconstructed blood vessels images.
Manyresearchers have developed methods to reconstruct
ID image of blood vessels from X-ray angiograms. In several
nethods, biplane and stereo angiographic system were utilized
mdit wasassumedthat cross sectional shapes of blood vessels are
:lliptic[ 1]-[3]. However,the information providedby only a few
~rojections is insufficient for reconstructing precise 3Dimageof
dood vessels. Furthermore, the assumption of cross sectional
hapes is not usually valid for abnormalblood vessels. Recentlya
ew studies have applied cone-beamalgorithms[4]-[8] to the 2D
~rojection images of blood vessel phantomsor patient’s blood
’essels collected from manyviews[9]-[14]. The main advantage
,f these cone-beamalgorithms is the direct reconstruction of 3D
mage from 2Dprojection images without the assumptionof cross
ectional shapes of blood vessels. Therefore, it can be useful for
aspection of abnormal blood vessels which have complicated
trucmre.
III.
ANALYSIS OF 3D RECONSTRUCTED BLOOD
Wehave developed a prototype system that provides
VESSELS IMAGES
.recise anatomicalinformationof patient’s blood vessels using an
169
In order to analyze the anatomical information such as the centedine are expressed as an undirected graph with vertexes
the orientation of blood vessels and cross sectional area, we and edges whichare obtained by using feature points and partial
transformed the 3Dreconstructed image of blood vessels into a centerlines segmented by vertexes respectively. The graph is
represented by the list representation with attributions
graph deseription[14]. Thesymbolic description of 3Dstructure
of blood vessels have been developedby G.Geriget al[24]. They represeoting geometrical features like edge points, branchings,
and loops.
employed the 3D thinning algorithm improved the 3D binary
thinning method proposed by Tsao and Fu[21] and a 3D
extension of the 2D 11 -graph concept. Wedeveloped the 3D IV. MEASUREMENT FUNCTION
thinning algorithm based on the necessary and sufficient
The measurementfunctions for analysis of anatomical
condition of l-voxel deletable in a 3D binary imageproposedby
blood vessels information are constructed based on the graph
Toriwaki et al [22] and a 3D extension of the distance
transformation of a line pattern[25]. Theprocedurefor the graph description of centerline of the 3D blood vessels image and are
description can be divided into four steps : segmentation, 3D described as follows.
thinning operation, distance transformation operation of a 3Dline (a) Extraction of the orientation of blood vessels: To estimate
pattern, and description of the centerline of blood vessels as a the cross section of blood vessels in planes perpendicular to the
graph.
local 3Dorientation of blood vessels, this function approximates
StepI (Segmentation)
: Astheregion
ofblood
vessels the local 3Dorientation of blood vessels by calculating local
a 3D reconstructed
bloodvesselimagetendsto havehigher direction of the 3Dcenterline of blood vessels image.
contrast
thanother
region,
weusedthethreshold
procedure
and (b) Measurementof the cross-sectional area : Before making
3Dobject
connectivity
toextract
theregion
ofblood
vessels
from treatment planning for lesions like stenoses, quantitative
the 3D reconstructed image. Hereafter, we consider a 3Dblood information such as vessels size or percent stenosis are required.
vessels imageto be a 3Dbinary imagein whichvoxel’s values of This function yields the cross-sectional area of blood vessels in
the region of blood vessels and backgroundregion are 1 and 0, planes perpendicular to the local 3Dorientation of blood vessels
obtained by the extraction of the orientation of blood vessels and
and are noted O-voxeland l-voxel respectively.
Step 2 (3D thinning operation) : A centerline of the
provides cross sectional shapes along 3Dcenterlines of 3Dblood
blood vessel imageis approximatelyextracted by thinning the 3D vessels.
blood vessel imageuntil a centerline of one voxel thickness with (c) Measurementof the distance between two points along a
preservation of the geometrical structure remains. This 3D blood vessel path : In interventional radiology procedures, it is
important to decide the adequate blood vessel path for guiding a
thinning operation must satisfies the following requirements: (i)
preserving connectivity of an original image, (ii) approximating catheter without damagingblood vessels. This function provides
the centerline of original 3Dimage, and (ill) consisting of the the shortest path between selected points on 3D blood vessels
maximal lines which are one point width. A 3D thinning
image and the distance between them is calculated using an
algorithm based on the constancyof the Euler’s number[19] or a algorithm calculating the shortest path problem. Thetrajectory of
straightforward extension of 2Dthinning algorithms[20] does not a guide line for the catheter is imposedon the 3Ddisplayed blood
necessarily meet the requirement of preserving connectivity[21]- vessels image.
[24]. Several technics have been proposed to satisfy the above (d) Evaluation of the volumeof a vascular decease: In surgical
requirements[2I]-[24]. Toriwakiet al providedthe necessary and operation or interventional radiology procedures for aneurysms,
sufficient condition of l-voxel deletable in a 3Dbinary image instrument like a clip or a balloon is used, then accurate position
using a 3Dextension of the connectivity numberin a 2Dimage of the neck of an aneurysm or its volume is an important
and the connectivity index[22]. Accordingto the necessary and information. This function allows doctors to detect a vascular
sufficient condition, the checkof l-voxel deletalbe is performed decease like an aneurysmin 3Dblood vessels image interactively
by using its 3x3x3 neighborhoods.Here, using the necessary and and calculate the volumeof the aneurysm.
sufficient condition of l-voxel deletable in a 3Dbinary imageand
EXPERIMENTAL RESULTS
the six sub-cyclesidea [22],a parallel 3Dthinning operation using V.
26-neighborhoods was examined for a 3D binary image.
Weapplied the proposedmethodto the patient’s cerebral
withan aneurysm.
Fig.1 showsthepartof 66
Especially, the parallel 3D thinning operation is modified to bloodvessels
removethe cause of elimination or separation of objects which projection
images
ofpatient’s
cerebral
blood
vessels
which
were
collected
bytheX-ray
rotational
angiographic
system.
Projection
usually occur with the parallel operation[22].
Step 3 ( distance transformation operation of a 3Dline
imagesarecomposed
of a 512x512
pixelmatrixandthe pixel
pattern) : We developed a 3D extension of the distance
pitch is 0.45 nun. These projection images were approximately
~ansformationof a 2Dline pattern[25] to obtain the information reconstructed into 320x320x320voxels ( voxel size 0.5x0.SxO.5
3) by the short scan cone-beamreconstruction algorithm. Fig.
of shape features of the centerlJne of blood vessels image. The mm
distance transformation operation of a 3Dline pattern change a 2 shows the cross section images of the 3D reconstracted blood
value at each voxel in the 3D line pattern into the distance
vessels image.
It is possible to observethe interior of blood vessels along
measuredalong the 3Dline pattern from that voxel to the furthest
end point. Thevalue at the voxel in the 3Dline pattern obtained the blood vessel path using the centerline of the 3Dblood vessels
allowedus to distinguish the feature points such as a edge point, a image as a guideline for the trajectory of a viewpoint. Fig. 3
shows the 3D visualization of 3Dblood vessels image by volume
branchingpoint, or a point on a loop.
Step4 (Graphdescription ) : Todescribe the centerline
rendering method[26],[27].The left image shows exterior
the 3D blood vessels image as a graph, the centerlines are structure of blood vessels in front of viewand the right ones show
disassembledinto line segmentsbetweenthe feature points. Then composite images of exterior structure and interior morphology
170
respectively of blood vcssels lying to the viewpoints in the
positions denoted by arrows. The rela6onship of the aneurysmto
the surrounding structure and to the root of its origin from the
vessels was observed.
Fig. 4 showsan exampleof selection of blood vessel path
for the measurement distance between two points along 3D
blood vessels path. After displaying 3D blood vessels image
from three different views, a user interactively inputs two points
on the 3Ddisplayed blood vessels images by crossh~dr cursor.
The two selected points in the screen coordinate system are
mappedinto the points in 3Dspace coordinate system of the 3D
¯ blood vessels image, Thenearest points from the obtained points
are detected on the centerline of 3Dblood vessels image. The
detected points on the centerline are considered as vertexes and
the graph of the centerline is changed and the new list
representation of the graph is produced. Using the new list
representation, the shortest path linked the two selected points
by the user is calculated and the shortest path is imposedto the
3Ddisplayed blood vessels images as the trajectory of a guide
line for catheter.
Fig.5 showsan exampleof an evaluation of the vascular
deceasein the 3Dreconstructed blood vessels imageof a patient.
Fig. 1 Projection imagesof cerebral blood vessels collected by
the X-ray rotational angiogrsphic system.
Fig. 4 An example of the selection of a blood vessel path.
After displaying 3Dblood vessels image from three differe~nt
views, a user interactively inputs two points on the 3Ddisplayed
blood vessels images by crosshair cursor. The shortest path be.
Fig. 2 Cross secdonimages of the 3D reconstructed blood
vessels image.
tween two points is automatically imposedto the 3Ddisplayed
bloodvessels imagesas the trajectory of the guide line for catheter.
Fig. 5 An example of the volume measurementof a vascular
Fig. 3 3D display of the 3Dreconstructed blood vessels image.
The left imageshowsexterior structure of blood vessels in front
decease. Manipulating the 3Dblood vessels image, a user can
of view and the right ones showcomposite images of exterior
structure and interior morphologyrespectively of blood vessels
interactively detect the region of interesting. The region of
aneurysmand the catheter guide line are colored red on the 3D
lying to the viewpointsin the positions denotedby arrows.
displays.
171
The region of the aneurysm and the catheter guide line are
colored red on the 3D displays. Manipulating the 3D blood
vessels image, a user can interactively detect the region of
interesting (ROI). The volume of the ROI (ie. aneurysm)
approximately obtained from the numberof voxel contained in
the detected region. Using the 3Ddisplay like Fig.5 it will be
easily to grasp anatomical relationship between the abnormal
region and others.
V. CONCLUSION
Wehave presented a methodto reconstruct precise 3D
blood vessels images using the X-ray rotational angiographio
systemand realize 3Dvisualization and 3Dquantitative analysis.
Theshort scan cone-beamreconstruction algorithm enabled us to
reconstruct 3D blood vessel images approximately from
insufficient samplingof cone-beamprojections. The analyses of
anatomical blood vessels information were constructed based on
the graph description of 3Dblood vessels image using the 3D
thinning algorithm and the distance transformation of a 3Dline
pattern. Theresults of the experimentsshowedthe feasibility of
the prototype system for providing 3Dquantitative analysis such
as accurate 3D localization of the neck of an aneurysm or
measurementof vessels size at a stenosis lesion, whichcould be
an effectivity tool for interventional radiology, surgical planning,
or quantitative diagnosis.
REFERENCES
[1] N. Niki, S. Horie, and K. Matsumoto,"High-precision 3-D
shaded display of the brain constructed
from CT and
angiogrephical images’, IEICETrans., Japan, vol. J70-D, no.12,
pp.2525-2534, 1987.
[2] K. Kitamura, J.M. Tobis, and J. Sklansky, "Estimating the
3-D skeletons and transverse areas of coronary arteries from
biplane angiograms", IEEE Trans. Med. Imaging , vol. MI-7,
oo.3, pp.173-187, 1988.
[3] J.A. Fessler and A. Macovski,"Object-based
3-D
reconstruction of arterial trees from magnetic resonance
angiograms", IEEE Trans., Med. Imaging, vol. MI-10, no.l,
pp.25-39,1991.
[4] H.K. Tuy : " An inversion formula for cone-beam
reconstruction ", SLMA
J. Appl. Math., 43, 3, pp.546-552, 1983.
[5] L.A. Feldkamp, L.C. Davis, and J.W. Kress, "Practical
cone-beamalgorithm", J. Opt. Soc. Am.A, vol.1, no.6, pp.612619, 1984.
[6] B.D. Smith, "Image reconstruction
from cone-beam
projections:
Necessary and sufficient
conditions and
reconstruction methods", IEEETrans. Meal. Imaging, voL MI-4,
no.l, pp.14-25, 1985.
[7] H. Kudo and T. Saito, "Three dimensional tomographic
image reconstruction from cone beam projections by helical
scanning method’, IEICETrans., Japan, vol. J72-D-II, no.6,
pp.954-962, 1989.
[8] G.T GuUberg arid G.L. Zeng, "A cone-beam filtered
backprojection reconstruction algorithm for cardiac single
photon emission computed tomography", IEEE Trans. Med.
Imaging, vol. MI-II, no.l, pp. 91-101, 1992.
[9] R. Ning, R.A. Kruger, H. Hu, "Image intensifier-based CT
volume imager for angiography: system evaluation", SPIE
vol.1090, pp.131-143, 1989.
[10] Y. Trousset, D. Saint-Felix, A. Rougeeand C. Chardenon,
"Multiscale cone-beamX-ray reconstruction", SPIE, voL 1231,
172
pp. 229-238, 1990.
[11] N. Niki, T. Mineyama,H. Satoh, C. Uyama, T.Kumazaki,
"3-D reconstruction
of blood vessels using a rotational
angiopraphy system’, Med.and Biol. Eng. and Comput., vol.29,
part l, p. 649, 1991.
[12] Y. Kawata, N. Niki, H. Satoh, T. Kumazaki, "3D
reconstruction of blood vessels using an X-ray rotational
projection system", 7th Int. Conf. on Biomed.Eng., Singapore,
pp.327-329, 1992.
[13] Y. Kawata, N. Niki, H. Satoh, T. Kumazaki, "Threedimensional blood vessels reconstruction using a high-speed Xray rotational projection system’, IEICETrans., Japan, vol. J76D-II, no.9, pp.2133-2142,1993.
[14] N. Niki, Y. Kawata, H. Satoh and T. Kumazaki, " 3D
imaging of blood vessels using X-ray rotational angiographic
system’, IEEEMedical Imaging Conference, in press 1993
[15] T. Kumazaki, "Rotational stereo-digital
angiographyimage processing and clinical usefulness’, Med. Imag. Tech.,
Japan, vol. 7, no.4, pp.433-439,1989.
[16] J.M. Boone, J.A. Seibert, W.A. Barrett and E.A. Blood,
"Analysis and correction of imperfections in the image
inteusifier-TV-digitizer
imaging chain", Med.Phys. vol.180,
no.2, pp. 236-242, 1991.
[17] G.T. Gullberg, B.M.W.Tsui, C.R. Crawford, J.G. Ballard
and J.T. Hagius, "Estimation of geometrical parameters and
collimator evaluation for cone beam tomography", Med. Phys.
vol. 17, no.2, pp.264-272, 1990.
[ 18] D.L. Parker, "Optimalshort scan convolution reconstruction
for fanbeamCT’, Med. Phys. vol.9, no.2, pp. 254-257, 1982.
[19] S. Lebregt, P.W. Verbeek, and F.C.A. Groen, " Threedimensional skeletonization, principle and algorithm.", IEEE
Trans.Pattem Anal. &Mach. Intell., vol. PAMI-2,no. 1 pp.7577, 1980.
[20] L. Lam, S.W. Lee, and C.Y. Suen, "Thinning
methodologies - A comprehensive survey", IEEE Trans.Pattem
Anal. &Mach.InteU., vol. PAMI-14,no. 9, pp.869-885, 1992.
[21] Y.F. Tsao and K.S. Fu, "A parallel thinning algorithm for 3D pictures", CGIP,vol. 17, pp. 315-331, 1981.
[22] J. Toriwaki and S. Yokoi, "Basics of algorithms for
processing three-dimensional digitized pictures", IECETrans.,
Japan, vol. J68-D. no.4, pp. 426-433, 1985.
[23] G. Malandain, G. Bertrand, and N.Ayache, " Topological
classification in digital space", IPMI’91, pp.300-313, BerlinHeidelberg, 1991, Springer-Verlag.
[24] G. Gerig, Th. Koller, G. Szekely, Ch. Brechbuhler and O.
Kubler, "Symbolic description of 3-D structures applied to
cerebral vessel tree obtained from MRangiography volume
data’, IPMI~)3, pp.94-111, Arizona,USA., 1993, SpringerVedag.
[25] J. Toriwaki, N. Kato, and T. Fukumura, "Parallel local
operations for a newdistance transformation of a line pattern and
their applications", IEEE Trans. Syst., Man. & Cybern., vol.
SMC-9,no.lO, pp.628-643, 1979.
[26] M.Levoy, "Volume rendering: display of surfaces from
volumedata", IEEECO&A, vol. 8, no. 3, pp.29-37, 1988.
[27] N.Niki, Y.Kawataand H.Satoh, "A 3-D display method of
fuzzy shapes obtained from medical images", IEICE Trans.,
Japan, vol.J73-D-II, no.10, pp.1707-1715,1990.