NOVEL TECHNIQUES FOR AUTOMATICALLY ENHANCED

VISUALIZATION OF CORONARY ARTERIES IN MSCT DATA AND

FOR DRAWING DIRECT COMPARISONS TO CONVENTIONAL

ANGIOGRAPHY

Marion J¨ahne, Christina Lacalli* and Stefan Wesarg*

Technische Universit¨at Darmstadt, Department of Graphic Interactive Systems, Rundeturmstr. 10, 64283 Darmstadt, Germany

*Fraunhofer IGD, Department of Cognitive Computing & Medical Imaging, Fraunhoferstr. 5, 64283 Darmstadt, Germany

Keywords:

Cardiac imaging, MSCT, coronary arteries, conventional coronary angiography, DRR, visualization.

Abstract:

The new generation of multi-slice computed tomography (MSCT) scanners enables the radiologist to assess

the coronary arteries in a non-invasive way. The question of particular interest is whether the quality of the

ﬁndings based on MSCT data can compete with the gold standard - the coronary angiography. In this work

we present novel automated methods for a reliable visualization of coronary arteries and for drawing direct

visual side-by-side comparisons to conventional angiograms. Our approach comprises a new method for

automatically extracting the heart from cardiac CT data and an advanced masking method for eliminating large

cardiac cavities to obtain a better visibility of the coronary arteries in the rendered CT data. For drawing direct

side-by-side comparisons we present a novel approach for simulating the conventional coronary angiography

in an easy-to-handle manner. The new methods have been developed for and tested with contrast-enhanced

cardiac CT datasets.

1 INTRODUCTION

In this work we present novel techniques for direct

volume rendering of the coronary arteries from car-

diac MSCT data and for drawing side-by-side com-

parisons to conventional angiographyin an automated

manner.

Coronary artery disease (CAD) and its compli-

cations such as myocardial infarction are among the

main causes of death in the western world. CAD most

often results from a decrease of the cross-sectional

area of the coronary arteries. Therefore, a reliable

detection of those narrowed areas – the so-called

stenoses – is of high importance for diagnosis and in

order to plan a possible treatment.

For years, the coronary angiography was the gold

standard for the assessment of the coronary artery tree

– in spite of the fact that it is an invasive modality and

therefore associated with potential negative implica-

tions for the patient. The new generation of MSCT

scanners allows the analysis of the coronary arteries

in a non-invasive way. The question of particular in-

terest is whether the quality of the ﬁndings based on

these image data can compete with the conventional

coronary angiography. To answer this question a re-

liable representation of the coronary arteries that is

comparable to the representation in conventional an-

giograms is required. Further on, registration and

matching of these representations allow transfering

positions of potential pathologies (stenoses, calciﬁ-

cation) detected in CT data to angiograms and vice

versa.

Typically, cardiac CT data also includes non-cardiac

structures such as ribs or the sternum which partially

block the view at the heart and at the coronary arter-

ies (see Figure 2 (a)). In this work we present an au-

tomated approach for extracting the heart from car-

diac CT data for a better visibility of the coronary ar-

teries. However, drawing direct visual comparisons

between corresponding pathologies remains difﬁcult.

This is because on one hand, only parts of the coro-

nary arteries are visible. On the other hand, manually

achieving a viewing direction adequate to that in con-

ventional angiography is very inaccurate. We present

an additional method, that automatically generates so

called digitally reconstructed radiographies (DRRs)

from the CT data by simulating the conventionalcoro-

nary angiography. Representations of the coronary ar-

290

Jähne M., Lacalli C. and Wesarg S. (2008).

NOVEL TECHNIQUES FOR AUTOMATICALLY ENHANCED VISUALIZATION OF CORONARY ARTERIES IN MSCT DATA AND FOR DRAWING

DIRECT COMPARISONS TO CONVENTIONAL ANGIOGRAPHY.

In Proceedings of the Third International Conference on Computer Vision Theory and Applications, pages 290-296

DOI: 10.5220/0001082102900296

 SciTePress

original CT data

threshold selection

method by Otsu

DRR generationvolume rendering

centre of gravity

computation

binarization

erosion

radial search rays

generation of

binary mask

neighborhood filter

dilatation

binary mask (3D)

to eliminate

cardiac cavities

binary mask (3D)

to eliminate non-cardiac

structures

combined mask

masked CT data

DRR with whole

coronary artery tree

volume rendering

CT data without

non-cardiac structures

and large cardiac cavities

CT data without

non-cardiac structures

Figure 1: Our three possibilities for the visualization of

coronary arteries from CT data. Left: Volume rendered CT

data without displaying spurious non-cardiac structures (see

Figure 2 (b)). Middle: Generated DRR from masked CT

data. Both, non-cardiac structures and large cardiac cavities

are eliminated (see Figure 6 (b)). Right: Volume rendered

CT data without displaying non-cardiac structures and large

cardiac cavities (see Figure 5 (b)).

teries generated this way, can easily be compared with

conventional angiograms.

2 STATE OF THE ART

2004 Lorenz et al. developed a method for the ex-

traction of the heart from cardiac CT datasets (Lorenz

et al., 2004). They locate the chest wall and the de-

scending aorta in all slices of the CT data in order to

roughly estimate the location of the heart. Afterwards

they use active contours to delineate the border of the

heart in a slice-by-slice manner.

A reliable visual comparison between cardiac

structures represented in 3D CT data and in 2D an-

giograms respectively, requires similar representa-

tions of those structures. In general, there are two

possibilities to meet this requirement. Either, 3D re-

construction of the coronary arteries from a series of

2D angiograms or the opposite way around, project-

ing the CT data to a 2D image plane and compare

these 2D images with conventional angiograms. In

(Sang et al., 2006), (Großkopf and Hildebrand, 1997)

and (Blondel et al., 2006) a 3D model of the coro-

nary artery tree is reconstructed by segmenting the

coronary artery tree. This 3D representation can be

compared with the coronary tree extracted from the

CT data. In (Schnapauff et al., 2007) emulation of

conventional angiography is performed by segment-

ing the coronary arteries in the CT data and displaying

the results by volume rendering techniques.

3 MATERIAL AND METHODS

In the following section, we describe our set of new

approaches for an enhanced visualization of the coro-

nary arteries in cardiac CT data (see Figure 1 for an

overview). First, we present a new method for ex-

tracting the heart, followed by an advanced masking

of large cardiac cavities to clearly visualize the whole

coronary artery tree in a volume rendering of the CT

data. The latter serves as a basis for the generation

of DRRs to simulate conventional coronary angiogra-

phy.

3.1 Image Acquisition

A total of 25 contrast-enhanced CT datasets were ac-

quired by 16-slice

and 64-slice

CT scanners. The

number of voxels in x- and y-direction constantly

amounts 512 × 512 voxels and varies in z-direction

from 176 to 441. The size of one voxel in x- and y-

direction is identically, varying from 0.41 to 0.54 mm,

and averages 0.5 mm along the z-axis. The gray level

values of the CT data are normalized values of the

computed X-ray attenuation coefﬁcients, expressed in

Hounsﬁeld Units (HU) (Hounsﬁeld, 1992). Typically,

the CT data includes 4096 gray level values (12 bit).

In 8 out of 25 cases X-ray coronary angiography

was performed

in addition to the CT data acquisition.

The resolution of the angiograms amounts 512 × 512

pixels and they were digitized with 8 bit, thus the gray

level values lie in between 0 and 255.

3.2 Enhanced Direct Volume Rendering

of the Heart

When displaying the original CT data by volume ren-

dering techniques non-cardiac structures such as ribs

SIEMENS Somatom Session 16

GE Light Speed VCT

SIEMENS HICOR/ACOM-TOP, PHILIPS Integris H

NOVEL TECHNIQUES FOR AUTOMATICALLY ENHANCED VISUALIZATION OF CORONARY ARTERIES IN

MSCT DATA AND FOR DRAWING DIRECT COMPARISONS TO CONVENTIONAL ANGIOGRAPHY

291

(a) (b)

Figure 2: Volume rendered cardiac CT data. (a) Non-cardiac structures such as ribs or the sternum block the view at the heart

and at the coronary arteries. (b) After applying the mask generated in 3.2 the heart and single branches of the coronary arteries

are clearly visible.

or the sternum block the view at the heart and at the

coronary arteries (see Figure 2 (a)). Here, we present

our new approach for eliminating those structures in

order to obtain a clear view at the heart and at parts

of the coronary artery tree in the volume rendered CT

data.

Our approach for automatically extracting the heart

from cardiac CT data consists of the following two

steps. First, we use a statistical method to compute a

point in the middle of the heart in all axial slices of

the CT data. From this point we send out a search ray

pattern to detect the outer boundary of the heart.

The gray level values of the CT data expressed in

HU represent a linear transformation of the measured

X-ray attenuation coefﬁcients. Thus, the gray level

values correlate with the different anatomical struc-

tures (air -1000 HU, water 0 HU, bone +1000 up to

+3000 HU). In order to obtain a clear partition of the

anatomical structures, we use Otsu’s method of auto-

matic threshold selection from gray level histogram

for image segmentation (Otsu, 1979). An optimal

set of thresholds is selected by the discriminant cri-

terion; namely by maximizing the measure of separa-

bility of the resultant classes in gray levels. With two

thresholds the voxels of the original CT data are di-

chotomized into three gray level classes. The darkest

gray level corresponds to air, the middle one corre-

sponds to fat and muscle tissue and the brightest gray

level corresponds to bones and the contrast-enhanced

cardiac structures (see Figure 3 (b)). As the result of

the computation of the center of gravitiy of the two

brightest gray levels we obtain a point in the middle

of the heart.

In the axial slices of the labeled CT data the out-

line of the heart can be approximated by a circle or

an ellipse. Therefore, from the computed center of

gravity in the middle of the heart a radial search ray

pattern (von Berg, 2005) is send out in order to detect

the outer boundary of the heart. Clear bright-to-dark

transitions from the heart to the surrounding tissue are

found. At locations with overlaying structures, like

the aorta and the sternum, the search rays are too long

(see Figure 3 (b)) and need to be cut in a subsequent

step. On each side of the aorta and the sternum the last

rays which hit on lung tissue and therefore have the

correct length are automatically detected . The length

of the rays which lie between these rays is corrected

by interpolating their new length from the length of

the detected rays (see Figure 3 (c)). Afterwards the

endpoints of the search rays are stored and then con-

nected to get the outline of the heart in every axial

slice. We obtain the binary mask by ﬁlling up the in-

ner of the 2D outer boundary of the heart.

By applying the generated mask to the original CT

data and by rendering these data, the heart and single

branches of the coronary artery tree are clearly visible

(see Figure 2 (b)).

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

292

(a) (b) (c)

Figure 3: Detection of the outer boundary of the heart using the example of a single axial slice. (a) Original CT data. (b)

Labeled CT data after performing Otsu’s method. The gray labels 0 to 2 correspond to the anatomical structures air, fat and

muscle tissues, bone and contrast-enhanced cardiac structures. At locations with overlaying structures (sternum, aorta) the

search rays are too long (1). On each side the last rays which hit on the lung tissue, and therefore have the correct length, are

automatically detected (2). (c) The length of the rays (1) is corrected by interpolating their new length from the rays (2).

3.3 Drawing Side-by-Side Comparisons

The generation of DRRs from the original CT data

leads to insufﬁcient results concerning the visibility

of the coronary artery tree in the projected images.

On one hand non-cardiac structures such as ribs block

the view at the heart. This problem can be ﬁxed by

eliminating those structures (see 3.2). On the other

hand large cardiac cavities like ventricles and atria are

in addition to the coronary arteries contrast-enhanced

and occlude the coronary arteries in the projected im-

ages. The latter is because in the case of MSCT the

contrast medium is applied systemically and is not in-

jected directly into the coronary artery branches like

in conventional coronary angiography. We present an

advancedmasking of spurious cardiac cavities for bet-

ter visibility of the whole coronary artery tree. The

masked CT data serves as input for generating DRRs

in order to draw direct side-by-side comparisons to

conventional angiography.

3.3.1 Advanced Masking

Our approach for masking out spurious cardiac cav-

ities is based on morphological image processing

techniques and neighborhood ﬁlters. First, we per-

form a thresholding operation on the labeled CT data.

The highest label correlates with both, the contrast-

enhanced cardiac cavities and the contrast-enhanced

coronary arteries (see Figure 4 (a)). Thus, the result-

ing binary mask eliminates large structures correlat-

ing with the cardiac cavities as well as small struc-

tures correlating with the coronary arteries (see Fig-

ure 4 (b)). To avoid the masking of the coronary ar-

teries, we perform an erosion operation on the thresh-

olded data. This allows a clear seperation of coronary

arteries and cardiac cavities, even when they are in

close proximity to each other. Small areas of con-

nected pixels can be assigned to the coronary arteries

in the eroded data. We then apply a neighborhood

ﬁlter along each of the three orthogonal axes of the

dataset to remove those areas. As the size of the car-

diac cavities was reduced by eroding the data, we per-

form a dilatation operation for resizing purpose.

By applying the generated mask to the CT data af-

ter extracting the heart, the data can be rendered that

way that the whole coronary artery tree is clearly vis-

ible (see Figure 5 (b)).

3.3.2 DRR Generation

Simulating conventional coronary angiography re-

quires an appropriate projection model as well as in-

formation about the X-ray attenuation coefﬁcients of

the different anatomical structures. In our approach,

a perspective projection model forms the geometrical

basis for the calculation of the DRRs. Obtaining com-

parable views to those of conventional angiograms,

requires general knowledge of the projection param-

eters such as the origin of the CT data, the position

of the center of the detector, or the position of the

virtual X-ray source. In our model we assume that

the central ray starting from the virtual X-ray source

passes directly through the virtual iso-center - that is

the center of the CT data - and the center of the de-

tector. This mimics the rigid geometry of the X-Ray

C-arm. The transformation to project one voxel of

the CT data to the detector plane can be expressed

in terms of a translation t, followed by a rotation

R = R

· R

around the transversal and longitudi-

nal axes of the volume, respectively, and a perspective

projection matrix P

perspective

. The angiograms are pre-

NOVEL TECHNIQUES FOR AUTOMATICALLY ENHANCED VISUALIZATION OF CORONARY ARTERIES IN

MSCT DATA AND FOR DRAWING DIRECT COMPARISONS TO CONVENTIONAL ANGIOGRAPHY

293

(a) (b) (c)

Figure 4: Advanced masking. (a) The highest gray level value in labeled CT data correlates with both, cardiac cavities and

coronary arteries. (b) Binary CT data after thresholding. The coronary arteries (arrows) must be excluded from masking.

(c) Masked CT data after applying the combined mask obtained in previous steps to the original CT data. Both, non-cardiac

structures and cardiac cavities are eliminated, whereas the coronary arteries (arrows) are still visible. The masked CT data

serves as input for the generation of DRRs.

sented in DICOM

format. Typically, its header ﬁle

contains acquisition speciﬁc information. By parsing

the DICOM header we obtain the relevant projection

parameters.

The transmitted intensity I of X-ray s, is given by,

I = I

· e

−

∑

, (1)

whereas I

is the incident X-ray intensity and µ

are

the linear attenuation coefﬁcients. The intensity of

one pixel of the DRR is computed by analysing the

exponent in Equation 1, i.e. by accumulating the gray

values of the CT data along each of the virtual X-

rays passing through the CT data. This procedure

conforms to a classical ray-casting approach as im-

plemented in (Schroeder et al., 2002).

4 RESULTS

We tested our approach for extracting the heart and

the advanced masking of large cardiac cavities with

all 25 CT datasets. In all cases we obtained clinical

useful results at a single press of a button (see Figure

2 (b) and Figure 5 (b)). The generation of DRR was

tested on those 8 CT datasets for which corresponding

conventional angiograms existed. In 5 out of 8 cases

spurious cardiac cavities could successfully masked,

while the run of the coronary arteries was clearly vis-

ible (see Figure 6 (b)). In the other 3 cases the quality

of the original CT data was reduced, e.g. the contrast

medium was not equally distributed. Therefore, the

coronary arteries were not sufﬁciently enhanced.

By generating the DRRs the radiologist obtains a

representation of the coronary tree that is morphologi-

Digital Imaging and Communications in Medicine

cally equivalent to conventional angiograms (see Fig-

ure 6 (a)-(b)). Right and left branches of the coronary

artery tree are visible in the projected images because

in the case of MSCT the contrast medium is applied

intravenously. For more similar representations we

clip the volume along the axial direction, projecting

only the relevant part of the CT data (see Figure 6

(c)).

5 CONCLUSIONS AND

DISCUSSION

The approaches presented in this work are new and

automated techniques for enhanced visualization of

coronary arteries in MSCT data. The extraction of

the heart and the masking of large cardiac cavities en-

ables the radiologist a clear view at the heart and at

the coronary artery tree in volume rendered CT data.

For drawing direct visual side-by-side comparisons

between representations of the coronary arteries ob-

tained by MSCT with that in conventionalangiograms

we present a novel automated technique based on the

generation of DRRs.

While (Lorenz et al., 2004) delineates the heart

from the outside, we extract the heart from within a

point in the middle of the heart. Thus, we need less

information about the position of the heart in the CT

data and therefore our approach is more ﬂexibly appli-

cable (e.g., regarding the portion of the rib cage that is

included in the CT data). For simulating the coronary

angiography we proposed the generation of DRRs,

which is a common technique to obtain radiograph-

ical representations from tomographic datasets. The

advantage of our approach is that we get along with-

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

294

(a) (b)

Figure 5: Volume rendered cardiac CT data. (a) By eliminating non-cardiac structures only parts of the coronary artery tree

are visible. (b) By eliminating also large cardiac cavities the whole coronary artery tree is clearly visible.

out an explicit segmentation of the coronary artery

tree, neither in the CT data nor in the conventional

angiograms. At a single press of a button the radiolo-

gist obtains a representation of the coronary arteries

analogous to the representation in conventional an-

giograms. To our knowledge, no comparable auto-

mated approach has been published yet.

Since we opperate on the entire CT data the mask-

ing of cardiac cavities and the generation of DRRs de-

pends on the quality of the original CT data. For in-

stance, if the CT dataset was reconstructed at a point

in time, where the contrast medium was not opti-

mally distributed, a clear assigment of the contrast-

enhanced cardiac cavities to the two brightest gray

level values in the labeled CT data is difﬁcult. To cir-

cumvent such problems the radiologist has the possi-

bility to adjust some parameters (e.g. the number of

thresholds).

Currently, we extend our approach by aligning the

generated DRR and the angiogram. By registrating

the two modalities it will be possible to highlite po-

sitions of potential pathologies (stenoses, calciﬁca-

tion) detected by quantitative 3D CT-analysis in the

angiograms. This would be a further step in enhanc-

ing the trust of the radiologist in the assessment of the

coronary arteries by MSCT.

An extended clinical evaluation together with our

clinical partners will be conducted in the near future.

ACKNOWLEDGEMENTS

We want to thank the Institute for Diagnostic and

Interventional Radiology of the Johann Wolfgang

Goethe University Frankfurt, Germany for provid-

ing us the cardiac CT datasets and conventional an-

giograms.

REFERENCES

Blondel, C., Malandain, G., Vaillant, R., and Ayache, N.

(2006). Reconstruction of coronary arteries from a

single rotational X-ray projection sequence. IEEE

Transactions on Medical Imaging, 25:653–663.

Großkopf, S. and Hildebrand, A. (1997). Three-

dimensional reconstruction of coronary arteries from

x-ray projections. In Lanzer, P., editor, Diagnostics of

Vascular Diseases. Principles and Technology, pages

307–314. Springer, Berlin; Heidelberg.

Hounsﬁeld, G. N. (1992). Nobel Lecture, December 8,

1979. In Lindsten, J., editor, Nobel Lectures, Physi-

ology or Medicine 1971-1980, pages 568–586. World

Scientiﬁc Publishing Co., Singapore.

Lorenz, C., Lessick, J., Lavi, G., Blow, T., and Renisch, S.

(2004). Fast automatic delineation of cardiac volume

of interest in msct images. Proceedings of the SPIE,

Medical Imaging 2004: Image Processing, 5370:456–

466.

Otsu, N. (1979). A threshold selection method from gray

level histograms. IEEE Trans. Systems, Man and Cy-

bernetics, 9:62–66.

NOVEL TECHNIQUES FOR AUTOMATICALLY ENHANCED VISUALIZATION OF CORONARY ARTERIES IN

MSCT DATA AND FOR DRAWING DIRECT COMPARISONS TO CONVENTIONAL ANGIOGRAPHY

295

(a) (b) (c)

Figure 6: (a) Conventional angiogram of the left coronary artery branch (RAO 30.9 / CRAN 0.1) (b) Generated DRR from

the pre-processed CT data with projection parameters extracted from DICOM header ﬁle of (a). Both, left and right coronary

artery branches are visible. (c) Generated DRR from clipped pre-processed CT data.

Sang, N., Peng, W., Li, H., Zhang, Z., and Zhang, T. (2006).

3D reconstruction of the coronary tree from two x-ray

angiographic views. Proceedings of the SPIE, Med-

ical Imaging 2006: Image Processing, 6144:1591–

1598.

Schnapauff, D., D¨ubel, H., Scholze, J., Baumann, G.,

Hamm, B., and Dewey, M. (2007). Multislice

computed tomography: angiographic emulation ver-

sus standard assessment for detection of coronary

stenoses. European Radiology, 17:1858–1864.

Schroeder, S., Martin, K., and Lorensen, B. (2002). The

Visualization Toolkit. An Object-Oriented Approach to

3D Graphics. Kitware, Inc., 3rd edition.

von Berg, L. (2005). Fast automated object detection by re-

cursive casting of search rays. International Congress

Series, 1281:230–235.

VISAPP 2008 - International Conference on Computer Vision Theory and Applications

296