PHYSICALLY MEANINGFUL VIRTUAL UNENHANCED IMAGERECONSTRUCTION FROM DUAL-ENERGY CT

Mahnaz Maddah, Paulo R. S. Mendonca, and Rahul Bhotika

GE Global Research, One Research Circle, Niskayuna, NY 12309, USA

ABSTRACT

Virtual unenhanced (VUE) image reconstruction tech-niques have the potential to eliminate the need for theacquisition of a non-contrast exam in imaging protocolsthat currently require both contrast and non-contrastcomputed tomographic (CT) scans. Here, we proposea new physically meaningful approach for the generationof VUE images as an application of a recently introducedtechnique for multi-material decomposition (MMD) fromdual-energy CT. Rather than subtracting an estimateof the contribution to the image intensity in Hounsfieldunits due to contrast agent from the original image, thenew algorithm replaces the estimated volume of contrastin each voxel by the same volume of the material thatthe contrast has displaced, namely, blood. Our results onboth phantom and patient data demonstrate that VUEimages can replace true unenhanced images effectively,thereby reducing radiation exposure and scan time ofCT exams.

Index Terms— Virtual unenhanced, multi-materi-al decomposition, computed tomography, fast kV switch-ing, dual energy

1. INTRODUCTION

Computed tomography (CT) has been extensively usedin assessing patients with suspected urinary stones [10],renal masses [3], and liver problems [4]. A common pro-tocol for such CT examinations involves true unenhanced(TUE) image acquisition, typically followed by at leastone more scan after the injection of an iodine-based con-trast agent.

Dual-energy computed tomography (DECT) allowsfor the production of density images for a given mate-rial basis. Typical material basis are water and calciumor water and iodine. Assuming the later case, the den-sity images correspond to the density map of water andthe density map of iodine that yields the same projec-tions as in the original DECT data [11]. These densitymaps can be use to produce monochromatic images, i.e.,images equivalent to those obtained from a narrow-band,monochromatic X-ray beam [11]. It has been shown that

(a) (b)

Fig. 1: Monochromatic images in Hounsfield units at (a) 70keV and (b) 140 keV.

the adoption of DECT systems may allow for the elim-ination of the unenhanced scan of certain imaging pro-tocols [10, 3]. This elimination can be achieved throughthe generation of virtual unenhanced (VUE) images [6],which are invariably described in the literature as someform of contrast or iodine subtraction based on a triplematerial decomposition method or simply the water im-age, converted to Hounsfield units, of the water/iodinepair [9, 2]. Significant reduction in radiation exposure(e.g. 35% in a study on the identification of renal masses[3]) and scan time as well as elimination of misregis-tration between two different acquisitions are the mostimportant benefits of virtual unenhanced image recon-struction. The generation of high-quality VUE images,required to achieve such benefits, is the focus of this pa-per.

It is important to observe that the simple removal ofcontrast agent is not physically meaningful; instead, itsreplacement with the volume of blood that the contrastagent has displaced is the appropriate protocol. How-ever, this protocol requires an accurate estimate of theblood volume displaced by the contrast agent. Recently,a novel multi-material decomposition technique based onDECT has been introduced [7]. In this technique thecontribution of each constituent material in a mix of ma-terials is expressed in terms of volume fractions, i.e., thepercentage contribution in volume of each constituent

808978-1-4244-4126-6/10/$25.00 ©2010 IEEE ISBI 2010

(a) (b) (c) (d)

Fig. 2: Multi-material decomposition of the DECT data. The images shown in Fig. 1 have been decomposed into (a)fat, (b) blood, (c) bone, and (d) contrast agent components

material of the mix. The method is based on a physico-chemical model that assumes that materials in the body(as well as typical contrast agents) mix to form an idealsolution, in which the volume of a mixture is equal tothe sum of the volume of its constituent parts [8].

In this work, we present a virtual unenhanced im-age reconstruction technique which is based on multi-material decomposition. We apply the method on bothphantom and anatomical data and provide qualitativeand quantitative evidence that TUE can be reliably re-placed by VUE images.

2. BACKGROUND ON MATERIALDECOMPOSITION

We refer the reader to [1] for an exposition on the funda-ments of dual-energy CT. Given a pair of monochromaticimages at two distinct energy levels (e.g, 70 keV and 140keV, depicted in Fig. 1), as produced by some DECTsystems, the multi-material decomposition method in [7]seeks an accurate estimation of volume fractions of Npre-selected materials for each voxel of the scanned vol-ume. X-ray linear attenuation, denoted by μL, is pri-marily a function of incident X-ray energy E, the massdensity ρ, and composition of the material being imaged.In [7] the mix of materials and contrast agents within thehuman body is assumed to form an ideal solution. Un-der this assumption it follows that the linear attenuationcoefficient μL can be expressed as:

μL(E) =

N∑

i=1

αiμL,i(E)

where αi = vi/(∑

vj) is the volume fraction of material

i and∑N

i=1 αi = 1. The goal of material decompositionis to find αi’s under constrain that they sum up to 1.Mathematically, N − 1 measurements are needed to de-compose the solution into N materials. Therefore, withDECT measurements that provide only two equations,

simultaneous material decomposition into N > 3 mate-rials is not well-posed. However, a sequential decomposi-tion into different triplets may still be possible, becausea poor choice of a material triplet may yield solutionsfor αi that violate the constraints 0 ≤ αi ≤ 1, and thisviolation can be used as a flag to indicate that anotherchoice of material triplet could be more appropriate fora given location in the image [7]. Figure 2 shows thedecomposition of the images in Fig. 1 into its fat, blood,bone, and contrast agent components.

3. VIRTUAL UNENHANCED IMAGING

The estimation of volume fractions described in [7] allowsfor the development of a virtual unenhanced techniquethat carries out a replacement rather than the removalof contrast agent from the contrast-enahanced images.Note that this operation does not have an elementarycorrespondent for kVp-based dual-energy imaging, be-cause of the non-linearity of the map between the tworepresentations [5].

The key piece of information necessary for the gener-ation of virtual unenhanced images via the replacementof contrast agent by blood is the volume of blood thatis displaced by the contrast. Under the same assump-tion that the mix of blood and contrast agent form anideal solution used in [7], this volume must be equal tothat of the contrast agent. Having the volume fractionsof each material, we therefore simply replace the volumeof contrast agent by the same volume of blood. At agiven energy, we then reconstruct the linear attenuationcoefficient μVUE

L (E) of the unenhanced image as

μVUE

L (E) =

N∑

i=1i �=i1

αiμL,i(E) + αi1μL,i2(E), (1)

where the index i1 refers to the contrast agent, and theindex i2 refers to blood. Contrast subtraction, if still

809

(a) (b) (c) (d) (e)

Fig. 3: CT images of phantom containing different materials. (a) Monochromatic image at 70 keV; Volume fractionimages for (b) fat, (c) calcium, (d) contrast agent and (e) reconstructed virtual unenhanced image. Note that the two whiterings containing a mixture of 5% iodine in saline, are disappeared in the generated VUE image.

desired, can be obtained by omitting the second term inthe sum on the right-hand side of (1). Furthermore, sub-traction or replacement of any material for another, evenif not a contrast agent, can be easily obtained throughthe appropriate change in (1).

As pointed out in the literature [3], an important limi-tation of current virtual unenhanced techniques based ontriple material decomposition is that the material tripletis pre-selected and kept fixed throughout the analysis.A typical selection is fat, soft tissue, and iodine, whichfaces problems in image areas depicting a high concen-tration of calcium. The automated selection of materialtriplet in [7] overcomes this limitation without the sepa-rate post-processing step suggested in [3].

4. RESULTS AND DISCUSSIONS

Figure 3 shows a monochromatic DECT image of a phan-tom data taken at 70 keV, the estimated volume fractionsfor fat, calcium, iodine and the reconstructed VUE im-age. The two white rings in Fig. 3(a) contain a salinesolution of 5% Visipaque, an iodine-based contrast agent.It is clear from 3(e) that the algorithm cancels the effectof the contrast agent without impacting other regions sig-nificantly. In particular, the calcium region indicated byone of the areas is much better preserved in the methodhere proposed. Quantitative results are shown in Fig.4, which depicts boxplots of the signal intensity in theenhanced, virtual unenhanced, calcium, and saline-onlyregions. The boxplot clearly shows the ability of the al-gorithm to cancel the effect of the contrast agent whilepreserving the areas with calcium. Moreover, the concen-tration of Visipaque can be estimated from the volumefractions computed by the algorithm, yielding a value of5.88%, a good approximation to the true value of 5%.

In an experiment with real patient data, two vol-ume acquisitions were taken before and after injectionof the contrast agent. Figure 2 shows a sample contrast-

enhanced DECT image taken at 70 keV and its decom-position to fat, blood, bone, and contrast agent. Fig-ures 5(a) and 5(b) show the reconstructed VUE and thetrue unenhanced images, respectively. A good quali-tative agreement between the two images is observed.For quantitative comparison, five ROIs were selected asshown in Figure 5(b). Figure 5(c) compares the boxplotsof the attenuation coefficient in these ROIs from the con-trast enhanced (CE), reconstructed VUE, and true un-enhanced images. The boxplots also demonstrate a goodagreement between VUE and true unenhanced images(TUE), irrespective of the contrast agent density.

5. CONCLUSION

We proposed a novel approach for virtual unenhancedimage construction, based on multi-material decomposi-tion technique from dual-energy CT data. The algorithm

CEN VUE SALINE WATER CEN VUE WATER

-100

0

100

200

300

400

Line

ar A

ttenu

atio

n C

oeffi

cien

t (H

U)

Iodine Calcium

Fig. 4: Boxplot of data for contrast enhanced(CEN), virtual unenhanced(VUE), and true unen-hanced (Saline) regions. In iodine regions both the waterimage of an water/iodine pair and the VUE image we pro-pose have attenuation values in good agreement with those ofa TUE image. However, calcium regions are appropriatelyleft untouched by our algorithm, whereas the values of thewater image are excessively low.

810

(a) (b)

(c)

Fig. 5: Comparison of VUE against TUE for realpatient data. (a) Reconstructed VUE image, (b) true un-enhanced image, and (c) boxplots of data for contrast en-hanced (CE), virtual unenhanced (VUE), and true unena-haced (TUE) scans for the five ROIs specified in (b).

replaces the estimated volume of contrast in each voxelby the same volume of the material that the contrast hasdisplaced. We demonstrated the accuracy of the algo-rithm qualitatively and quantitatively on both phantomand anatomical data. Our results show that VUE imagescan replace true unenhanced images effectively, leadingto reduction in both radiation exposure and scan timeof CT exams and suggest a need for a change in currentpractices for the generation of VUE images.

References

[1] R. E. Alvarez and A. Macovski. Energy-selectivereconstructions in X-ray computerized tomography.Phys. Med. Biol., 21(5):733–744, Sept. 1976.

[2] E. J. Chae, J.-W. Song, J. B. Seo, B. Krauß, Y. M.Jang, and K.-S. Song. Clinical utility of dual-energyCT in the evaluation of solitary pulmonary nodules:Initial experience. Radiology, 249(2):671–681, Nov.2008.

[3] A. Graser, T. R. C. Johnson, E. M. Hecht, C. R.

Becker, C. Leidecker, M. Staehler, C. G. Stief,H. Hildebrandt, M. C. B. Godoy, M. E. Finn,F. Stepansky, M. F. Reiser, and M. Macari. Dual-energy CT in patients suspected of having renalmasses: Can virtual nonenhanced images replacetrue nonenhanced images? Radiology, 252(2):433–440, Aug. 2009.

[4] J. P. Heiken, J. A. Brink, B. L. McClennan, S. S.Sagel, H. P. Forman, and J. DiCroce. Dynamiccontrast-enhanced CT of the liver: Comparison ofcontrast medium injection rates and uniphasic andbiphasic injection protocols. Radiology, 187(2):327–331, May 1993.

[5] J. Hsieh, N. Chandra, D. Langan, M. S. Kulpins,X. Wu, and P. E. Licato. Dual-energy X-ray CTwith fast-kVp switch. In 51st AAPM Anual Meet-ing, Anaheim, CA, USA, July 2009.

[6] T. Johnson, B. Krauß, M. Sedlmair, M. Grasruck,H. Bruder, D. Morhard, C. Fink, S. Weckbach,M. Lenhard, B. Schmidt, T. Flohr, M. Reiser, andC. Becker. Material differentiation by dual en-ergy CT: Initial experience. European Radiology,17(6):1510–1517, June 2007.

[7] P. R. S. Mendonca, R. Bhotika, B. W. Thomsen,P. E. Licato, and M. C. Joshi. Multi-material de-composition of dual-energy CT data. In SPIE Med-ical Imaging, San Diego, CA, USA, Feb. 2010.

[8] E. B. Millard. Physical Chemistry for Colleges: ACourse of Instruction Based Upon the FundamentalLaws of Chemistry. International Chemical Series.McGraw-Hill, 1921.

[9] B. Schmidt and C. H. McCollough. Dual-energycomputed tomography. In T. C. Gerber, B. Kan-tor, and E. E. Williamson, editors, Computed To-mography of the Cardiovascular System, chapter 34,pages 451–462. Informa Healthcare, London, UK,Dec. 2007.

[10] N. Takahashi, R. P. Hartman, T. J. Vrtiska,A. Kawashima, A. N. Primak, O. P. Dzyubak, J. N.Mandrekar, J. G. Fletcher, and C. H. McCollough.Dual-energy CT iodine-subtraction virtual unen-hanced technique to detect urinary stones in aniodine-filled collecting system: A phantom study.Am. J. Roentgenol., 190(5):1169–1173, May 2008.

[11] X. Wu, D. A. Langan, D. Xu, T. M. Benson, J. D.Pack, A. M. Schmitz, J. E. Tkaczyk, J. Leverentz,and P. E. Licato. Monochromatic CT image repre-sentation via fast switching dual kVp. In E. Sameiand J. Hsieh, editors, SPIE Medical Imaging, vol-ume 7258, Lake Buena Vista, FL, USA, Feb. 2009.

811