An outlook on x-ray CT research and developmentGe Wanga� and Hengyong Yub�

Biomedical Imaging Division, VT-WFU School of Biomedical Engineering and Science, Virginia PolytechnicInstitute and State University, Blacksburg, Virginia 240601

Bruno De Manc�

CT and X-ray Laboratory, GE Global Research, Niskayuna, New York 12309

�Received 5 October 2007; revised 20 December 2007; accepted for publication 20 December 2007;published 25 February 2008�

Over the past decade, computed tomography �CT� theory, techniques and applications have under-gone a rapid development. Since CT is so practical and useful, undoubtedly CT technology willcontinue advancing biomedical and non-biomedical applications. In this outlook article, we shareour opinions on the research and development in this field, emphasizing 12 topics we expect to becritical in the next decade: analytic reconstruction, iterative reconstruction, local/interior reconstruc-tion, flat-panel based CT, dual-source CT, multi-source CT, novel scanning modes, energy-sensitiveCT, nano-CT, artifact reduction, modality fusion, and phase-contrast CT. We also sketch severalrepresentative biomedical applications. © 2008 American Association of Physicists in Medicine.�DOI: 10.1118/1.2836950�

Key words: analytic reconstruction, iterative reconstruction, local/interior reconstruction, flat-panelbased CT, dual-source CT, multi-source CT, scanning mode, energy-sensitive CT, nano-CT, artifact

reduction, modality fusion, phase-contrast CT

I. INTRODUCTION

The evolution of civilization has been largely driven by theneed for extending human’s capabilities. Arguably, the mostimportant way for us to sense the world is by visual percep-tion. However, the human vision is severely limited by theopaqueness of many natural and artificial objects. Hence,how to achieve “an inner vision” was often pondered in vari-ous scenarios through history. Mainly credited to the pio-neering works of Cormack1 and Hounsfield,2 in the last cen-tury, x-ray computed tomography �CT� is the first imagingmodality that allows accurate non-destructive interior imagereconstruction of an object from a sufficient number of x-rayprojections. Hounsfield’s x-ray CT prototype immediatelygenerated a tremendous excitement in the medical commu-nity and inspired a rapid technical development with an everstrong momentum. Also, x-ray CT as the first trans-axial to-mography model promoted the development of other tomog-raphic modalities for biomedical applications and beyond,such as magnetic resonance imaging, ultrasound tomography,nuclear tomography, optical tomography, molecular tomog-raphy, and so on.

Since its introduction in 1973,2 x-ray CT has revolution-ized radiographic imaging and become a cornerstone of ev-ery modern radiology department. Closely correlated to thedevelopment of x-ray CT, the research for higher perfor-mance system architectures and more advanced image recon-struction algorithms has been intensively pursued for impor-tant biomedical applications. A famous CT scanner is thedynamic spatial reconstructor �DSR� built at the Mayo Clinicin 1979.3,4 With the ability to acquire 240 contiguous 1 mmslices in a time window of 0.01 s, the heart, lung, and bloodflow can be vividly observed. This DSR system is considered

as the precursor of the electron-beam CT scanner, the dual-

1051 Med. Phys. 35 „3…, March 2008 0094-2405/2008/35„3…

source CT scanner and other similar systems. Early 1990s,single-slice helical/spiral CT became the standard scanningmodel,5 and helical/spiral cone-beam CT was proposed.6,7

Elscint developed a two-slice CT scanner in the mid 90’s andGE came out with the first four-slice CT scanner a few yearslater, quickly followed by all other major CTmanufacturers.8,9 In 2004, Toshiba first successfully devel-oped 256-slice CT systems.10 With the fast evolution of thetechnology, true volumetric cone-beam CT scanners in heli-cal and other scanning modes are emerging as the next gen-eration biomedical CT.11,12 On the algorithm side, over thepast several years, triggered by an exact and efficient solu-tion to the long-standing helical cone-beam CT problem,13,14

there has been a remarkable surge in studies on image recon-struction algorithms.15 Now, there are already a number ofexact reconstruction schemes and algorithms that deal withgeneral scanning trajectories and appropriately truncateddata.

Will the progress of this magnitude in the CT field con-tinue in the next decade? Based on the impressive trackrecord in this field, we firmly believe it will. Then, what arethe most important research topics/directions for the x-rayCT research and development? The answer to this big ques-tion is difficult to formulate because any prediction of thatnature must be limited by our partial knowledge and inad-equate ability to look into the future. Having acknowledgedthat, in this article we would like to share our opinionsfor the purpose of information and inspiration, whichwould hopefully guide our efforts and facilitate the futureadvancement.

This outlook article is organized as follows. In Sec. II, wewill quantitatively overview the CT literature to give a sense

of the dynamics of this field. In Sec. III as the main part of

1051/1051/14/$23.00 © 2008 Am. Assoc. Phys. Med.

1052 Wang, Yu, and De Man: CT outlook 1052

this article, we will present 12 topics to cover the maintrends in the CT field. In Sec. IV, we will discuss biomedicalimplications of the proposed research and development. Fi-nally, in Sec. V we will make concluding remarks.

II. QUANTITATIVE LITERATURE ANALYSIS

Our methodology for literature analysis is to search theISI Web of Knowledge �http://portal.isiknowledge.com�: Sci-ence Citation Index Expanded �SCI-EXPANDED� from1975 until now �search performed on September 4, 2007�.Each topic search is conveniently performed with one ormore terms within article titles, keywords, and abstracts. Ouranalysis was intended to give a quantitative impression but isnecessarily selective and by no means exhaustive. A readercan easily verify our data, modify our searches and reachhis/her own conclusions.

Does the CT research have an increasing momentum?The answer is definitely “yes,” despite the great develop-ments in all competing modalities. With the search rule “x-ray” and “CT,” there are 5857 hits for 1975–2007, of whichwe have only 1210 hits for 1987–1996 and 4553 hits for thepast ten years 1997–2007. With the search rule “Cone-beam”and “CT,” we have 896 hits for 1975–2007, of which wehave only 117 hits for 1992–2001, and 777 hits for the pastfive years 2002–2007. More detailed yearly numbers of theretrieved articles are shown in Fig. 1. Some further represen-tative search results over the period 1997–2007 are summa-rized in Table I. Evidently, the above data are quite informa-tive, indicating steady and increasing efforts for the past tenyears over a wide spectrum of CT research, development andapplications. However, there should be some noise and arti-facts in the results since the keywords and their combinationscould be misleading in some cases. Therefore, these datashould not be interpreted without caution.

We then retrieved all 363 “x-ray” and “CT” papers for theyear 2007, excluded irrelevant ones, and categorized them,as shown in Table II. Clearly, the biomedical applicationsremain the mainstream for x-ray CT, and a major portion ofefforts is being devoted to research and development ofmethods and systems. Also, there are some non-biomedicalapplications, which are about one third of the medical appli-cations.

FIG. 1. Numbers of retrieved papers during 1997–2007 with the searchingrules �a� “x-ray” and “CT” and �b� “cone-beam” and “CT,” respectively.Exponential fitting shows that there is a 5% increment yearly for the former,while a 39% increment for the latter. Note that the numbers for 2007 are

incomplete.

Medical Physics, Vol. 35, No. 3, March 2008

III. TWELVE TOPICS FOR THE NEXT DECADE

III.A. Analytic reconstruction

Since Katsevich’s 2002 paper on exact and efficient heli-cal cone-beam CT reconstruction,14 intensified research ef-forts have been made in this area.15 Various sophisticatedformulas have been proposed for exact reconstruction fromprojection data that can be longitudinally truncated and eventransversely truncated, and for not only a standard helicaltrajectory but also a quite general class of scanning loci �Fig.2�.16–24 However, in the general case such as saddle-likescanning paths24 and nonstandard helical trajectories, thesealgorithms are far less computationally efficient as comparedto the popular filtered backprojection �FBP� with spatiallyinvariant filtering. In the 9th International Meeting on Fully3D Image Reconstruction in Radiology and Nuclear Medi-cine �Lindau, Germany, July 9–13, 2007�, Katsevich pre-sented an important progress towards exact yet efficient gen-eral cone-beam reconstruction algorithms for two classes ofscanning loci.25 The first class curves are smooth and of posi-tive curvature and torsion. The second class consists ofcircle-plus curves, with the segment for the plus part startingbelow the circle-like trajectory and ending above it.26 How-ever, there are other important classes of trajectories forwhich exact and efficient algorithms are desired and highlynontrivial. Therefore, it remains an open challenge to formu-late such exact and efficient algorithms for many other typesof scanning curves. Theoretical unification of these new ex-act algorithms is also worthwhile.27–29

TABLE I. Representative search data on CT studies and application, wherethe percentage means the number of hits for the past five years over that forthe past ten years �searched on 9/4/07�.

RuleNo. of Hitsfor 97-07

No. of Hitsfor 02-07 %

“CT” and “algorithm�” 2785 1829 65.7“CT” and “cancer” 11 468 7642 66.6“CT” and “cardiac” 2252 1517 67.4“CT” and “dose” 5595 3609 64.5“CT” and “dynamic” 2572 1456 56.7“CT” and “dual-energy” 235 142 60.4“CT” and “screen�” 2771 1812 65.4“CT” and “therap�” 12 099 7296 60.3“CT” and “diagno�” 27 025 15604 57.7“CT” and “fusion” 1586 1134 71.5“CT” and “mouse” 787 416 52.9

TABLE II. Categorization of 262 relevant papers selected from the 363 ar-ticles in English retrieved under the rule “x-ray” and “CT” for 2007�searched on 9/4/07�.

Category No. of Papers %

Methods & systems 72 27.5Medical applications 127 48.5Pre-clinical applications 14 5.3Non-biomedical applications 49 18.7

1053 Wang, Yu, and De Man: CT outlook 1053

Furthermore, it has been well recognized that the way toformulate exact and efficient reconstruction is notunique.17,27,30 The key is to select an appropriate weightfunction. Different choices of the weights would have dis-tinct impacts on the image noise distribution. Depending oneach specific clinical application, the preferred image noisedistribution may be uniform or least noisy in a region ofinterest �ROI� while being more tolerable outside the ROI. Inother words, an ideal exact cone-beam reconstruction for-mula needs to be not only efficient but also produce the bestpossible image noise distribution, which is application de-pendent. The governing theory guiding this procedure hasbeen studied from different aspects but it has not been thor-oughly developed yet.

In contrast to impressive progress in solving the so-calledlong objection problem �reconstruction of a long object fromlongitudinally truncated cone-beam data�, cardiac cone-beamCT is what we call the quasi-short object problem �recon-struction of a short portion of a long object from longitudi-nally truncated cone-beam data involving the short object�and deserves more research efforts. To solve the quasi-shortobject problem, the circular cone-beam scan only permitsapproximate reconstruction and the helical cone-beam scanhas inefficient photon utilization. The saddle-like cone-beamscan can combine exact reconstruction and good photon uti-lization, representing a promising direction.24,31,32 Again, wewill need exact, efficient and optimized algorithms in thecontext of solving the quasi-short object problem.

In the CT applications, approximate cone-beam recon-struction algorithms have been dominating despite the rapiddevelopment of exact cone-beam reconstruction algorithms.6

FIG. 2. Representative general scanning trajectories. �a� A non-standard he-lix, �b� a saddle curve and �c� a circle-plus-line combination.

Medical Physics, Vol. 35, No. 3, March 2008

There are two reasons. First, in a good number of applica-tions, the data completeness required for exact reconstructioncannot be satisfied due to physical constraints. Hence, theonly choice is to perform an approximate reconstruction.Second, even if an exact reconstruction is practically fea-sible, oftentimes the approximate algorithms can deliversimilar or even better performance as compared to the exactreconstruction counterparts, which are not necessarily opti-mal in terms of the noise characteristics. Figure 3 shows twoapproximate reconstructions of excellent quality.33

It is our opinion that the popularity of approximate cone-beam reconstruction algorithms will definitely sustain in thenear future. With the development of exact cone-beam recon-struction algorithms, the approximate cone-beam reconstruc-tion algorithms will be surely improved as well. In additionto various approximations to be improved or made on anindividual basis, a promising research direction for approxi-mate cone-beam reconstruction would be to adapt new exactcone-beam reconstruction algorithms.34–36 By doing so, themerits of exact cone-beam reconstruction algorithms can belargely inherited at a much lower computational cost andmodified to have certain practically desirable features. Thisapproach should be particularly attractive in the cases of in-complete and/or inconsistent data such as cardiac CT, con-trast studies, artifact reduction, and so on.

III.B. Iterative reconstruction

While the very first CT scanners used algebraic iterativereconstruction �ART�,37,38 FBP39 soon became the gold stan-dard for CT reconstruction. A few years later, statistical it-erative reconstruction was successfully introduced for emis-sion tomography,40 because FBP reconstruction from a lowSNR �signal-noise-ratio� emission data set would producequite poor image quality. A simultaneous update variation ofART, called simultaneous ART, was presented in 1984.38 Inthe past decade, thanks to the increasing computationalpower, statistical iterative reconstruction has become a hotresearch topic for CT,41–46 with a focus on noise suppression,artifact reduction and dual energy/energy-sensitiveimaging.46,47,92 Recent advances in statistical iterativereconstruction48 promise to achieve dramatic improvementsin image quality, as illustrated in Fig. 4. Figure 4 shows the

FIG. 3. Images reconstructed from clinical data usingan approximate helical cone-beam FBP algorithm �seeRef. 33�. �a� A reformatted abdominal image �displaywindow width/length: 381/83�HU�� and �b� a reformat-ted cardiac image �display window setting: 800/216�HU�� duplicated from Ref. 33 with the permissionof the Institute of Physics Publishing.

1054 Wang, Yu, and De Man: CT outlook 1054

impact of a 4x reduction in x-ray tube current �mA�. WithFBP, the noise level increases dramatically �the standard de-viation of the noise increases by a factor of 2 theoretically�.With iterative reconstruction, the noise level is not increasedor even descreased relative to the full-dose FBP counterpart,suggesting a 4x dose reduction in this particular case. Newiterative reconstruction schemes will surely emerge, as willbe further discussed for Topics J and K. We believe that it isonly a matter of time for statistical iterative reconstruction tobecome available on commercial CT scanners.

Several factors have prevented statistical iterative recon-struction from being deployed on CT products so far. Thefirst is the huge computational cost, about two to three ordersof magnitude larger than that of FBP. Despite recent ad-vances in high-performance image reconstruction based onfield-programmable gate arrays, multi-core processors suchas IBM’s Playstation 3 Cell Broadband Engine, and graphicsprocessing units, we expect that first product realizations willbe slow �tens of seconds per slice�. The second is the robust-ness of iterative algorithms across different protocols andapplications. Currently, some significant parameter tuningneeds to be done for practical use of an iterative method in aspecific application. The best imaging performance is usuallyvery sensitive to the particular choice of parameters. Thethird is the potentially new look of statistically reconstructedimages. Since images are reconstructed in a statistically op-timal fashion, their noise and artifact characteristics can berather different from those of FBP. Radiologists are used tothe FBP image appearance and associated various imagequality trade-offs. Consequently, statistical reconstructionmay give an impression of somewhat reduced diagnosticvalue in certain cases. In fact, its spatial resolution charac-teristics are indeed very different from the FBP counterpartand depend on many factors. A careful quantitative analysisof image noise and spatial resolution as a function of loca-tion, contrast, and measurement statistics shows indeed thatstatistical iterative reconstruction has the potential to im-prove the information content of reconstructed images.49

In perspective, iterative reconstruction offers distinct ad-vantages relative to analytic reconstruction in important

FIG. 4. Images analytically and iteratively reconstructed from clinical data. �a 75 mAs dataset, and �c� an iterative reconstruction from the same 75 mA

cases when data are incomplete, inconsistent and rather

Medical Physics, Vol. 35, No. 3, March 2008

noisy. Even in the cases where analytic reconstruction per-forms well, there is no fundamental reason why iterative re-construction would perform any worse. Hence, we predictthat the fast development in computing methods and hard-ware will lead to a shift from analytic to statistical recon-struction or at least a fusion of analytic and iterative ap-proaches. This evolution will happen gradually, because ofthe high computational cost as well as the time needed forthe radiologists to adopt this new technology.

III.C. Local/interior CT

There is a long track record for the exact reconstruction ofan ROI from a minimum dataset, starting from the work onhalf-scan fan-beam reconstruction.50 The benefits includeshorter data acquisition time, less radiation dose and moreimaging flexibility. The most remarkable recent finding is thework by Katsevich in 2002 that demonstrates the feasibilityof the exact regional reconstruction within a long object fromlongitudinally truncated data collected along a PI ���-turn ofa helical scanning trajectory.14 The subsequent backprojec-tion filtration variant51 and generalization into the quite arbi-trary scanning case16,18 significantly enriched the local CTtheory. As illustrated in Fig. 5, the latest results along thisdirection are the one-sided inverse Hilbert transform basedlocal reconstruction algorithm52 and the truly truncated Hil-bert transform based interior reconstruction algorithmcoupled with some a priori knowledge in an ROI to bereconstructed,53,54 which is somehow challenging the con-ventional wisdom that an interior problem �reconstruction ofan interior region from projection data along lines throughthe region� does not have a unique solution.55 Clearly, exactsolutions to the interior problem have tremendous implica-tions for numerous biomedical and other applications.56

It is highly desirable to establish a unified theory for exactlocal/interior reconstruction that covers the exact reconstruc-tion schemes and methods from a minimum dataset in thetwo-dimensional �2D� and three-dimensional �3D� casesfrom parallel- and divergent-beam geometries. It should bevaluable to refine the reported stability analysis and reflect

FBP reconstruction from a 300 mAs dataset, �b� a FBP reconstruction fromset �Ref. 48� �Courtesy of J.-B. Thibault�.

a� As data

the sampling geometry and data noise in an optimal fashion.

1055 Wang, Yu, and De Man: CT outlook 1055

Also, it is critical to design numerically stable, robust andefficient algorithms for this purpose. The state-of-the-artframework for local/interior reconstruction is the Hilberttransform analysis. Perhaps other possibilities exist for us togain a thorough understanding of this amazing problem.

As an alternative technology for local/interior reconstruc-tion, theoretically exact fan-beam lambda tomography �LT�was also available for a general scanning locus based on the2D Calderon operator.57,58 However, exact cone-beam LT isgenerally impossible based on the 3D Calderon operator forpractical scanning loci such as a helix.58 It is promising todevelop exact LT reconstruction methods in terms of the 2DCalderon operator in some planes for data collected along ageneral scanning locus. As a related area, partial derivativesof a dynamically changing image volume can be directlyreconstructed from original cone-beam data. Thus, velocitytomography may be improved from these partial derivativesbased on the general velocity field constraint equation.59

III.D. Flat-panel based CT

While clinical x-ray systems are conventionally limited toprojection images, digital, solid-state flat-panel x-ray detec-tors �FPDs� have enabled 3D reconstruction in these modali-ties. In 1999, GE Healthcare first introduced FPD-basedx-ray systems, in which FPDs replaced film in radiographyand mammography. Then, GE introduced an FPD-based car-diology system, replacing the conventional image intensifier

FIG. 5. Theoretically exact solution to the interior problem. The exactlyrecoverable regions are not the same according to the different sufficientconditions-based on the truncated Hilbert Transform: �a� The exact recon-struction region by Defrise et al. �Ref. 52�, �b� that by Ye et al. �Ref. 53�where f�x� is the object image along a given PI line within a compactsupport and g�x� is the corresponding Hilbert Transform, �c� a slice of thethorax phantom with a T-shaped local ROI, and �d� the image reconstructedfrom a purely local dataset plus prior knowledge on a sub-region in the ROIusing the POCS method �Ref. 53�.

�II� tube. Today, all of the major diagnostic imaging system

Medical Physics, Vol. 35, No. 3, March 2008

manufacturers �GE, Hologic, Philips, Siemens, and Toshiba�offer FPD-based x-ray systems. GE, Hologic, Canon andShimadzu make FPDs independently. FPDs in use today em-ploy pixel sizes ranging from 70�70 �m2 to 200�200 �m2, in arrays of up to 14 million pixels, and activeareas of up to 43�43 cm2. Some FPDs are of the indirectconversion type, in which a scintillator converts the x rays tovisible light, and a 2D array of photodiodes converts the lightto electronic charges to be digitized. In an alternative directconversion technology, x-rays are converted directly to elec-tronic charges and then digitized.

Even prior to the development of FPDs, 3D reconstruc-tion of multi-angle digital projection images from an II-based C-arm cardiology system was demonstrated.60 OnceFPDs were incorporated into such systems, the non-idealcharacteristics of the II tube were circumvented, and CT re-construction of data from C-arm systems became practical.Today most clinical FPD-based C-arm systems have the abil-ity to acquire and reconstruct 3D CT images. Digital BreastTomosynthesis61 is a new modality using key elements ofFPD-based mammography systems. The x-ray source ismoved along an arc and a relatively small number of projec-tions are acquired, enabling reconstruction of approximately1.0-mm-thick image planes. This technology is currently inadvanced development. Researchers have also incorporatedFPDs into a conventional CT gantry to achieve high spatialresolution in medium to large field-of-view scans.62 This ar-chitecture was primarily intended for small animalimaging.63,64 FPDs are also under investigation in a fewother application-specific areas, such as breast CTscanners,65,66 infra-operative imaging systems,67 and com-pact or mobile CT systems for head or dental imaging.68

FPD-based CT is still in its infancy, and substantial develop-ments are expected in the future.

III.E. Dual-source CT

Most commercial clinical scanners until today are basedon the so-called third-generation architecture. A single x-raytube and a single detector assembly are positioned face toface and rotated jointly around the patient. Berninger andRedington69 presented the idea of replicating the source anddetector, as illustrated in Fig. 6�a�. The dynamic spatial re-constructor may have been the first real multi-sourceprototype.4 About 25 years later, Siemens developed the firstcommercial two-tube-two-detector scanner, also known asdual-source CT.70 The main advantage of this architecture isits improved temporal resolution. In today’s state-of-the-artCT scanners, the gantry rotation time is reduced to about0.35 s, but it is mechanically challenging to reduce that timeeven further, which justifies the renewed interest in multi-source architectures. For the dual-source CT system, theminimum rotation interval is 90° +�, compared to 180° +�for a third-generation scanner. For a cardiac fan angle � of25°, this gives about 44% reduction in acquisition time.

The dual-source CT has been claimed to have a dose ben-efit because the second detector is smaller and hence the dose

is reduced at the edge of the field of view. However, the

1056 Wang, Yu, and De Man: CT outlook 1056

exact same dose profile is achieved with a third-generationarchitecture and a more aggressive bowtie. A more fair com-parison would compare patient dose at an identical noiselevel throughout the field of view. In fact, the cross scatterfrom the second source into the first detector is reported toresult in a scatter-to-primary ratio as high as 100% for obesepatients,71 corresponding to a severe dose penalty. Therefore,more research will be needed on both hardware �scatter re-jection� and software �scatter correction� methods for dual-source CT.

A next step in this area is to use three sources and threedetectors.72–74 For a cardiac fan angle � of 25°, a triple-source system with symmetric spacing would result in 59%reduction in acquisition time. In terms of exact cone-beamreconstruction, unlike the dual-source cone-beam geometrythe triple-source cone-beam scanner allows a perfect mosaicpattern of truncated cone-beam data to satisfy the Orlovcondition,75 as shown in Fig. 6�b�. In the near future, manu-facturers may or may not adopt the multi-source-multi-detector strategy, depending on their ability to conquer tem-poral resolution with other means, such as faster rotation,virtual rotation �see next sub-section�, or motion compensa-tion techniques.76 Triple-source CT has unique theoreticalmerits relative to dual-source CT. However, the brute-forcehardware approach of dual-source and triple-source CTcomes with important technical and cost trade-offs, andmight limit them to be niche cardiac scanners, just likeelectron-beam CT �EBCT�.

III.F. Multi-source CT

Almost all modern CT architectures are based on one ormore single-spot x-ray tubes �possibly with focal spotwobble�. The best known counter example is the EBCTscanner,77 which uses a huge x-ray source surrounding thepatient with an electron beam that is deflected to producex-rays along four half-circle target rings. The distributed na-ture of the x-ray source makes it possible to rotate the focalspot freely, thereby eliminating any mechanical bottlenecks.This architecture is fairly complex, limited in source inten-sity, and not so compatible with true volumetric scanning,

but it is extremely fast and therefore it is very useful for

Medical Physics, Vol. 35, No. 3, March 2008

cardiac scanning. A miniature version of the e-beam CTscanner was also conceptualized for micro-CT cardiacstudies.78 Most of the drawbacks of EBCT are eliminated bygoing to a distributed source with discrete electron emittersand appropriate source-detector topologies.79 As shown inFig. 7�a�, another class of x-ray sources has multiple spotsdistributed along the longitudinal axis �in the z direction�.80

It has been shown that the combined information from themultiple spots in the z direction effectively eliminates cone-beam artifacts.81

Another example of a distributed x-ray source with a de-flected e beam is the transmission x-ray source developed byNovaRay �formerly known as Cardiac Mariners and Nexray,Palo Alto, California, USA�, with thousands of focal spots.This area source was first used to demonstrate the concept ofinverse-geometry CT.82 In addition to eliminating cone-beamartifacts, this architecture also has the benefit of a smallphoton-counting detector and very good detection efficiency.A related architecture is based on discrete electron emitters,83

resulting in a 2D array source with tens of focal spots �Fig.7�b��. This source architecture is more compact than theabove and perhaps more compatible with the concept of avirtual bowtie, where the operation of each spot is modulatedin real time to optimize image quality and minimize dose,depending on the patient anatomy. In the past decades, ad-vances in detector technologies defined the so-called “slicewars.” We expect that in the next decade dramatic advancesin distributed x-ray sources may define a new revolution inCT and give birth to a wide class of new multi-source CTarchitectures, including line sources, inverse-geometry CT,and ultimately a rebirth of stationary CT, which means thatneither the source nor the detector is in motion during thedata acquisition process.

III.G. Novel scanning modes

A very important trend in the CT field is the evolution inscanning modes for improved imaging performance at re-duced patient dose. The introduction of single-/multi-detector-row/cone-beam helical CT was only the first steptowards fully volumetric CT. In practice, a number of more

FIG. 6. CT architecture with three sources and threedetectors. �a� Berninger and Redington first proposed toreplicate the x-ray tube and detector assembly �Ref. 69�and �b� the triple-source system in the helical cone-beam geometry allows a perfect data mosaic pattern forexact reconstruction. Figure 6�a� is duplicated fromRef. 69 with the permission of GE Company.

flexible scanning schemes are needed. An earliest example of

1057 Wang, Yu, and De Man: CT outlook 1057

a dynamic helical pitch trajectory is bolus-chasing CTangiography.84,85 In this application the table speed is ad-justed on the fly to synchronize the CT imaging aperture withthe contrast bolus peak. Also, the so-called shuttle mode is ascanning scheme where the patient table travels back andforth, enabling better dynamic applications than with con-ventional scanning, including liver perfusion and brain per-fusion. These applications are based on cone-beam recon-struction algorithms specifically designed for dynamichelical pitch source trajectories.86 Another example is a com-bined cardiac and thorax scan, where the table speed is re-duced in the cardiac region for phase-coherent imaging, andincreased as soon as the scan aperture is past the cardiacregion. Finally, the latest efforts on saddle-like cone-beamscanning seem also promising for exact solution of a quasi-short object problem �reconstruction of a short portion of along object from longitudinally truncated cone-beam data in-volving the short object�.24,31 Especially, the composite-circling scanning mode �Fig. 8� has some advantages overthe existing scanning modes from perspectives of both engi-neering implementation and clinical applications because ofits symmetry in mechanical rotation and the compatibilitywith the physiological conditions.24,31

To minimize patient dose in CT, it is critical to detectevery transmitted photon and to treat it in a statistically op-timal fashion during reconstruction. It is equally important tooptimize the flux profile through the patient based on thepatient attenuation profile and the organ sensitivities. A firstway to achieve this is to pre-attenuate the x-rays with a so-called bowtie filter. Bowtie filters are becoming more ad-vanced in targeting regions of interest, sometimes even usingone or more moving parts.87 A second way is to modulate thetube current �and possibly other scan parameters such as kVpor filtration� as a function of rotation angle, longitudinal po-

88

FIG. 7. Novel source configurations. �a� A distributed source with multiple sinverse-geometry architecture with tens of spots combining the benefits of aFigure 7�a� is duplicated from Ref. 80 with the permission of GE Company

sition, and cardiac phase. A major advantage of an inverse-

Medical Physics, Vol. 35, No. 3, March 2008

geometry architecture with an array source is the ultimateflexibility to customize the flux profile with the so-calledvirtual bowtie.83

Key to all the above methods is knowledge about thepatient anatomy. This knowledge may be based on scoutscans, patient data or atlases.89 One can take huge advantageof this knowledge not only to perform simple tasks such as

along the z direction to eliminate cone-beam artifacts �Ref. 80�, and �b� anource in z, a small photon-counting detector, and a virtual bowtie �Ref. 83�.

FIG. 8. Compositing-circling scanning mode. In such a CT system, the scan-ning trajectory is a composition of two circular motions: While an x-rayfocal spot is rotated on a plane facing an object to be reconstructed, thex-ray source is also rotated around the object on the gantry plane. Once aprojection dataset is acquired, exact or approximate reconstruction can bedone �Copyright by G. Wang and H. Y. Yu, US Provisional Patent Applica-

potsline s.

tion, 2007�.

1058 Wang, Yu, and De Man: CT outlook 1058

bowtie selection and patient centering but also ultimately tocompute optimal pulse sequences, avoid sensitive organs,and perform true ROI imaging.

III.H. Energy-sensitive CT

The idea of decomposing the linear attenuation coefficientinto two or more basis functions in order to eliminate beamhardening artifacts for material discrimination was proposedsoon after the invention of CT.90 The physical foundation isthat the linear attenuation coefficients are a function of en-ergy and that this energy dependence is different for differenttissues and elements. Figure 9 also illustrates that variouselements have different attenuation properties as functions ofphoton energy.91 Previously, two measurements were ac-quired at different tube kVp settings, also called dual-energyCT. It was then realized that iterative reconstruction caneliminate the need for a second measurement by incorporat-ing prior information.43,46,92 Technological advances are nowmaking it possible to acquire two or more measurementssimultaneously at different effective x-ray energies. Thesemethods can be primarily evaluated by signal-to-noise ratioand measurement registration. The process of material de-composition results in a noise amplification. Noise can beminimized if the effective energies of the measurements arewell separated, and better algorithms are designed.

Dual-layer detectors are fairly mature93,94 and the corre-sponding measurements are perfectly registered but of allsolutions have the poorest energy separation. Dual-sourceCT is an existing technology,70 its energy separation is betterthan dual-layer detectors but there is a significant time mis-registration �about 100 ms� between corresponding measure-ments, not suitable for dynamic applications. Fast kVpswitching is an existing technology,95 with better energy

FIG. 9. Energy dependence of the linear attenuation coefficient varies acrossdifferent elements and tissues. The plot shows the linear attenuation coeffi-cients for bone �continuous curve� and iodine �curve with a discontinuity atthe K edge�. While for the shown iodine concentration the average attenu-ation is very similar to that of bone; their energy dependence is verydifferent.

separation than the dual-layer detector, and very recently it

Medical Physics, Vol. 35, No. 3, March 2008

has been shown that its time misregistration can be as goodas one or a few views ��1 ms�.96 Advances in compactmonochromatic x-ray sources may result in improved energyselectivity, but this solution does not address the time mis-registration. The ultimate architecture for energy-sensitiveCT may use an advanced polychromatic source to cover aspectrum of interest and a photon-counting-based energy-discriminating detector to record all these polychromaticphotons and sort them into respective spectral bins.97 Whilethis solution would give perfect energy separation and regis-tration, the technology is still immature and suffers fromsevere count rate limitations. Along with dedicated hardwaredevelopment, algorithm development, especially statisticalmethods, will definitely facilitate this type of application.The fact that all manufacturers are pursuing different ap-proaches makes the area of energy-sensitive CT an extremelyinteresting area. None can change the laws of physics butenormous efforts in the coming years will show how dra-matic advances in CT source, detector, and reconstructiontechnologies will help us all to approach the fundamentalperformance limits of energy-sensitive CT.

III.I. Nano-CT

The world’s first and only sub-100 nm nano-CT scannernanoXCT™98 was recently developed by the Xradia company�Concord, CA�. This system is a revolutionary microscopefor non-invasive investigations involving semiconductoranalysis, drug discovery, molecular imaging, stem-cell re-search and materials development. It allows 50 nm reso-lution using proprietary condenser and objective optics. Formost samples in nanotechnology, the x-ray attenuation lengthfor low Z materials is very long, resulting in poor imagecontrast. The Xradia system can significantly increase imagecontrast in the Zernike phase contrast mode.

Since this century, nanotechnology has gained tremendousmomentum through both governmental and private invest-ments. Clearly, nano-CT may be a strategic enabling compo-nent for the immediate future research and education. In abroad range of nano-CT applications, interior tomography isnot only valuable but also necessary. For example, in thecase of in-situ imaging of cells or tissue specimens at thecellular/molecular level, we require little morphologicalchanges and minimum radiation exposure, have the water orair component as reference and a volume of interest �VOI�much smaller than the specimen. Since the recent theoreticaland numerical results53,54,56,99 demonstrated that the interiorproblem can be solved in a theoretically exact and stablefashion assuming that a small subregion within the interiorregion is known, it becomes now feasible to meet the above-mentioned interior reconstruction need for nano-CT.

In the next decade, we believe that the existing nano-CTscanners will be further advanced, unique reconstruction al-gorithms will be developed with multi-scale and interior im-aging capabilities, and more nano-CT applications will beidentified in the fields including but not limited to life sci-ence, preclinical and clinical imaging studies in vitro, phar-

maceutical research, and so on.

1059 Wang, Yu, and De Man: CT outlook 1059

III.J. Artifact reduction

Reduction of image artifacts has been a central topic inthe CT field.100 The paradigm shift towards volumetric CT,novel architectures and dynamic and quantitative imagingdemands a more effective reduction of various artifacts. Thewell-known scattering artifacts become more and more seri-ous with the increasing cone angle and dual-source CT con-figurations. Beam hardening artifacts must be suppressed toextract energy-dependent information.101 Motion artifacts re-main a main challenge for cardiac CT and contrast-enhancedstudies.102,103 More than a dozen types of artifacts are wellknown to the field �Fig. 10�, most of which assume newforms and present new problems associated with the devel-opment of new CT technologies. Therefore, the fight againstthese artifacts remains active and requires new tools.

While many traditional artifact reduction algorithms aread hoc and approximate, the future efforts may rely on moresolid physical models and more rigorous inherent dataintegrity.101,104,105 In this context, data consistency conditionswere proposed to suppress motion artifacts, minimize beamhardening and so on. The invariance of the integral was sug-gested as a possible mechanism for this purpose. Neverthe-less, much more work is required to advance this frontier. Toa large degree, artifact reduction is very similar to imagereconstruction. In both cases, the goal is to find an optimumsubject to a set of constraints. Given the complexities im-posed by the artifact-related constraints, the iterative ap-proach will play a more important role. Several iterativeschemes have been well studied so far.106–108 New iterativeschemes deserve major attention and refinement, such as thealternating iteration scheme for metal artifact reduction.109

Only with optimized reduction of various artifacts, the futureCT technology will deliver its ultimate performance thatshould be spatially, dynamically, spectrally and quantita-

FIG. 10. Representative CT image artifacts. �a� A shoulder phantom imaggenerating metal artifacts, and �c� a moving head leading to motion artifactNorth America�.

tively correct.

Medical Physics, Vol. 35, No. 3, March 2008

III.K. Modality fusion

A distinguished trend in modern biomedical imaging isthe area of multi-modal imaging, in which two or more im-aging systems are synergistically integrated for much betterperformance, improving or enabling biomedical applications.A primary example is the positive emission tomography�PET�/CT systems.110 Another example is the hybrid opticaltomography systems such as those proposed for magneticresonance imaging �MRI�-based diffuse opticaltomography,111 CT/MRI integration,112 and CT-based biolu-minescence tomography.113 Recently, the Siemens micro-CT-PET-single photon photo emission computed tomography�SPECT� system Inveon �Fig. 11� became commerciallyavailable as an integrated preclinical imaging platform. Fromthe perspective of x-ray CT research and development, anunprecedented potential would be unlocked by identifyingnew combinations of complementary imaging modes and im-proving the existing multi-modal systems, such as hybridCT-angio and CT-cardiac cath systems. There are good pos-sibilities for one-stop imaging centers or suites to emergewhere all the imaging tasks can be streamlined in a task-specific fashion, which is in some sense an extension of thecurrently already available trail based multi-modal small ani-mal imager.

In addition to the architectural issues, we emphasize thatthere are excellent opportunities for algorithm developmentin this area. Traditionally, each component modality of afusion-based system can be independently considered for im-age reconstruction. Then, all reconstructed images are retro-spectively combined via post-processing for further analysis.However, there is generally some or strong correlationamong the datasets acquired by multiple imaging tools ap-plied to study the same individual object. Ideally, we shouldnot solve the imaging problems for these modalities sepa-

th streaks caused by photon starvation, �b� a patient with spine implantsplicated from Ref. 100 with the permission of the Radiological Society of

e wis �Du

rately but couple these imaging problems with implicit or

1060 Wang, Yu, and De Man: CT outlook 1060

explicit relationships describing dependence among the in-volved datasets. Such an integrated inverse problem may re-quire an iterative solution containing several loops each ofwhich assumes other image reconstructions known and re-fines an intermediate image, or have more sophisticatedforms like a truly simultaneous iterative solution.114 Theoret-ical studies are needed to establish the solution existence,uniqueness and stability with new iterative reconstructionschemes, as already mentioned for Topics B and J. In thecases of no unique solutions, regularization issues must beaddressed with the aid of a priori knowledge in the form ofconstraints, penalty terms and so on.

III.L. Phase-contrast CT

In all mainstream x-ray CT imaging modalities, the con-trast mechanism has been attenuation based for over a cen-tury. As a result, weak-absorbing tissues are not imaged well,and radiation dose has been a general concern. On the otherhand, x-ray phase-contrast imaging relies on the diffractionproperties of structures, promising to have much better con-trast at much lower dose. Specifically, for x-rays the refrac-tion index of biological soft tissues is approximately 1.0,namely, n=1−�+ i�, where � and � represent attenuationand diffraction, respectively. In the range of 15–150 KeV, �is about three orders of magnitude greater than �, makingphase-contrast imaging about three orders of magnitudemore sensitive than attenuation-based methods.115 Typicalvalues of refraction indexes of breast tissue are plotted in

FIG. 11. Imaging platform Inveon for fusion of CT, PET and SPECT inpreclinical applications �Courtesy of Siemens Company�.

Ref. 116.

Medical Physics, Vol. 35, No. 3, March 2008

X-ray phase-contrast imaging has been a hot topic overthe past decade.117–120 It is traditionally implemented via in-terferometry, diffractometry and in-line holography, respec-tively. Interferometry and diffractometry are restricted by theavailability of a synchrotron radiation facility. The in-lineholography approach was originally proposed by Gabor in1948, for which he was awarded the Nobel Prize. Wilkins etal. developed in-line holography techniques with a micro-focus polychromatic x-ray source.120 However, a micro-focus ��100 �m� x-ray tube offers limited x-ray flux. In2006, Pfeiffer et al. made a major improvement so that sucha phase-contrast imaging system can be built on a hospital-grade x-ray tube.121,122 As shown in Fig. 12, the key idea isto utilize the Talbot effect generated by x-ray gratings forsensing waveform deformation. It is anticipated that moreresearch efforts will be devoted to x-ray phase-contrast im-aging including x-ray phase-contrast tomography using themicro-focus or the grating-based framework.123 The forwardmodel will be a theoretically and computationally challeng-ing problem, particularly for relatively large specimen. Ac-cordingly, the development of phase-contrast reconstructionalgorithms will become an area of intense research.

IV. BIOMEDICAL APPLICATIONS

Image quality can essentially be summarized by four mainperformance characteristics. First, high-contrast image reso-lution �spatial resolution� is the ability to distinguish adjacenthigh-contrast features. Second, low-contrast image resolution�contrast resolution� measures the ability to differentiate alow-contrast feature from its background. Image noise im-poses a grainy appearance due to random fluctuations of thex-ray photon flux, and is a key factor in limiting low-contrastresolution. Third, temporal resolution determines the abilityto capture structures in motion. Fourth, quantitative accuracyis desired to relate the image pixel values to physicallymeaningful quantities in the absolute sense, which is particu-larly important in dual/multi-energy CT. Image artifacts are

FIG. 12. Ordinary-x-ray-tube-based phase-contrast imaging system withthree 1D gratings G0, G1 and G2. The key idea is to utilize the so-calledTalbot effect, which is a periodic self-imaging phenomenon, to extract phaseshift information �Duplicated from Ref. 121 with the permission of NaturePublishing Group�.

structured interferences of any type and clearly need to be

1061 Wang, Yu, and De Man: CT outlook 1061

minimized at all times, but are generally hard to quantify.The aforementioned 12 topics of x-ray CT research and de-velopment are all aimed at major improvements in imagequality and are well motivated by the imaging needs of im-portant biomedical applications.

Just like for clinical CT, there is a strong need for im-proved technologies for micro-CT of small animals, espe-cially genetically engineered mice.124–126 Most CT researchresults can be similarly applied in either clinical or preclini-cal scenarios. CT of large animals will remain important forveterinary medicine and research.

In terms of spatial resolution, CT has come a long way.Only a few years ago routine studies were done at 5–7 mmslice thickness, while today’s scanners offer half-mm isotro-pic spatial resolution. Still, there is an opportunity to dra-matically improve upon the state-of-the-art CT resolutionclinically available. As an example, we are working to estab-lish the feasibility of a clinical micro-CT �CMCT� prototypetargeting 60–100 �m resolution for temporal bone imaging.There has been an explosive growth in the development ofmicro-CT scanners with an emphasis on high spatialresolution126 but none of them allow patient studies due tothe narrow imaging aperture, long acquisition time, and doselimitations. Our proposed CMCT system consists of a specialmicro-CT scanner, a clinical CT scanner, and a cross-modality registration mechanism involving both hardwareand software.127 This system will integrate the strengths ofmicro-CT and clinical CT techniques to achieve the spatialresolution that is several times or even an order of magnitudefiner than any commercially available scanner for clinicaltemporal bone imaging. Our preliminary system design andsimulation results indicate that such a system is possible at aclinically acceptable dose, provided motion artifacts can beeffectively eliminated. An enabling technology for CMCT ata minimum patient dose is to perform local CT reconstruc-tion and/or local tomosynthesis of a volume of interest�VOI�, by physically monitoring the patient motion and com-pensating for it during the reconstruction, subject to variousconstraints. This technique is particularly useful for imagingof the inner ear, because of its small dimensions, high-contrast structures, fine features, and stationary nature, andmay be extended for other applications such as bone cancerstudies. Another potentially useful solution would be to ap-ply exact local reconstruction,53,99 as discussed earlier, whichcan be immediately applied to dental CT as well. In such aconfiguration, x-rays would only go through a tooth of inter-est and its neighborhood. Based on the known CT numbersin the neighborhood �air�, the tooth can be reconstructed ac-curately, allowing characterization of bony loss and so on.

Improvements in low-contrast resolution will lead to re-duced radiation dose and superior diagnostic performance.Worldwide there are active discussions and public concernson radiation induced genetic, cancerous and otherdiseases.128,129 CT is considered a radiation-intensive proce-dure, yet becoming more and more widespread. In the mid-1990s, CT scans only accounted for 4% of the total x-rayprocedures but they contributed to 40% of the collective

130

dose. The recent introduction of helical, multi-slice and

Medical Physics, Vol. 35, No. 3, March 2008

cone-beam technologies has increased and continues increas-ing the usage of CT. There is a substantial room for dosereduction using algorithmic means for attenuation-based CT.Other possible directions include more dose-efficient scan-ning protocols, better a priori knowledge, more effective re-construction methods and smarter post-processing strategies.Phase-contrast based CT, if successfully developed and clini-cally utilized, will revolutionize its x-ray imaging applica-tions by virtually eliminating the dose problems and greatlyboosting contrast resolution, which may open doors to novelfunctional and molecular imaging studies, and overtake orcomplement some MRI applications. X-ray phase-contrastCT/micro-CT is a technically very challenging and relativelyimmature area, but it has the potential to have huge impact incertain pre-clinical and clinical applications.

As far as temporal resolution is concerned, cardiac CT isby far the most prominent application.131 Most common CTtargets are relatively motionless, such as bony structures,head, neck, and abdomen. However, the closer we get to theheart, the poorer the image quality becomes, because of thecardiac motion. The �coronary� arteries are too elusive to beclearly captured in challenging cases such as when the car-diac rate is too high or irregular, since the relative rest phaseof the coronary arteries is only about 60 ms. Most demand-ing cardiac imaging used to be done by EBCT, which isexpensive and rarely available. In addition to its remarkableapplications for dynamic anatomical imaging of cardiacstructures, EBCT is also a powerful tool for physiologicalimaging. Following the introduction of the Siemens dual-source scanner, it seems that more efforts are warranted todesign better multi-source or distributed source systems andmethods for cardiac CT. The ideas and schemes already existon triple-source and multi-source CT systems, and hopefullywill be refined and realized in the not too distant future.

Finally, CT images with a high quantitative accuracy willbe more informative in general and extremely useful withenergy-sensitive CT in particular. It is anticipated that moreadvanced x-ray sources such as portable synchrotron radia-tion devices will be much more accessible one or two de-cades from now. In the meantime, innovative sources anddetectors will be developed to generate spectrally resolvedprojection data directly or indirectly. Statistical reconstruc-tion can be applied to design and refine iterative algorithmsthat fully utilize all detected photons, and add a spectral di-mension to CT images. Presumably, this type of energy-sensitive CT will dramatically improve the characterizationof heart and lung diseases as well as various cancers. Alongthis direction, scattering-based CT, x-ray induced fluores-cence CT, and CT in other hybrid modes may also becomeuseful.

V. CONCLUDING REMARKS

When the first author of this paper graduated with a CTdissertation in 1992, he considered to switch to a differentimaging field. His former mentor, Dr. Vannier, who is anauthority in x-ray imaging and a certified radiologist, advised

him not to do so because CT was “just too practical and

1062 Wang, Yu, and De Man: CT outlook 1062

useful.” After over a decade, Dr. Vannier’s advice seems stillvalid. Even if just a few of the aforementioned possibilitiesare realized, CT will become much more powerful than it isnow. The ideal CT system should waste no information re-lated to absorption, scattering, diffraction, energy, time offlight, prior information, and so on. Correspondingly, the im-aging model will become more complicated: going from asimple line integral model and the radiation transport equa-tions for describing attenuation and scattering to Maxwell’sequations for accurately modeling the wave nature of phase-contrast imaging. Although the computational obstacles ap-pear overwhelming at this time, we remain optimistic aboutour collective creativity and capabilities. The dream of wholebody CT imaging could come through in the future to pro-vide thorough, detailed and individualized image data when-ever needed for healthcare.

As indicated earlier, this outlook paper is meant to besuggestive and not exhaustive. As a result, the citations arenot comprehensive relative to the huge related literaturebase. Nevertheless, to the best of our knowledge it reflectsthe current trends in CT and, while our insights are unavoid-ably biased, the reality should not be too far from the targetswe hope for and believe in. It is our intension to keep refin-ing our predictions in this framework as time goes by and toupdate this outlook in a few years. Thus, we highly welcomecomments and critiques from colleagues and peers.

ACKNOWLEDGMENTS

This work is partially supported by the NIH Grant Nos.EB002667, EB004287, EB007288, and EB006837. The au-thors also express their gratitude to Dr. Michael Cable, Dr.Paul Fitzgerald, Dr. Ming Jiang, Dr. Xuanqin Mou, Dr. BobSenzig, Dr. Xiangyang Tang, Dr. Xizeng Wu, Dr. WenbingYun, and Dr. Jun Zhao for their valuable comments and sug-gestions.

a�Electronic mail: ge-wang@ieee.orgb�Electronic mail: hengyong-yu@ieee.orgc�Electronic mail: deman@ieee.org1A. M. Cormack, “Representation of a function by its line integrals, withsome radiological applications,” J. Appl. Phys. 34�9�, 2722–2727 �1963�.

2G. N. Hounsfield, “Computerized transverse axial scanning �tomogra-phy�: Part I. Description of system,” Br. J. Radiol. 46, 1016–1022 �1973�.

3R. A. Robb et al., “High-sped three-dimensional x-ray computed tomog-raphy: The dynamic spatial reconstructor,” Proceedings of the IEEE,71�3�, 308–319 �1983�.

4E. L. Ritman, R. A. Robb, and L. D. Harris, Imaging Physiological Func-tions: Experience with the DSR, Praeger, Philadelphia, �1985�.

5W. A. Kalender, “Thin-section three-dimensional spiral CT: Is isotropicimaging possible?,” Radiology 197�3�, 578–580 �1995�.

6G. Wang, T. H. Lin, P. C. Cheng, and D. M. Shinozaki, “A general conebeam reconstruction algorithm,” IEEE Trans. Med. Imaging 12�3�, 486–496 �1993�.

7G. Wang et al., “Scanning cone-beam reconstruction algorithms for x-raymicrotomography,” Proceedings of SPIE, Vol. 1556, pp. 99–112 �1991�.

8K. Taguchi and H. Aradate, “Algorithm for image reconstruction in multi-slice helical CT,” Med. Phys. 25�4�, 550–561 �1998�.

9M. Kachelriess, S. Schaller, and W. A. Kalender, “Advanced single-slicerebinning in cone-beam spiral CT,” Med. Phys. 27�4�, 754–772 �2000�.

10S. Mori et al., “Physical performance evaluation of a 256-slice CT-scanner for four-dimensional imaging,” Med. Phys. 31�6�, 1348–1356�2004�.

11

G. Wang, C. R. Crawford, and W. A. Kalender, “Multirow detector and

Medical Physics, Vol. 35, No. 3, March 2008

cone-beam spiral/helical CT,” IEEE Trans. Med. Imaging 19�9�, 817–821�2000�.

12L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beamalgorithm,” J. Opt. Soc. Am. A 1�A�, 612–619 �1984�.

13A. Katsevich, “An improved exact filtered backprojection algorithm forspiral computed tomography,” Adv. Appl. Math. 32�4�, 681–697 �2004�.

14A. Katsevich, “Theoretically exact filtered backprojection-type inversionalgorithm for spiral CT,” SIAM J. Appl. Math. 62�6�, 2012–2026 �2002�.

15G. Wang, Y. B. Ye, and H. Y. Yu, “Appropriate and exact cone-beamreconstruction with standard and nonstandard spiral scanning,” Phys.Med. Biol. 52�6�, R1–R13 �2007�.

16Y. B. Ye, S. Y. Zhao, H. Y. Yu, and G. Wang, “Exact reconstruction forcone-beam scanning along nonstandard spirals and other curves,” Pro-ceedings of SPIE, Vol. 5535, pp. 293–300, Denver, CO �2004�.

17Y. B. Ye and G. Wang, “Filtered backprojection formula for exact imagereconstruction from cone-beam data along a general scanning curve,”Med. Phys. 32�1�, 42–48 �2005�.

18Y. B. Ye, S. Y. Zhao, H. Y. Yu, and G. Wang, “A general exact recon-struction for cone-beam CT via backprojection-filtration,” IEEE Trans.Med. Imaging 24�9�, 1190–1198 �2005�.

19J. Pack and F. Noo, “Cone-beam reconstruction using 1D filtering alongthe projection of M-lines,” Inverse Probl. 21�4�, 1105–1120 �2005�.

20J. D Pack, F. Noo, and R. Clackdoyle, “Cone-beam reconstruction usingthe backprojection of locally filtered projections,” IEEE Trans. Med. Im-aging 24�1�, 70–85 �2005�.

21T. L. Zhuang, S. Leng, B. E. Nett, and G. H. Chen, “Fan-beam andcone-beam image reconstruction via filtering the backprojection image ofdifferentiated projection data,” Phys. Med. Biol. 49�24�, 5489–5503�2004�.

22Y. Zou, X. Pan, and E. Y. Sidky, “Theory and algorithms for imagereconstruction on chords and within regions of interest,” J. Opt. Soc. Am.A 22�11�, 2372–2384 �2005�.

23G. Wang and Y. B. Ye, “Nonstandard spiral cone-beam scanning methods,apparatus, and applications,” US Provisional Patent Application 60/588,682, Filing date: July 16, 2004 �2004�.

24G. Wang and H. Y. Yu, “Scanning method and system for cardiac imag-ing,” Patent disclosure submitted to Virginia Tech. Intellectual Propertieson July 17 �2007�.

25A. Katsevich and M. Kapralov, “Theoretically exact FBP reconstructionalgorithms for two general classes of curves,” 9th International Meetingon Fully Three-Dimensional Image Reconstruction in Radiology andNuclear Medicine, pp. 80–83, Lindau, Germany �2007�.

26A. Katsevich, “Image reconstruction for a general circle-plus trajectory,”Inverse Probl. 23�5�, 2223–2230 �2007�.

27A. Katsevich, “A general scheme for constructing inversion algorithmsfor cone-beam CT,” Int. J. Math. Math. Sci. 21, 1305–1321 �2003�.

28Y. C. Wei, H. Y. Yu, J. Hsieh, W. X. Cong, and G. Wang, “Scheme ofcomputed tomography,” J. X-Ray Sci. Technol. 15�4�, 235–270 �2007�.

29S. Y. Zhao, H. Y. Yu, and G. Wang, “A unified framework for exactcone-beam reconstruction formulas,” Med. Phys. 32�6�, 1712–1721�2005�.

30G. H. Chen, “An alternative derivation of Katsevich’s cone-beam recon-struction formula,” Med. Phys. 30�12�, 3217–3226 �2003�.

31H. Y. Yu and G. Wang, “Cone-beam composite-circling scan and exactimage reconstruction for a quasi-short object,” Int. J. Biomed. Imaging2007, Article ID 87319, 10 pages �2007�.

32J. D. Pack, F. Noo, and H. Kudo, “Investigation of saddle trajectories forcardiac CT imaging in cone-beam geometry,” Phys. Med. Biol. 49�11�,2317–2336 �2004�.

33X. Y. Tang, J. Hsieh, R. A. Nilsen, S. Dutta, D. Samsonov, and A. Hagi-wara, “A three-dimensional-weighted cone beam filtered backprojection�CB-FBP� algorithm for image reconstruction in volumetric CT - helicalscanning,” Phys. Med. Biol. 51�4�, 855–874 �2006�.

34X. Y. Tang and J. Hsieh, “A filtered backprojection algorithm for conebeam reconstruction using rotational filtering under helical source trajec-tory,” Med. Phys. 31�11�, 2949–2960 �2004�.

35X. Y. Tang and J. Hsieh, “Handling data redundancy in helical cone beamreconstruction with a cone-angle-based window function and itsasymptotic approximation,” Med. Phys. 34�6�, 1989–1998 �2007�.

36X. Y. Tang, J. Hsieh, R. A. Nilsen, and S. M. McOlash, “Extendingthree-dimensional weighted cone beam filtered backprojection �CB-FBP�algorithm for image reconstruction in volumetric CT at low helical

pitches,” Int. J. Biomed. Imaging 2006, Article ID 45942, 8 pages �2006�.

1063 Wang, Yu, and De Man: CT outlook 1063

37R. Gordon, R. Bender, and G. T. Herman, “Algebraic reconstruction tech-niques �ART� for three-dimensional electron microscopy and x-ray pho-tography,” J. Theor. Biol. 29, 471–482 �1970�.

38A. H. Andersen and A. C. Kak, “Simultaneous algebraic reconstructiontechnique �SART�: a superior implementation of the ART algorithm,”Ultrason. Imaging 6, 81–94 �1984�.

39L. A. Shepp and B. F. Logan, “The Fourier reconstruction of a headsection,” IEEE Trans. Nucl. Sci. NS21�3�, 21–34 �1974�.

40A. J. Rockmore and A. Macovski, “A maximum likelihood approach toemission image reconstruction from projections,” IEEE Trans. Nucl. Sci.23�4�, 1428–1432 �1976�.

41J. Nuyts et al., “Iterative reconstruction for helical CT: A simulationstudy,” Phys. Med. Biol. 43�4�, 729–737 �1998�.

42B. De Man et al., “Reduction of metal streak artifacts in x-ray computedtomography using a transmission maximum a posteriori algorithm,”IEEE Trans. Nucl. Sci. 47�3�, 977–981 �2000�.

43I. A. Elbakri and J. A. Fessler, “Statistical image reconstruction for poly-energetic x-ray computed tomography,” IEEE Trans. Med. Imaging 21�2�,89–99 �2002�.

44J. A. Fessler, “Statistical image reconstruction methods for transmissiontomography,” in Medical Image Processing and Analysis, edited by M.Sonka and J. M. Fitzpatric �SPIE Press, Bellingham, Washington, 2000�,Vol. 3, pp. 1–70.

45J. A. Fessler and A. O. Hero, “Space-alternating generalized expectation-maximization algorithm,” IEEE Trans. Signal Process. 42�10�, 2664–2677 �1994�.

46I. A. Elbakri and J. A. Fessler, “Segmentation-free statistical image re-construction for polyenergetic x-ray computed tomography with experi-mental validation,” Phys. Med. Biol. 48�15�, 2453–2477 �2003�.

47P. E. Kinahan, A. M. Alessio, and J. A. Fessler, “Dual energy CT attenu-ation correction methods for quantitative assessment of response to can-cer therapy with PET/CT imaging,” Technol. Cancer Res. Treat. 5�4�,319–327 �2006�.

48J. B. Thibault, A. K. Sauer, A. C. Bouman, and J. Hsieh, “Three-dimensional statistical modeling for image quality improvements inmulti-slice helical CT,” 8th Meeting on Fully Three–Dimensional ImageReconstruction in Radiology and Nuclear Medicine, pp. 271–274, SaltLake City, UT �2005�.

49M. Iatrou, B. De Man, and S. Basu, “A comparison between filteredbackprojection, post-smoothed weighted least squares, and penalizedweighted least squares for CT reconstruction,” IEEE Nuclear ScienceSymposium Conference Record, 2006, pp. 2845–2850, San Diego, CA�2006�.

50D. L. Parker, “Optimal short scan convolution reconstruction for fanbeamCT,” Med. Phys. 9�2�, 254–257 �1982�.

51Y. Zou and X. C. Pan, “Exact image reconstruction on PI-lines fromminimum data in helical cone-beam CT,” Phys. Med. Biol. 49�6�, 941–959 �2004�.

52M. Defrise, F. Noo, R. Clackdoyle, and H. Kudo, “Truncated Hilberttransform and image reconstruction from limited tomographic data,” In-verse Probl. 22�3�, 1037–1053 �2006�.

53Y. B. Ye, H. Y. Yu, Y. C. Wei, and G. Wang, “A general local reconstruc-tion approach based on a truncated Hilbert transform,” Int. J. Biomed.Imaging 2007, Article ID 63634, 8 pages �2007�.

54Y. B. Ye, H. Y. Yu, and G. Wang, “Exact interior reconstruction withcone-beam CT,” Int. J. Biomed. Imaging 2007, Article ID 10693, 5 pages�2007�.

55F. Natterer, The Mathematics of Computerized Tomography �Society forIndustrial and Applied Mathematics, Philadelphia, 2001�.

56G. Wang, Y. B. Ye, and H. Y. Yu, “General VOI/ROI reconstructionmethods and systems using a truncated Hilbert transform,” Patent disclo-sure submitted to Virginia Tech. Intellectual Properties on May 15 �2007�.

57H. Y. Yu and G. Wang, “A general formula for fan-beam lambda tomog-raphy,” Int. J. Biomed. Imaging 2006, Article ID 10427, 9 pages �2006�.

58H. Y. Yu, Y. C. Wei, Y. B. Ye, and G. Wang, “Lambda tomography withdiscontinuous scanning trajectories,” Phys. Med. Biol. 52�14�, 4331–4344�2007�.

59H. Y. Yu and G. Wang, “A general scheme for velocity tomography,”Signal Process. 88�5�, 1165–1175 �2008�.

60D. Saintfelix et al., “In-vivo evaluation of a new system for 3D comput-erized angiography,” Phys. Med. Biol. 39�3�, 583–595 �1994�.

61L. T. Niklason et al., “Digital tomosynthesis in breast imaging,” Radiol-

ogy 205�2�, 399–406 �1997�.

Medical Physics, Vol. 35, No. 3, March 2008

62W. Ross, D. D. Cody, and J. D. Hazle, “Design and performance charac-teristics of a digital flat-panel computed tomography system,” Med. Phys.33�6�, 1888–1901 �2006�.

63S. Greschus et al., “Potential applications off lat-panel volumetric CT inmorphologic and functional small animal imaging,” Neoplasia 7�8�, 730–740 �2005�.

64F. Kiessling et al., “Volumetric computed tomography �VCT�: A newtechnology for noninvasive high-resolution monitoring of tumor angio-genesis,” Nat. Med. 10�10�, 1133–1138 �2004�.

65J. M. Boone et al., “Computed tomography for imaging the breast,” J.Mammary Gland Biol. Neoplasia 11�2�, 103–111 �2006�.

66B. Chen and R. Ning, “Cone-beam volume CT breast imaging: Feasibilitystudy,” Med. Phys. 29�5�, 755–770 �2002�.

67L. T. Holly and K. T. Foley, “Image guidance in spine surgery,” Orthop.Clin. North Am. 38�3�, 451–461 �2007�.

68N. Clinthorne, “Intra-operative imaging,” J. Laparoendosc Adv. Surg.Tech. A 15�5�, 549–555 �2005�.

69W. Berninger and R. Redington, “Multiple purpose high speed tomogra-phic x-ray scanner,” US Patent No. 4,196,352A �1980�.

70T. G. Flohr et al., “First performance evaluation of a dual-source CT�DSCT� system,” Eur. Radiol. 16�2�, 256–268 �2006�.

71Y. Kyriakou and W. A. Kalender, “Intensity distribution and impact ofscatter for dual-source CT,” Phys. Med. Biol. 52�23�, 6969–6989 �2007�.

72J. Zhao, M. Jiang, T. Zhuang, and G. Wang, “An exact reconstructionalgorithm for triple-source helical cone-beam CT,” J. X-Ray Sci. Technol.14�3�, 191–206 �2006�.

73J. Zhao, M. Jiang, T. G. Zhuang, and G. Wang, “Minimum detectionwindow and inter-helix PI-line with triple-source helical cone-beam scan-ning,” J. X-Ray Sci. Technol. 14�2�, 95–107 �2006�.

74J. Zhao, Y. N. Jin, Y. Lu, and G. Wang, “A reconstruction algorithm fortriple-source helical cone-beam CT via filtered backprojection,” 9th Inter-national Meeting on Fully Three-Dimensional Image Reconstruction inRadiology and Nuclear Medicine, pp. 205–208, Lindau, Germany �2007�.

75S. S. Orlov, “Theory of three dimensional reconstruction I: Conditions fora complete set of projections,” Sov. Phys. Crystallogr. 20, 312–314�1975�.

76B. De Man, P. Edic, and S. Basu, “An iterative algorithm for time-resolved reconstruction of a CT scan of a beating heart,” 8th Meeting onFully Three–Dimensional Image Reconstruction in Radiation andNuclear Medicine, Salt Lake City, UT, 356–359 �2005�.

77D. Boyd et al., “High-speed, multi-slice, x-ray computed tomography,”Proceedings of SPIE, Vol. 372, pp. 139–150, Pacific Grove, CA �1982�.

78G. Wang et al., “Top-level design and preliminary physical analysis forthe first electron-beam micro-CT scanner,” J. X-Ray Sci. Technol. 12�4�,251–260 �2004�.

79B. De Man et al., “Stationary computed tomography system and method,”US Patent No. 7,280,631 B2�2005�.

80J. S. Price, W. Block, and M. Vermilyea, “Extended multi-spot computedtomography x-ray source,” US Patent No. 6,983,035 B2�2006�.

81Z. Yin, B. De Man, and J. Pack, “Analytical cone-beam reconstructionusing a multi-source inverse geometry CT system,” Proceedings of SPIE,Vol. 6510, San Diego, CA �2007�.

82T. G. Schmidt, R. Fahrig, N. J. Pelc, and E. G. Solomon, “An inverse-geometry volumetric CT system with a large-area scanned source: A fea-sibility study,” Med. Phys. 31�9�, 2623–2627 �2004�.

83B. De Man et al., “Multi-source inverse geometry CT: A new systemconcept for X-ray computed tomography,” Proceedings of SPIE, Vol.6510, San Diego, CA �2007�.

84G. Wang and M. W. Vannier, “Bolus-chasing angiography with adaptivereal-time computed tomography,” U.S. Patent No. 6,535,821, allowed on11/26/2002, issued on 3/18/2003 �2003�.

85E. W. Bai et al., “Study of an adaptive bolus chasing CT angiography,” J.X-Ray Sci. Technol. 14�1�, 27–38 �2006�.

86A. Katsevich, S. Basu, and J. Hsieh, “Exact filtered backprojection recon-struction for dynamic pitch helical cone beam computed tomography,”Phys. Med. Biol. 49�14�, 3089–3103 �2004�.

87E. Tkaczyk et al., “Simulation of CT dose and contrast-to-noise as func-tion of bowtie shape,” Proceedings of SPIE, Vol. 5368, pp. 403–410, SanDiego, CA �2004�.

88M. Kalra, N. Naz, S. Rizzo, and M. Blake, “Computed tomography ra-diation dose optimization: Scanning protocols and clinical applications ofautomatic exposure control,” Curr. Probl Diagn. Radiol. 34�5�, 171–181

�2005�.

1064 Wang, Yu, and De Man: CT outlook 1064

89B. De Man and B. Senzig, “Method and system for radiographic imagingwith organ-based radiation profile prescription,” US Patent No.20070147579 A1�2007�.

90R. E. Alvarez and A. Macovski, “Energy-selective reconstructions inx-ray computerized tomography,” Phys. Med. Biol. 21�5�, 733–744�1976�.

91J. H. Hubbell and S. M. Seltzer, Tables of X-ray mass attenuation coef-ficients 1 keV to 20 MeV, for elements Z�1 /4� 1 to 92 and 48 additionalsubstances of dosimetric interest. �National Institute of Standards andTechnology, Gaithersburg, MD, 1995�.

92B. De Man et al., “An iterative maximum-likelihood polychromatic algo-rithm for CT,” IEEE Trans. Med. Imaging 20�10�, 999–1008 �2001�.

93R. Carmi, G. Naveh, and A. Altman, “Material separation with dual-layerCT,” IEEE Nuclear Science Symposium Conference Record, pp. 1876–1878, Wyndham El Conquistador Resort, Puerto Rico �2005�.

94J. E. Tkaczyk et al., “Atomic number resolution for three spectral CTimaging systems,” Proceedings of SPIE, 6510, paper No. 651009, SanDiego, CA �2007�.

95S. Kuribayashi, “Dual energy CT of peripheral arterial disease withsingle-sorurce 64–slice MDCT,” presented at 9th Annual InternationalSymposium on Multidetector-Row CT, San Francisco, CA �2007�.

96X. Wu et al., “Dual kVp computed tomography via view-to-view kVswitching: Temporal resolution and energy separation �submitted�.

97P. M. Shikhaliev, “Tilted angle CZT detector for photon counting/energyweighting x-ray and CT imaging,” Phys. Med. Biol. 51�17�, 4267–4287�2006�.

98nanoXCT, available from: http://www.xradia.com/Products/nanoxct.html.99H. Y. Yu, Y. B. Ye, and G. Wang, “Local reconstruction using the trun-

cated Hilbert transform via singular value decomposition,” Phys. Med.Biol. �under review�.

100J. F. Barrett and N. Keat, “Artifacts in CT: Recognition and avoidance,”Radiographics 24�6�, 1679–1691 �2004�.

101X. Q. Mou, S. J. Tang, and H. Y. Yu, “A beam hardening correctionmethod based on HL consistency,” Proc. of SPIE, Vol. 6318, San Diego,CA �2006�.

102R. P. Zeng, J. A. Fessler, and J. M. Balter, “Estimating 3-D respiratorymotion from orbiting views by tomographic image registration,” IEEETrans. Med. Imaging 26�2�, 153–163 �2007�.

103R. P. Zeng, J. A. Fessler, and J. M. Balter, “Respiratory motion estimationfrom slowly rotating X-ray projections: Theory and simulation,” Med.Phys. 32�4�, 984–991 �2005�.

104Y. C. Wei, H. Y. Yu, and G. Wang, “Integral invariants in fan-beam andcone-beam computed tomography,” IEEE Signal Process. Lett. 13�9�,549–552 �2006�.

105H. Y. Yu and G. Wang, “Data consistency based general motion artifactreduction in fan-beam CT,” IEEE Trans. Med. Imaging 26�2�, 249–260�2007�.

106G. Wang, M. W. Vannier, and P. C. Cheng, “Iterative X-ray cone-beamtomography for metal artifact reduction and local region reconstruction,”Microsc. Microanal. 5, 58–65 �1999�.

107J. A. Fessler, “Penalized weighted least-squares image-reconstruction forpositron emission tomography,” IEEE Trans. Med. Imaging 13�2�, 290–300 �1994�.

108Bruno De Man, “Iterative reconstruction for reduction of metal artifactsin CT,” Ph.D. Dissertation, University of Leuven, Belgium �2001�.

109J. A. O’Sullivan and J. Benac, “Alternating minimization algorithms fortransmission tomography,” IEEE Trans. Med. Imaging 26�3�, 283–297

Medical Physics, Vol. 35, No. 3, March 2008

�2007�.110T. Beyer et al., “A combined PET/CT scanner for clinical oncology,” J.

Nucl. Med. 41�8�, 1369–1379 �2000�.111D. A. Boas and A. M. Dale, “Simulation study of magnetic resonance

imaging-guided cortically constrained diffuse optical tomography of hu-man brain function,” Appl. Opt. 44�10�, 1957–1968 �2005�.

112K. Kagawa et al., “Initial clinical assessment of CT-MRI image fusionsoftware in localization of the prostate for 3D conformal radiationtherapy,” Int. J. Radiat. Oncol., Biol., Phys. 38�2�, 319–325 �1997�.

113G. Wang, Y. Li, and M. Jiang, “Uniqueness theorems in bioluminescencetomography,” Med. Phys. 31�8�, 2289–2299 �2004�.

114J. Nuyts et al., “Simultaneous maximum a posteriori reconstruction ofattenuation and activity distributions from emission sinograms,” IEEETrans. Med. Imaging 18�5�, 393–403 �1999�.

115A. Momose and J. Fukuda, “Phase-contrast radiographs of nonstained ratcerebellar specimen,” Med. Phys. 22�4�, 375–379 �1995�.

116R. A. Lewis, “Medical phase contrast X-ray imaging: Current status andfuture prospects,” Phys. Med. Biol. 49�16�, 3573–3583 �2004�.

117A. Momose, “Recent advances in x-ray phase imaging,” Jpn. J. Appl.Phys., Part 1 44�9A�, 6355–6367 �2005�.

118X. Z. Wu and H. Liu, “Clarification of aspects in in-line phase-sensitiveX-ray imaging,” Med. Phys. 34�2�, 737–743 �2007�.

119X. Z. Wu, H. Liu, and A. M. Yan, “x-ray phase-attenuation duality andphase retrieval,” Opt. Lett. 30�4�, 379–381 �2005�.

120S. W. Wilkins et al., “Phase-contrast imaging using polychromatic hardX-rays,” Nature �London� 384, 335–338 �1996�.

121F. Pfeiffer, T. Weitkamp, O. Bunk, and C. David, “Phase retrieval anddifferential phase-contrast imaging with low-brilliance x-ray sources,”Nat. Phys. 2�4�, 258–261 �2006�.

122C. Kottler, C. David, F. Pfeiffer, and O. Bunk, “A two-directional ap-proach for grating based differential phase contrast imaging using hard xrays,” Opt. Express 15�3�, 1175–1181 �2007�.

123M. Jiang, C. L. Wyatt, and G. Wang, “X-ray phase-contrast imaging withthree 2D gratings,” Int. J. Biomed. Imaging 2007, Article ID 27152�2007�.

124D. W. Holdsworth, “Micro-CT in small animal and specimen imaging,”Trends Biotechnol. 20�8�, S34–S39 �2002�.

125M. J. Paulus, “A review of high-resolution x-ray computed tomographyand other imaging modalities for small animal research,” Lab Anim. 30,36–45 �2001�.

126G. Wang, “Micro-CT scanners for biomedical applications: An over-view,” Adv. Imaging 16, 18–27 �2001�.

127G. Wang et al., “Design, analysis and simulation for development of thefirst clinical micro-CT scanner,” Acad. Radiol. 12�4�, 511–525 �2005�.

128D. J. Brenner, C. D. Elliston, E. J. Hall, and W. E. Berdon, “Estimatedrisks of radiation-induced fatal cancer from pediatric CT,” AJR, Am. J.Roentgenol. 176�2�, 289–296 �2001�.

129A. Berrington de Gonzalez and S. Darby, “Risk of cancer from diagnosticx rays: Estimates for the UK and 14 other countries,” Lancet 363, 345–351 �2004�.

130O. W. Linto and F. A. Mattler, Jr. , “National conference on dose reduc-tion in computed tomography, emphasis on pediatrics,” AJR, Am. J.Roentgenol. 181, 321–329 �2003�.

131G. Wang, S. Y. Zhao, and D. Heuscher, “A knowledge-based cone-beamx-ray CT algorithm for dynamic volumetric cardiac imaging,” Med. Phys.29�8�, 1807–1822 �2002�.