multislice helical ct: image temporal resolution

384 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 19, NO. 5, MAY 2000

Multislice Helical CT: Image Temporal ResolutionHui Hu*, Member, IEEE, Tinsu Pan, and Yun Shen

Abstract—A multislice helical computed tomography (CT) half-scan (HS) reconstruction algorithm is proposed for cardiac appli-cations. The imaging performances (in terms of the temporal res-olution, -axis resolution, image noise, and image artifacts) of theHS algorithm are compared to the existing algorithms using the-oretical models and clinical data. A theoretical model of the tem-poral resolution performance (in terms of the temporal sensitivityprofile) is established for helical CT, in general, i.e., for any numberof detector rows and any reconstruction algorithm used. It is con-cluded that the HS reconstruction results in improved image tem-poral resolution than the corresponding 180 LI (linear interpola-tion) reconstruction and is more immune to the inconsistent dataproblem induced by cardiac motions. The temporal resolution ofmultislice helical CT with the HS algorithm is comparable to thatof single-slice helical CT with the HS algorithm. In practice, the180 LI and HS-LI algorithms can be used in parallel to generatetwo image sets from the same scan acquisition, one (180LI) forimproved -resolution and noises, and the other (HS-LI) for im-proved image temporal resolution.

Index Terms—Helical CT, multidetector-row CT, multislice CT,temporal resolution.

I. INTRODUCTION

M AJORITY computed tomography (CT) applications inmedical diagnosis are, in many ways, dictated by the

scanner’s volume coverage speed, which refers to its capabilityof rapidly scanning a large volume of interest(e.g., an entirelung and/or liver)with good image quality(i.e., thin-slice andlow-image artifacts). The scanning time for these applicationsis usually a fraction of a minute, which is imposed by 1) thetime window for optimal contrast enhancement and 2) a patientbreathholding duration for reduced image degradation inducedby breath-related motions.

Helical (or spiral) CT, introduced a decade ago [1], [2], repre-sents a substantial improvement in the volume coverage speed.Helical CT involves simultaneously transporting the patient ata constant speed through the gantry while CT data are continu-ously acquired over multiple gantry rotations.

A further and substantial improvement in the volume cov-erage speed can be achieved by a combination of helical CT withthe so-called multislice (or multidetector row) CT. The multi-slice CT scanner refers to a special CT system equipped witha multiple-row detector array (Fig. 1), as opposed to today’s

Manuscript received August 30, 1999; revised February 24, 2000. Thiswork was supported partially by the National Science Foundation AwardDMI-9661717. The Associate Editor responsible for coordinating the reviewof this paper and recommending its publication were M. Defrise.Asteriskindicates corresponding author.

*H. Hu is with ImagingTech, Inc., Waukesha, WI 53186 USA (e-mail:[email protected]).

T. Pan and Y. Shen are with the General Electric Company, Milwaukee, WI53201 USA.

Publisher Item Identifier S 0278-0062(00)05271-X.

commonly used single-row detector array. We refer to a mul-tislice CT scanner with N-slice simultaneous data collection asa N-slice CT scanner. Most CT scanners in use are the one-slice(or single slice) CT scanner. A two-slice CT scanner was in-troduced several years ago [3]. Several four-slice CT scannershave been recently introduced. The volume coverage speed offour-slice helical CT has been discussed in recent publications[4]–[7].

Besides the applications dictated by the volume coveragespeed, there is another type of applications, namely, cardiac ap-plications, that require high performance of the image temporalresolution. The image temporal resolution refers to the temporalresolution of each CT image. Examples of cardiac applicationsinclude cardiac and pulmonary imaging. The time window forcardiac applications is a fraction of one cardiac cycle (i.e., afraction of one second) in order to reduce the inconsistent dataproblem induced by cardiac-related motions. Traditionally,cardiac CT applications can only be performed with electronbeam CT (EBCT) scanners [8]–[10] and the applications arehindered by the limited availability of EBCT scanners. As thespeed of general-purpose CT scanners increases, the cardiacapplications are being actively developed on general-purposeCT scanners [11]–[15].

In this paper, we will propose a multislice helical interpo-lation algorithm for cardiac imaging applications. We willestablish a general theoretical model to predict the temporalsensitivity profile of CT images, i.e., regardless the number ofdetector rows and the reconstruction algorithm used. We willcompare the performances (in terms of the temporal resolution,-axis resolution, image noise, and image artifacts) of the

proposed multislice helical interpolation algorithm with theexisting ones using theoretical models and clinical data.

II. M ULTISLICE HELICAL INTERPOLATION ALGORITHM FOR

IMPROVED IMAGE TEMPORAL RESOLUTION

Multislice helical CT calls for new reconstruction algorithmsthat take advantage of the simultaneous data acquisition by mul-tiple detector rows. It has been demonstrated that for a smallnumber of detector rows (e.g., ), the actual cone-beam CTimaging geometry can be approximated by the multiple, parallelfan-beams, as illustrated by dark dashed lines in Fig. 1. This ap-proximation allows the use of the existing fan-beam-based com-puting system.

Thus, the multislice helical CT reconstruction consists ofthe following two steps: helical interpolation and conventionalstep-and-shoot (axial) CT reconstruction. The helical interpola-tion algorithm involves estimating a complete set of projectiondata at each prescribed-location from a multislice helical dataset . The projection along a given projection path is estimated by

0278–0062/00$10.00 © 2000 IEEE

HU et al.: IMAGE TEMPORAL RESOLUTION 385

Fig. 1. A sideview of the multislice CT scanner.

weighted average (interpolation) of those measurements, whichwould be along the same projection path if the differences in-sampling positions and cone angles were ignored.We denote multislice projection measurements as ,

where denotes the gantry rotational angle of a given view,the fan angle of a given detector channel andthe detector

row index (ranging from 1 to ). We use and to de-note the gantry rotational angle and the table-location, respec-tively, when the X-ray focal spot is passing the slice to be recon-structed. Thus, the-sampling position of the multislice projec-tion measurement is given by the following equation:

(1a)

where

(1b)

In (1a), the second term describes the table translation as thegantry rotates, where denotes the table transporting distanceper rotation. The third term depicts thedisplacements from themidplane due to the use of different detector rows (referring toFig. 1), where represents the detector row spacing measuredat the axis of rotation. is given in (1b). For example,for the single slice system; for thefour-slice system.

The existing multislice helical interpolation algorithms [4],[7], referred to as the 180LI (linear interpolation) algorithm,allow interpolation between the measurements acquired alongopposing projection paths, denoted as and , whichsatisfy the following relationship:

(2a)

(2b)

With reference to (1a) and (2a), since the measurements of op-posing projection paths are collected around 180gantry ro-tation apart, tend to provide measurements at dif-ferent locations from due to the table translation.Thus, including the measurements of the opposing projectionpath in the interpolation will potentially increase theeffective -sampling density and will result in a more accurateprojection data estimate. Consequently, the 180LI algorithmwill improve the -axis resolution and/or the volume coveragespeed performance. However, the 180LI algorithm results in

relatively poor image temporal resolution because combiningdata that are 180gantry rotation apart will degrade the datatemporal resolution and therefore the image temporal resolu-tion.

To improve the image temporal resolution, the halfscan (HS)reconstruction algorithm [16] is adapted for multislice helicalCT reconstruction. In this algorithm, only 180 fan-angleworth of data [16] is used to reconstruct each image. The inter-polation occurs between the measurements acquired by differentdetector rows within the same projection view and themeasurements of the opposing projection path areexcluded. Depending on the interpolation type, the HS algo-rithm can be further categorized as the halfscan with the linearinterpolation (HS-LI), the halfscan with the nearest neighbor in-terpolation (HS-NI), and so on. The HS-LI algorithm can be de-scribed as follows:

(3a)

where

(3b)

and

(3c)

In (3), and are indexes of the two adjacent detectorrows whose locations, denoted as and , respectively,are closest to the location of the reconstructed slice. The

and are the helical interpolation function applied toand , respectively. In (3b) and (3c), the

and are explicitly expressed as a function of using(1). is the halfscan weighting function [16] centeredat the location of the reconstructed slice. The mathematicalformula of is given in [16].

Unlike the 180 LI algorithm, the HS algorithm eliminatesthe degradation of the data temporal resolution by excludingthe measurements of the opposing projection path inthe interpolation. Thus, the HS algorithm improves the imagetemporal resolution. However, the HS algorithm may result indegraded -axis resolution and image noise due to the exclusionof the measurements of the opposing projection path.


III. T HEORETICAL MODEL OF TEMPORAL SENSITIVITY

PROFILE OFHELICAL CT IMAGE

The image temporal resolution is primarily determined bythe temporal span of the projection data used to reconstruct theimage. The temporal response of an image can be further char-acterized by the image sensitivity to a structure that exists onlyin a very short moment (i.e., a temporal impulse). The imagesensitivity as a function of the time of the temporal impulse isreferred to as the temporal sensitivity profile (TSP).

For simplicity, we will study TSP along the axis of rotation.We use to denote the time of the temporal impulse andthecenter of the time window of the projection data used to recon-struct the image. It follows that

and (4)

where is the gantry rotational speed. We use ,or to denote the weighting function applied to the mea-surement of the th detector row at the center detector channel

—the channel which the axis of rotation is projectedonto. The time representation of the weighting function isnew and can be readily obtained from the commonly used rep-resentations or by substitution of variable using(4) and (1).

The formula for predicting TSP can be derived by drawingthe parallelism between the temporal sensitivity profile (TSP)and the slice sensitivity profile (SSP). It has been proven [4]that SSP of a CT image, denoted as , can be modeled asfollows:

(5)

where is the profile of individual X-ray beam and isthe distance to the reconstructed image. The width of canbe characterized by the detector row collimation. Equation (5)can be rewritten as the following convolution:

where (6)

Equation (6) has a clear physical meaning. It states that the-resolution of a CT image is limited by two-averaging ef-

fects, due to a finite thickness of fan beam used in acquisitionof each projection view, and the average of all projection viewsin image reconstruction, respectively. These two effects can becharacterized by theprofile of individual X-ray beam andthe weighting function in reconstruction, . As illustrated in(6) and predicted by the linear system theory, the overall SSP isa convolution of SSP of each contributing factor. In practice, thewidth of is comparable to that of , both in the order ofmillimeters.

By drawing the parallelism between SSP and TSP, one canexpress TSP of a CT image, denoted as , as the followingconvolution:

where (7)

In (7), represents the TSP due to acquisition of each pro-jection view, and denotes the TSP due to combination ofall projection views in image reconstruction. In other words,(7) states the physical fact that the temporal resolution of a CTimage is determined by two time-averaging effects, due to thesignal integration over a finite time duration in each projec-tion view acquisition, and the average of all projection views inimage reconstruction, respectively. Note that the width of,which is associated with the signal integration time of each view,is about 1 ms or less. On the other hand, the width of ,which is associated with the temporal span of a complete set ofviews used to reconstruct an image, is about several hundreds ofmillisecond. Thus, unlike the case of SSP, the width of ismuch (1000 times) narrower than that of and the wideningof due to can be ignored. Consequently, (7) can bereduced as follows:

(8)

where, is the weighting function applied to the measure-ment of the th detector row at the center detector channel

.The above derivation can be confirmed by an alternative

derivation as follows. The projection views, representing CTmeasurements at different moments, are weighted equallyin step-and-shoot (axial) CT reconstruction, resulting in arectangular TSP with its width equal to the temporal span ofthe data used to reconstruct the image. In helical CT recon-struction, on the other hand, the projection views are weighteddifferently, determined by the weighting function. Since eachview contributes to the CT image in a linear fashion, the TSP ofhelical CT is that of step-and-shoot CT (a rectangle) modulatedby the weighting function, as illustrated in (8).

Like the SSP model (5) or (6), the TSP model (8) is general,applicable to any CT system with any number of detector rows,any helical pitch, and any reconstruction algorithm used.

IV. PERFORMANCEEVALUATION OF FOUR-SLICE CT IN

CARDIAC IMAGING

The performance of the proposed HS algorithm was com-pared with the existing 180LI algorithm for several helicalCT imaging protocols of practical interest. The imaging charac-teristics compared include the temporal resolution performance(in terms of TSP), the-axis resolution performance (in termsof SSP), the image noise, and image artifacts. These imagingcharacteristics were studied using both the theoretical modelsand/or clinical data.

Specifically, the TSP’s of several helical CT protocols werederived from (8) and are displayed in Fig. 2. The full widths ofthese TSP’s at 0, 10, and 50% of the maximum were measuredand are tabulated in Table I. It is shown from Fig. 2 and TableI that the HS algorithms provide substantially better temporalresolution performance (more compact TSP) than the 180LIalgorithms. The temporal resolution of a four-slice helical CTwith the HS-LI algorithm is comparable to that of single slicehelical CT with the HS-NI algorithm.


Fig. 2. Temporal sensitivity profile (TSP) of the HS (including both LI and NI) algorithms for CT systems of any number of detector rows and for any pitchisshown by the thick solid line. In comparison, TSP’s of the 180LI algorithms are displayed for the four-slice 3 : 1 pitch helical CT (the long-dashed line), thefour-slice 6 : 1 pitch helical CT (the thin solid line), and the single-slice helical CT at any helical pitch (the short-dashed line). The horizontal axis is the time of thetemporal impulset relative to the center of the time window of the projection data used to reconstruct the imaget normalized to the period of a complete gantryrotationT . The full-widths of the TSP’s at 0, 10, and 50% of the maximum are tabulated in Table I.

TABLE ITSP’s, SSP’s,AND NOISE RATIOS OF SEVERAL HELICAL CT PROTOCOLS

The SSP’s of these helical CT imaging protocols were de-rived from (5) and are displayed in Fig. 3. The function isassumed to be a rectangle with a full width of. The full widthsof these SSP’s at 0, 10, and 50% of the maximum were measuredand are tabulated in Table I. It is shown in Fig. 3 and Table I thatthe -resolution performance of the HS-LI algorithm is worse(wider SSP) than that of the 180LI algorithm in the four-slice3 : 1 pitch helical CT and is comparable to four-slice 6 : 1 pitchhelical CT. (The helical pitch is defined as a ratio of the tabletransporting distance per rotation, to the detector row collima-tion .) Corresponding single-slice CT results are also includedin Fig. 3 and Table I, and will be discussed in the next section.

The noise ratio of each helical CT protocol to the corre-sponding full rotation step-and-shoot CT protocol acquiredwith the same beam collimation and mAs setting from the samescanner were derived using the equation given in the Appendix

[4]. They are also tabulated in Table I. The noise ratios ofthe HS algorithms are higher than those of the corresponding180 LI algorithms. This is because to reduce the imagetemporal span, the HS algorithms use less data than the 180LI algorithms, resulting in higher noises.

A set of data from an aorta study using four-slice helical CTwas used to further compare the performance of the HS-LI al-gorithm with the existing 180LI algorithm, in terms of, thetemporal resolution, the-axis resolution, and image artifacts.The data were collected at a gantry rotational speedof 0.8s/rotation and with 3 : 1 helical pitch 11.25 mm/rotation tablespeed (i.e., with 3.75-mm detector row collimation). Two setsof images were reconstructed from the same data set using the180 LI algorithm and the HS-LI algorithm respectively. Fromthe Table I and given that s and mm, theimage temporal duration (the full width at the 0% maximum of


Fig. 3. Slice sensitivity profiles (SSP) of the HS-LI algorithms are shown for the four-slice 3 : 1 pitch helical CT (the dark-thick solid line) and the four-slice6 : 1 pitch helical CT (the light-thick solid line). In comparison, SSP’s of the 180LI algorithms are shown for the four-slice 3 : 1 pitch helical CT (the dark-thinsolid line) and the four-slice 6 : 1 pitch helical CT (overlap with the light–thick solid line). The horizontal axis is the distance to the reconstructed image,z � z ,normalized to the detector row collimationd. For completion, slice sensitivity profiles of single-slice helical CT are displayed for 1 : 1 pitch 180LI (overlap withthe dark-thin solid line), 2 : 1 pitch 180LI (overlap with the light-thick solid line), 1 : 1 pitch HS-NI (the light-thin solid line), and 2 : 1 pitch HS-NI (the dottedline). The full widths of SSP’s at 0, 10, and 50% of the maximum are tabulated in Table I.

Fig. 4. Fig. 4 shows a pair of images of the identicalz or time location reconstructed from the same data set using the 180LI algorithm (left) and the HS-LIalgorithm (right). The image field of view is28� 22 cm and is displayed with a window of [full-width: 350 HU, center: 40 HU]. The aorta in the image appearsto be dissected in the 180LI reconstruction (arrow). The HS-LI reconstruction shows a normal aorta and proves that the aorta dissection appearance is due toartifacts of the 180 LI reconstruction induced by cardiac motions. However, the HS-LI reconstruction shows more shading artifacts (some highlighted by thecircles) and more overlap of structures from adjacent slices (highlighted by the rectangle) than the 180LI reconstruction.

TSP) is 1064 ms for the 180LI reconstruction and 520 ms forthe HS-LI reconstruction, while the slice thickness in terms ofthe full width at the 10% maximum of SSP is 5.9 mm for the180 LI reconstruction and 8.2 mm for the HS-LI reconstruc-tion.

Fig. 4 shows a pair of cm images at the identicalor time location reconstructed with the 180LI algorithm

(left) and the HS-LI algorithm (right). Fig. 5 displays sevenpairs of zoom-in ( cm) images of the 180LI reconstruc-tion (top) versus the HS-LI reconstruction (bottom) at seven lo-

cations with a time spacing of 133 ms, corresponding to aspacing of 1.88 mm. (The time spacing of 133 ms was chosen bygenerating six evenly spaced images over one period of rotationof 0.8 s.) The Figs. 4 and 5 demonstrate that the 180LI imagesfrequently show the artifacts resembling to the aorta dissection.The HS-LI reconstruction reduces the occurrence of this typeof artifacts. This is because the image temporal duration of theHS-LI reconstruction (520 ms) is much shorter than that of the180 LI reconstruction (1064 ms). Thus, the HS-LI reconstruc-tion is more immune to the inconsistent data problem induced by


Fig. 5. Fig. 5 shows seven pairs of zoom-in (8� 7:2 cm) images of the 180LI reconstruction (top) versus the HS-LI reconstruction (bottom) at seven locationswith a time spacing of 133 ms, corresponding to az spacing of 1.88 mm. The 180LI reconstruction frequently show the artifacts resemble to the aorta dissection.The HS-LI reconstruction reduces the occurrence of this type of artifacts.

cardiac motions. However, Fig. 4 shows that the HS-LI recon-struction has more shading artifacts and more overlap of struc-tures from adjacent slices than the 180LI reconstruction. Thisis because by excluding the measurements from the opposingpath, the HS-LI algorithm deals with the measurements that arefurther apart and therefore results in degraded interpolation ac-curacy and SSP comparing to the 180LI.

V. DISCUSSION

In majority of CT applications, such as abdominal or neu-roimaging, the cardiac motion is not a concern and the 180LIalgorithm is a method of choice due to its superior image qualityin terms of compact SSP, low-image artifacts and noise. How-ever, the 180 LI algorithm results in degraded temporal reso-lution performance (wider TSP). For some clinical studies suchas thoracic imaging, the 180LI algorithm may produce specialartifacts induced by cardiac motions, causing potential misdiag-nosis. For this type of applications the HS algorithm should beused due to its improved temporal resolution. With the HS algo-rithm, the temporal resolution of multislice helical CT is com-parable to that of single-slice helical CT.

At a gantry rotational speed (of 0.8 s/rotation), the temporalspan of even the HS reconstruction (520 ms) is still too largeto eliminate the artifacts induced by cardiac motions. Thus,the image quality of the HS reconstruction strongly dependson what phase of a cardiac cycle corresponds to the temporalwindow of the data used to reconstruct the image. When thedata temporal window corresponds to the diastole phase of acardiac cycle, relatively good image quality can be obtained.Otherwise, the image may contain the distortions or theartifacts resembling to the aorta dissection. This observationis confirmed by the variation of image quality of the HS-LIreconstruction in Fig. 5. The temporal resolution of both the180 LI and HS reconstruction will be improved proportionallyas the gantry rotational speed further increases.

In practice, the 180LI and HS-LI algorithms can be usedin parallel to generate two image sets from the same scan ac-quisition, one (180 LI) for improved SSP and reduced imagenoise, and the other (HS-LI) for improved image temporal reso-lution. In particular, when the 180LI reconstruction shows thestructure resembling to the aorta dissection, one should always

examine multiple HS images reconstructed at the adjacent-lo-cations with a temporal spacing of a fraction of one second. Thestructure may be the artifacts of the 180LI reconstruction if itdisappears in the majority of HS images.

The variation of image quality of the HS reconstruction (asshown in Fig. 5) will be substantially reduced if the ECG dataare collected with the helical CT data and are used to chooseboth and temporal locations of the reconstructed images torepresent the same cardiac phase [12]–[15].

The HS algorithm has been proposed for single slice helicalCT [2], [12], [13]. Since the single slice CT can only afford thenearest neighbor interpolation within each projection view, theHS algorithm is a HS-NI algorithm. The imaging performances(TSP’s, SSP’s and the noise ratios) of single-slice helical CTwith HS-NI are also included in this study (refer to Figs. 2 and 3,and Table I). Lacking a properinterpolation, the HS-NI recon-struction, though having slightly more compact SSP’s, containsmore image artifacts and higher noises than other imaging pro-tocols. In multislice helical CT, as simultaneous measurementsacquired by multiple detector rows within each projection vieware available and as the patient transporting speed is typicallyseveral times faster than that in single-slice CT, incorporationof linear interpolation (LI) becomes possible and necessary, andtherefore, the nearest neighbor interpolation (NI) is not used.

The halfscan algorithm has been proposed for various typesof approximate CB helical CT reconstruction algorithms [17],[18] to improve SSP. With the advent of the four-slice helicalCT, the multislice HS-LI reconstruction algorithm has also beenproposed by other researchers [15], besides the authors. Notethat the recent research is aimed at improving TSP, with a de-graded SSP.

In this study, the TSP and SSP of the HS algorithm were com-puted assuming the fan-angle to be 54 , corresponding toa scan field of view (FOV) of 50 cm in diameter on the Light-Speed scanner (General Electric Company). As has been dis-cussed by Kachelriess and Kalender [13], since the cardiac mo-tions are usually contained within only the central region of thescan FOV, a much smaller can be used for the temporal per-formance consideration. To investigate the impact of theonimaging characteristics, we also considered a theoretical sce-nario where the is 0 . In this case and in comparison to theresults shown in Figs. 2 and 3, the TSP of the HS algorithm is


in a rectangular shape with a full-width of 0.50. The SSP’s ofthe HS algorithm are similar to those shown in Fig. 3 becausethe variation in due to the change is mostly averaged outafter the convolution with in (6).

The general TSP model (8) is derived at the center location.Analyzing TSP at an off-centered location (i.e., ) is muchmore complex, similar to analyzing SSP at an off-centered lo-cation [19]. As previously discussed, the actual temporal per-formance has more to do with the TSP at the center than thatnear the edge of the scan FOV, because the cardiac motions areusually contained within a central region.

The proposed HS algorithm for multislice helical CT recon-struction is applicable to any CT systems with any number ofdetector rows and any helical pitch. In this study, we presentthe theoretical and clinical results for a four-slice helical CTat pitches of 3 : 1 and 6 : 1 as examples. Recent studies indi-cated that cardiac applications might benefit from lower helicalpitches. The imaging performance at these helical pitches canbe studied in a similar fashion.

The -filtering reconstruction algorithm has been proposedfor single-slice [20] and multislice [4], [7] helical CT for an im-proved tradeoff of -resolution versus image artifacts and noise.However, the -filtering reconstruction algorithm results in de-graded temporal resolution and therefore substantial-filteringshould not be used in cardiac applications.

VI. CONCLUSION

A multislice helical CT with HS-LI reconstruction algorithmhas been proposed for cardiac applications. The HS reconstruc-tion results in improved image temporal resolution than the cor-responding 180LI reconstruction and is more immune to theinconsistent data problem induced by cardiac motions. The tem-poral resolution of a four-slice helical CT with the HS-LI algo-rithm is comparable to that of single slice helical CT with theHS-NI algorithm. In practice, the 180LI and HS-LI algorithmscan be used in parallel to generate two image sets from the samescan acquisition, one (180LI) for improved SSP and reducedimage noise, and the other (HS-LI) for improved image tem-poral resolution.

APPENDIX

The formula used to predict the noise ratio of helical tostep-and-shoot CT, denoted as , is given as follows [4]:

(A1)

where is the detector row index, ranging from 1 tostands for the helical interpolation function at the center detectorchannel of the th detector row, denotes the gantry rotationalangle.

ACKNOWLEDGMENT

The authors would like to thank Dr. H. Rigauts from Depart-ment of Radiology, St. Jan Hospital, Belgim, for providing theclinical data for study shown in Figs. 4 and 5.

REFERENCES

[1] W. A. Kalender, W. Seissler, E. Klotz, and P. Vock, “Spiral volumetricCT with single-breath-hold technique, continuous transport, and contin-uous scanner rotation,”Radiology, vol. 176, no. 1, pp. 181–183, 1990.

[2] C. R. Crawford and K. F. King, “Computed tomography scanningwith simultaneous patient translation,”Med. Phys., vol. 17, no. 6, pp.967–982, 1990.

[3] J. S. Areson, R. Levinson, and D. Freundlich, “Dual slice scanner,”,1993.

[4] H. Hu, “Multi-slice helical CT: Scan and reconstruction,”Med. Phys.,vol. 26, no. 1, pp. 5–18, 1999.

[5] H. Hu, H. He, and S. Foxet al., “Imaging characteristics and tradeoff se-lections of four-slice CT: Theoretical and experimental results,” inSci-entific Assembly Annu. Meet. Radiological Soc. North America, 1998,p. 283.

[6] H. Hu, H. He, W. Foley, and S. Fox, “Helical CT imaging performanceof a new multislice scanner,” inProc. SPIE Medical Imaging, vol. 3661,San Diego, CA, 1999, pp. 450–461.

[7] K. Taguchi and H. Aradate, “Algorithm for image reconstruction in mul-tislice helical CT,”Med. Phys., vol. 25, no. 4, pp. 550–561, 1998.

[8] D. P. Boyd and M. J. Lipton, “Cardiac computed tomography,”Proc.IEEE, vol. 71, p. 281, 1983.

[9] A. S. Agatston, W. R. Janowitz, F. J. Hildner, N. R. Zusmer, M. ViamonteJr., and R. Detrano, “Quantification of coronary artery calcium usingultrafast computed tomography,”J. Amer. Coll. Cardiol., vol. 15, no. 4,pp. 827–832, 1990.

[10] C. H. McCollough, R. B. Kaufmann, B. M. Cameron, D. J. Katz, P. F.Sheedy, 2nd, and P. A. Peyser, “Electron-beam CT: Use of a calibrationphantom to reduce variability in calcium quantitation,”Radiology, vol.196, no. 1, pp. 159–165, 1995.

[11] J. Shemesh, S. Apter, and J. Rozenmanet al., “Calcification of coronaryarteries: Detection and quantification with double-helix CT,”Radiology,vol. 197, no. 3, pp. 779–783, 1995.

[12] C. E. Woodhouse, W. R. Janowitz, and M. Viamonte Jr., “Coronary ar-teries: Retrospective cardiac gating technique to reduce cardiac motionartifact at spiral CT,”Radiology, vol. 204, no. 2, pp. 566–569, 1997.

[13] M. Kachelriess and W. A. Kalender, “Electrocardiogram-correlatedimage reconstruction from subsecond spiral computed tomographyscans of the heart,”Med. Phys., vol. 25, no. 12, pp. 2417–2431, 1998.

[14] H. Hu, J. Carr, C. Woodhouse, W. Janowitz, R. Buchanan, and T. Mitsa,“Technical evaluation of the calcification scoring using general purposehelical CT with retrospective gating,” inScientific Assembly Annu. Meet.Radiological Soc. North America, 1998, p. 323.

[15] M. Kachelriess and W. A. Kalender, “Imaging of the heart by ECG-oriented reconstruction from subsecond spiral multi-row detector CTscans,” inScientific Assembly and Annu. Meet. Radiological Soc. NorthAmerica, 1998, p. 323.

[16] D. L. Parker, “Optimal short scan convolution reconstruction for fan-beam CT,”Med. Phys., vol. 9, no. 2, pp. 254–257, 1982.

[17] G. Wang, Y. Liu, T. H. Lin, and P. C. Cheng, “Half-scan cone-beamx-ray microtomography formula,”Scanning, vol. 16, no. 4, pp. 216–220,1994.

[18] F. Noo, M. Defrise, and R. Clackdoyle, “Single-slice rebinning methodfor helical cone-beam CT,”Phys. Med. Biol., vol. 44, no. 2, pp. 561–570,1999.

[19] H. Hu and S. H. Fox, “The effect of helical pitch and beam collimation onthe lesion contrast and slice profile in helical CT imaging,”Med. Phys.,vol. 23, no. 12, pp. 1943–1954, 1996.

[20] H. Hu and Y. Shen, “Helical CT reconstruction with longitudinal filtra-tion,” Med. Phys., vol. 25, no. 11, pp. 2130–2138, 1998.

multislice helical ct: image temporal resolution

Documents