predictive respiratory gating: anew method toreduce motion
TRANSCRIPT
Cameron J. Ritchie, PhD #{149}Jiang Hsieh, PhD #{149}Michael F. Gard, PhD #{149}J. David Godwin, MDYongmin Kim, PhD #{149}Carl R. Crawford, PhD
Predictive Respiratory Gating: A New Methodto Reduce Motion Artifacts on CT Scans’
I From the Departments of Bioengineering
(C.J.R.), Radiology (J.D.G., C.R.C.), and Electrical
Engineering (Y.K.), University of Washington,
SB-05, Seattle, WA 98195; and GE Medical Sys-tems, Milwaukee, Wis OH., M.F.G., C.R.C.). Re-ceived July 27, 1993; revision requested Septem-ber 22; revision received November 2; accepted
November 8. Supported in part by GE MedicalSystems and the Keck Foundation. Address re-print requests to C.J.R.
, � RSNA, 1994
847
PURPOSE: To evaluate a gating sys-tern, called predictive respiratorygating (PRG), that reduces motion-induced artifacts on computed tomo-graphic (CT) scans of patients whocannot suspend respiration.
MATERIALS AND METHODS: PRGuses a respiration monitor and a newalgorithm to predict when a motion-less period is about to occur. It auto-matically starts scanning so the scanis temporally centered around themotionless period at end inspirationor end expiration. To demonstratePRG, CT was performed on a motionphantom and a quietly breathing vol-unteer with and without gating.
RESULTS: Scans of the phantom ob-tamed with PRG contained less mo-tion-induced streaking and blurringthan did scans acquired withoutPRG. Scans of the volunteer gated atend expiration contained signifi-cantly less artifact than nongatedscans (P < .03).
CONCLUSION: PRG reduced motionartifact on scans of a spontaneouslybreathing volunteer. PRG may beable to reduce motion artifacts onscans of patients unable to suspendrespiration.
Index terms: Computed tomography (CT),artifact . Computed tomography (CT),
physics . Computed tomography (CT), tech-
nology
Radiology 1994; 190:847-852
R ESPIRATORY motion during com-puted tomography (CT) causes
artifacts (1,2) that can mimic diseaseand lead to misdiagnosis (3). Theseartifacts are particularly frequent onscans of unconscious or pediatric pa-
tients who are breathing spontane-
ously; however, they occur even on
scans of conscious, cooperative pa-tients (4). Herein, we describe a scan-ning method, called predictive respi-ratony gating (PRG), that reduces
respiratory motion artifacts on scansof spontaneously breathing patients.PRG uses a respiration monitor and anew algorithm to predict when a mo-tionless period is about to occur. It
then automatically starts scanning sothat the scan is temporally centered
around the quiescent periods thatoccur at end inspiration or end expi-ration.
Other methods for reducing nespi-ratory motion artifacts at CT havebeen proposed. One method is to usea scanner with a very short (less than100 msec) scan time; however, such
short scan times lead to lower signal-
to-noise ratios than those providedwith conventional (1 second) scantimes (5), and thus conventional scan-
ners are preferred for CT of the chest.
Another method is to use a modified
filtered back-projection algorithm thatcorrects for the motion that occursduring scanning; however, thismethod is limited because a techniquefor estimating this motion has notbeen developed (6).
A third method is to use prospec-
tive gating to collect segments of the
required projection data over several
scans in which a different segment ofeach scan is acquired during a quies-cent period in respiratory or cardiacmotion (7). This multiple-scan methodis not used because it increases the
radiation dose and the time requiredfor examination. To our knowledge,PRG with a single scan has not beenaccomplished with CT because scantimes are longer than the quiescent
periods in respiration. Kalender et at(8) acquired full scans during nespina-
tion, but they artificially prolongedthe quiescent period by mechanicallyarresting the subject’s respiration.
Crawford et at (9) described a PRGmethod for CT that did not requiremultiple scans and in which artifactswere reduced even when the scantime was longer than the quiescentperiod. With computer simulations,artifacts were shown to be reducedwhen the acquisition of a full scanwas timed so the midpoint of thescan coincided with the midpoint of aquiescent period (ie, scanning was“centered” on the quiescent period).Crawford et at recognized scanningmust be started before the quiescentperiod in order to center the scan.Therefore, they proposed that theoccurrence of the quiescent periodcould be predicted by assuming thatrespiratory motion was periodic. Thiskind of PRG was not applied on a realscanner.
Recently, slip-ring CT scannerswith 1-second scan times have be-come widely available. The quiescentperiods at end expiration (1-2 sec-onds) and end inspiration (less than 1second) are approximately equal tothese scan times (10). In PRG, we ap-ply the concept of centering the scanon a quiescent period to these fasterscanners. To deal with aperiodicitiesin respiratory motion (11), we me-placed the assumption of periodicrespiration with an algorithm thatpredicts when a quiescent period isabout to occur and then appropriatelytimes the start of scan acquisition.
MATERIALS AND METHODS
We first characterized respiratory mo-tion in a small group of volunteers to as-
Abbreviations: AMC = adaptive movingcorrelation, LVDT = linear variable differentialtransducer, PRG = predictive respiratorygating.
Figure 1. Motion phantom. The phantom consists of three stepping motor and table units.
The units are connected so that simultaneous motions along the x, y, and z axes can be speci-fled. The object to be scanned is attached to the long arm (solid arrow). In this photograph,the test object is a plastic cone (open arrow). The motion of each table is controlled by a spe-cialty designed circuit board and a computer (not shown).
848 #{149}Radiology March 1994
certain the periodicity of spontaneousbreathing. Next, we developed a predic-
tion algorithm for computing the time atwhich quiescent periods would occur. We
then combined the prediction algorithmwith a respiration monitor and a methodfor automatically starting the scanner. Theresultant system, which we designatedPRG, was tested by scanning a motionphantom and a volunteer.
Characterization of RespiratoryMotion
We characterized the anteroposteniommotion of the chest wall in a small groupof volunteers by monitoring the positionof the anterior wall of the chest. We usedtwo different motion monitors to track theposition of the chest watt. In this part ofour study, an infrared rangefinder (model214 Laser Rangefinder; Spectronics, Bea-
verton, Ore) was used.
Seven healthy men aged 26-50 yearswere studied. Each subject lay supine, anda plastic disk was placed on the subject’sclothing over the xiphoid process. The
rangefinder was placed 10 cm above thedisk. Each subject was asked to breathnormally and did so for several minutesbefore data collection was begun. Datacollection was begun without the subject’sknowledge (to ensure that the measure-ment was of spontaneous breathing) andwas continued for 40 seconds.
Plots of chest-wall position versus time
(designated “respiratory waveforms”)were generated. For each waveform, themean and standard deviation of the pe-nod of the respiratory cycle, the averagedifference between adjacent periods, andthe maximum difference between periodswere computed.
Prediction Algorithm
Preliminary experiments demonstratedthat artifacts would not be reduced if themidpoint of a scan occurred more than0.25 seconds from the midpoint of a quies-
cent period (Appendix). From our motioncharacterization, we found that differ-ences between the longest breath and theshortest breath were greater than thistime. Therefore, respiratory motion couldnot be modeled as periodic for the pur-poses of computing the time at which tostart the scanner, and thus the assumptionthat respiratory motion was periodic wasdiscarded.
Consequently, we developed a predic-tion algorithm that enabled us to analyzethe respiratory waveform and predict thetime at which a quiescent period wouldnext occur. The algorithm then computedwhen the scanner should be started so thescan would be centered around the quies-cent period. The prediction algorithm(designated adaptive moving correlation[AMCJ) exploits the fact that, whereas thequiescent periods of the respiratory wave-form do not occur in a strictly periodicfashion, the shape of the inspiratory orexpiratory segment of the waveform is
similar from breath to breath. The algo-rithm is adaptive in that it adjusts to thewaveform being analyzed. The AMC algo-rithm is described in more detail in theAppendix.
PRG System
We integrated a respiratory monitorwith the AMC prediction algorithm and aCT scanner. In this part of the study, weused a linear variable differential trans-ducer ([LVDT] ACTI000C; RDP Electron-ics, Pottstown, Pa) to track the position of
the chest wall. The linear resolution of theLVDT was comparable to that of the infma-red rangefinder.
The output of the LVDT was monitoredand analyzed by a computer (Sun 3/60;Sun Microsystems, Mountain View, Calif)running the AMC program. The worksta-tion read in the measurements of the chestwall position from the motion monitor,performed the prediction calculations, andthen generated the signal to start the scan-ner. When scanning the motion phantom,we modified the start-scan button on thescanner console (9800 HiLight Advantage;GE Medical Systems, Milwaukee, Wis) sothe scanner could be started automaticallyby the signal from the AMC program.When scanning the volunteer, we did notalter the start-scan button on the scanner(HiSpeed Advantage, GE Medical Sys-tems) because it was not feasible to modifyour busy clinical scanner.
Motion Phantom
We tested the PRG system by scanninga motion phantom (12) (Fig 1). The GE
9800 scanner was used because it wasavailable for the modifications needed tocontrol the start of scanning. Respiratorywaveforms measured from three of the
subjects monitored during the character-
ization of respiratory motion were repro-duced with the motion phantom. The fast-est scan time was 2.0 seconds, but scan
times on the order of the quiescent period
in respiration (1.0 second) were required
to test PRG. Therefore, we decreased thefrequency at which the motion phantomreproduced the subject’s motion to one-fourth that of real respiration, and we pro-portionately increased the scan time to 4.0seconds. Therefore, 1.0-second scans weresimulated.
A section of dried, inflated human lungwas used as the test object, and the probeof the LVDT was placed on the moving
arm of the motion phantom. Four-secondscans were obtained with 3-mm collima-
tion at 120 kVp and 40 mA and were re-
constructed into a 512 x 512 image with a15-cm field of view. The standard recon-
struction algorithm without underscan
(also known as the peristalsis option) wasused. We first scanned the lung sectionwhile it was stationary. Next, we scannedthe lung section during inspiration whenthe phantom velocity was at maximum.Finally, we allowed PRG to control the
scanning of the moving section.
Breathing Volunteer
To demonstrate the clinical feasibility ofPRG, we performed CT in a healthy 47-year-old male volunteer and had a panel
of radiologists evaluate the images. PRGwas used retrospectively because we did
Volume 190 #{149}Number 3 Radiology #{149}849
a.
V
C. d.
Figure 2. CT scans show grades 0 (a), I (b), 2 (c), and 3 (d) artifacts. Motion-induced streaking (open arrows) grows progressively worse fromimages b to d. Doubling can be seen in d (solid arrows).
not modify the scanner for automaticscanning; however, retrospective gatinghad the benefit of providing both gatedand nongated images for comparison fromthe same projection set.
The LVDT probe was placed high onthe patient’s chest to ensure it did not cre-
ate artifacts, and the volunteer’s respira-lion was measured for 5 minutes. The CTscan was obtained through the lung base,
but clear of the diaphragm. Four minutesinto the continuous LVDT measurement,the volunteer was scanned continuouslyfor 30 seconds at the same location with120 kVp, 40 mA, and 5-mm collimation.Images were reconstructed from the 30
seconds of scan data at 0.1-second inter-vats (291 images total) by using a 512 x
512 matrix with a 48-cm field of view. Thestandard reconstruction algorithm withand without underscan was used. Lowmilliamperage was used to limit the doseto the volunteer, but this low value didnot affect our results because scans withlow milliamperage have been shown toprovide chest images of sufficient qualityfor demonstrating alt but low-contrast de-tails (13).
We retrospectively analyzed the respira-tory waveform of the volunteer with AMC
and computed the start-scan times that
would have centered projection acquisi-
tion over the quiescent periods. For each
start-scan time, we selected the images
acquired at the corresponding times fromthe set of 291 images. This set of selectedimages represented the images that would
have been acquired if PRG had been oper-ating in real time and had started the
scanner prospectively. AMC selected im-ages gated at both end inspiration and
end expiration.The volunteer’s respiratory waveform
was analyzed, and the lengths of the qui-
escent periods were measured. Quiescent
periods were defined as those periods
around end inspiration and end expira-tion in which the motion was less than theminimum detectable motion as deter-
mined in a previous study (14).To determine if the gated images were
of significantly higher quality than thoseacquired without gating, a panel of fourexperienced chest radiologists scored 50images (39 nongated, five gated at end
inspiration, and six gated at end expira-
tion). Nongated images were chosen ran-domly from the set of 291 images and rep-
resented possible outcomes if a patientwho could not suspend respiration was
scanned without attention to the phase of
the respiratory cycle. A different set ofnongated images was selected for eachradiologist.
For each radiologist, the 50 images wereplaced randomly onto three sheets of filmand were scored on a scale of 0 (no anti-fact) to 3 (severe artifact). Each radiologistwas also provided with a set of standard
images obtained from the same volunteerrepresenting each grade of motion anti-fact against which to compare the experi-mental images (Fig 2). These benchmarkimages had been selected by a fifth radi-ologist U.D.G.). Images gated at end inspi-ration and end expiration were compared
with nongated images, and P values werecomputed for each comparison by usingthe nonpamametric Mann-Whitney test(15).
RESULTS
Characterization of RespiratoryMotion
In the seven volunteers, plots ofrespiratory motion demonstrated
variations in period and amplitudefor each subject. One subject demon-
�‘Y\JLf�k�t%�IJ\f
a li) 20 34) 40 � �#{216} 20 30 4() 0 () 20 0) 40
Time (sec) Time (sec) Time (st-c)
3. 4. 5.
Figures 3-5. (3) Respiration waveform for subject 1. The peaks correspond to inspiration, and the valleys correspond to expiration. Nearlyconstant period and variable end-inspiratory and end-expiratory chest-wall positions are seen. (4) Respiration waveform for subject 2. A nearlyconstant period and end-inspiratory chest-walt position are seen. (5) Respiration waveform for subject 3. A variable period and variable end-inspiratory chest-wall position are seen.
850 #{149}Radiology March 1994
E
S
strated nearly periodic motion and
moderately variable motion ampli-tude (Fig 3). In another subject, theperiod and end-inspinatomy chest-wallposition were nearly constant overseveral breaths, but end-expiratoryposition was not (Fig 4). In a thirdsubject, the period changed, and
there was substantial variation inchest-wall position at end inspiration(Fig 5).
The average period of a breath, thestandard deviation in the period of abreath, the mean difference betweenadjacent periods, the difference be-tween the longest and shortest pen-ods, and that difference expressed as
a percentage difference of the shortest
period were computed for eachsubject’s waveform (Table 1). The
standard deviation in the period of abreath ranged from 0.22 to 0.72 sec-onds. The difference between the
longest and shortest periods ranged
from 0.52 to 1.78 seconds.
Motion Phantom
The motion phantom was pro-grammed to simulate the breathing ofsubjects 1, 2, and 3 (Table 1). Images
of the lung section obtained withoutmotion were free of artifact andshowed expected detail (Fig 6a). Im-ages of the phantom simulating the
respiratory motion of subject 1, ob-
tamed during the inspiratory portionof the waveform, had substantial mo-tion artifacts, including doubling ofsmall vessels, black voids, and thick
white streaks around the high-attenu-ation structures (Fig 6b). Moving thesection in the pattern of the mespira-
tony motion of subject 1 but using
PRG to start the scanner resulted inan image with reduced streaking anddoubling (Fig 6c). Similar results were
C
C
S
obtained when the phantom was pro-grammed to simulate the respiratorymotion of subject 2. Use of PRG onthe respiratory motion of subject 3yielded less artifact reduction than for
subjects 1 and 2, but the image did
contain less artifact than the non-gated scan.
Breathing Volunteer
The volunteer’s respiration wave-form obtained in the 30-second scanperiod was recorded and analyzed.The volunteer’s respiratory rate wasabout 12 breaths pen minute, and theaverage length of the quiescent pen-ods at end inspiration and end expina-
tion were 0.63 (r = 0.12) and 1.42
(if = 0.28) seconds, respectively.
The degree of motion artifact oneach image was graded by the panelof radiologists. The mean scores of thenongated, gated at end expiration,and gated at end inspiration imageswere 1.09, 0.11, and 2.23, respec-tively (Table 2). The use of underscanweighting did not substantially affectthe image scores. Each of the four ma-diologists scored the images gated at
end expiration as containing less anti-
C
C
0
0
fact than the nongated images (P <
.034) (Table 2). Images gated at end
inspiration, conversely, scored signifi-
cantly worse than nongated images
(P < .052) (Table 2).
This poor scoring was due to anti-
facts caused from a coincidental syn-
chmonization of cardiac systole withend inspiration. Three of the five end-inspiration scans were obtained at
systole, when cardiac motion artifacts
are most intense, whereas none of the
end-expiration scans were obtained atsystole. The occurrence of systole was
indicated by the doubled appearanceof the left ventricular wall on the
gated images. Scans gated at end in-
spimation in which systole occurredreceived a mean score of 2.45, whilescans at end inspiration in which sys-tote did not occur received a mean
score of 1.85.
DISCUSSION
To our knowledge, theme has been
no effective strategy for reducing anti-fact on scans of spontaneously breath-
ing patients. Theme has also been noeffective strategy for ensuring thatcooperative patients have complied
Table 2
Scores for Gated and Nongated Images
Mean Scores P Value
Radiologist NongatedGated at
InspirationGated at
ExpirationNongated vs
Gated at InspirationNongated vs
Gated at Expiration
1 0.98 (0.77) 1.84 (0.39) 0.00 (0.00) .010 .0022 0.89 (0.84) 2.26 (0.73) 0.00 (0.00) .004 .0063 1.34 (1.06) 2.36 (0.56) 0.38 (0.49) .052 .0344 1.15 (1.00) 2.44 (0.38) 0.05 (0.12) .007 .001
Mean 1.09 2.23 0.11 . . . ...
Note.-Numbers in parentheses are standard deviations.
a. b. C.
Volume 190 #{149}Number 3 Radiology #{149}851
Figure 6. (a) CT scan of dry lung phantom with no motion. (b) CT scan oflung phantom with motion simulating that of subject 1. (c) Opti-
mally gated CT scan of lung phantom with motion simulating that of subject I.
with the scanner technician’s instmuc-
tion to hold their breath. We have
demonstrated that scans with sub-
stantially reduced artifact can be ob-
tamed from spontaneously breathing
patients by using PRG. Additionally,the use of PRG could eliminate the
need for patients to hold their breath.
Finally, PRG could be used to im-prove registration of contiguous scans
for volume display.The PRG system consists of a mespi-
ration monitor, software, and modifi-
cations to the scanner start-scan but-
ton. Because PRG requires onlyminimal scanner modifications, it
should be applicable to scanners with
sufficiently short scan times so an en-
time scan can be acquired during a qui-
escent period.
PRG was performed in both a mov-ing lung slice and a volunteer. PRGscans of the moving lung slice dem-onstmated reduced motion artifact
compared with nongated scans fortwo of the three subjects’ respiration
waveforms. PRG scans gated at endexpiration demonstrated reduced mo-
tion artifact compared with nongated
scans. Although radiologists generally
prefer to acquire CT scans at full in-
spination, we found that images gated
to end tidal volume adequately pre-
served lung detail and were superior
to nongated images even though thenongated images were obtained at
greater lung volumes.
We were unable to gate all the
phantom scans appropriately withPRG because we performed ourphantom experiments on a scanner
with a long scan time (4.0 seconds)
and a long scanner delay (2.26 sec-onds). (Scanner delay is the time be-
tween pressing the start-scan button
and the start of projection acquisi-
tion.) These long delays cause errorsin the prediction of the time the qui-escent period in breathing will occur.Although prediction accuracy suffers
somewhat when using a scanner with
a long scan time and a long scanner
delay, some artifact reduction can still
be obtained by using PRG with this
type of scanner, as shown by our me-
sults with subjects 1 and 2.
The ability of PRG to reduce motionartifacts depends on the respiration
waveform of the patient. We tested
PRG on waveforms with respiratoryperiods as short as 4.18 seconds. For
respiratory periods much shorter than
this, PRG scans will not demonstrate
as much motion artifact reduction
because the quiescent periods will be
necessarily shorter. However, because
normal respiration has a period ofapproximately 4 seconds (10), the
PRG system should perform satisfac-tonily with most patients.
We did not address artifacts caused
by cardiac motion. Cardiac motion
artifacts could be eliminated by gating
both on the cardiac cycle and on the
respiratory cycle (9); however, the
quiescent period in cardiac motion (at
end diastote) is too short for conven-tional scanners with 1-second scan
times. Fortunately, cardiac motion is
only a problem when scanning near
the diaphragm. Cardiac motion is notas great a problem for scans of the
abdomen or the upper chest.
For PRG to gain wide clinical accep-tance, a clinically suitable respirationmonitor is needed. The monitor must
be noninvasive and easy to use and
should not cause artifacts if it inter-sects the scan plane. The infrared and
LVDT respiration monitors used inthis study were easy to use. They
were also able to monitor the chestmotion from well outside the scanplane because the AMC algorithm
852 #{149}Radiology March 1994
does not base its prediction on theamplitude of the respiration signal.
We report a promising new methodfor scanning patients who do not sus-pend breathing. Artifacts were signifi-cantly reduced in scans gated at endexpiration in a volunteer. These pre-liminary results need confirmation ina larger number of subjects, includingpatients who cannot suspend breath-ing for a variety of reasons.
APPENDIX
Determination of Maximum Error
In PRG, the timing of the start of scanacquisition is based on the time a quies-cent period is predicted to occur. Craw-ford et at (9) proposed this predictioncould be made by assuming respiration isperiodic over the course of several breaths.To ascertain if this assumption was valid,we determined how much time couldseparate the midpoint of the quiescentperiod and the midpoint of the scan be-
fore artifacts began to appear on images.This time is designated the allowableerror.
To determine the allowable error, a plas-tic peg (1-cm diameter) was scanned 16times while it moved in a pattern that re-produced the motion of a single cycle ofrespiratory motion as measured in subject1 (Table 1). The midpoints of the 16 scanswere equally spaced in time through the4-second respiratory cycle. Streaking anti-facts on the images of the peg were quan-titated by measuring the standard devia-
tion in a region of interest placed far fromthe peg on each image. Images that wereoptimally centered on end inspiration or
end expiration demonstrated minimalstandard deviation measurements. Theallowable error was equal to the amountof time that separated these optimally cen-tered images from the first image that con-tamed visible motion artifacts.
The maximum errors for expiration andinspiration were 0.5 and 0.25 seconds, re-spectively. From Table 1, the differencesbetween the shortest period and the long-est period in respiratory motion rangedfrom 0.52 to 1.78 seconds and thus ex-
ceeded the maximum errors for both in-spiration and expiration. Therefore, reli-ably gated images at either end inspirationor end expiration could not be obtained ifpeniodicity was assumed when perform-ing start-scan time calculations.
Description of AMC
The AMC algorithm computes the timeat which quiescent periods in the respira-tory waveform will occur by computingthe correlation coefficient between an mi-tial segment F and the most recent mea-surements C of the respiratory waveform,where F and G are vectors. A correlationcoefficient equal to 1 indicates that thevectors are identical, while a coefficient ofzero indicates that the vectors are unre-tated.
We first measure the position of thechest watt for several breaths. The vectorG (called the kernel) is formed from thisinitial btock of data and is equal to thosevalues of the waveform that immediatelypreceded the quiescent period. The lastpoint in the kernel, kend, is positioned one-
half the scan time plus the scanner-delaytime before the midpoint of the quiescentperiod. A vector F is formed each time anew measurement is returned from therespiration monitor, and a correlation co-efficient is computed between F and G.When the correlation coefficient becomesgreater than 0.99, it indicates that thepoint corresponding to kend in the nextrespiratory cycle has just been measured,and thus AMC outputs a start-scan signal.
When a start-scan signal is output, anew kernel is formed by averaging F withG. This averaging allows AMC to adapt tochanges in the breathing pattern overtime. The averaging step is constrained soonly measurements representative of nor-mat respiration are averaged into the newkernel. This constraint ensures that faultymeasurements (ie, measurements obtainedduring a cough or sneeze) are filteredout. #{149}
Acknowledgments: We thank Eric Stern, MD,Thomas Winter, MD,Jutie Takasugi, MD, andStephen Marglin, MD, for scoring the imagesand Jack Estrada, RT, for operating the scanner.
References
2. Mayo JR. Muller NL, Henkelman RM.The double-fissure sign: a motion artifacton thin-section CT scans. Radiology 1987;
165:580-581.
3. Tarver RD. Conces DJ, Godwin JD. Mo-tion artifacts on CT simulate bronchiectasis.AJR 1988; 151:1117-1119.
4. Dupuy DE, Costello P. Ecker CP. SpiralCT of the pancreas. Radiology 1992; 183:815-818.
5. Gould RG. Principles of ultrafast corn-puted tomography: historical aspects,mechanism of action, and scanner charac-teristics. In: Stanford W, Rumberger JA,eds. Ultrafast computed tomography incardiac imaging: principles and practice.Mount Kisco, NY: Futura, 1992; 1-15.
6. Ritchie CJ, Kim Y, Crawford CR, GodwinJD. CT motion artifact correction usingpixel-specific back-projection. Proc IEEEEng Med Biot Soc 1992; 14:1782-1783.
7. Moore SC, Judy PF, GarnicJD, Karnbic CX,Bonk F, Cochran C. Prospectively gatedcardiac computed tomography. Med Phys1983; 10:846-855.
8. Kalender WA, Fichte H, Bautz W, Skatej M.Semiautomatic evaluation procedures forquantitative CT of the lung. J Cornput As-sist Tornogr 1991; 15:248-255.
9. Crawford CR, Godwin JD, PeIc NJ. Re-duction of motion artifacts in computedtomography. Proc IEEE Eng Med Biol Soc1989; 11:485-486.
10. Buikman D, Helzel T, R#{246}schrnann P. TheRF coil as a sensitive motion detector formagnetic resonance imaging. Magn ResonImaging 1988; 6:281-289.
11. Vannier MW. Respiratory gating by irn-pedance plethysmography. J Nuct Med1984; 25:1142-1143.
12. Ritchie CJ, Peterson E, Yee D, Kim Y, God-win JD, Crawford CR. A 3-D motion con-trol system for simulation of CT motionartifacts. Proc IEEE Eng Med Biot Soc 1989;11:487-488.
13. Naidich DP, Marshall CH, Gribben C,Arams RS, McCauley DI. Low-dose CT ofthe lungs: preliminary observations. Radi-otogy 1990; 175:729-731.
14. Ritchie CJ, Godwin JD, Crawford CR, Stan-ford W, Anno H, Kim Y. Minimum scanspeeds for suppression of motion artifactsin computed tomography. Radiology 1992;185:37-42.
15. ZarJH. Biostatistical analysis. Englewood
Cliffs, NJ: Prentice-Hall, 1974; 109-114.
1. Shepp LA, Hilal 5K, Schulz RA. The tun-ing fork artifact in computerized tomogra-
phy. Comput Graph Imaging Processing1979; 10:246-255.