The recent introduction of multirow detector computed

ORIGINAL ARTICLE Gauging Effective Spatial Resolution in Multirow Helical Cardiac Computed Tomography With a Dynamic Phantom Friedrich D. Knollmann, MD,* Tarkan Cangöz, MA,* Erdogan Cesmeli, PhD, Thomas Toth, Peter Edic, PhD, Johannes Müller, PhD, MD, and Roland Felix, MD, PhD* Purpose: To devise a numerical indicator of image quality for multirow helical cardiac computed tomography (CT) and its relation to temporal resolution. Materials and Methods: A pulsatile cardiac assist device was used to simulate cardiac wall motion by mechanically transmitting the device dynamics to a piece of tungsten wire. Wire motion induced by different device rates was captured with an 8-row subsecond helical CT scanner operating with various scanning parameters. Image artifacts were visually assessed and compared with the image point spread function (PSF) using the full width at half maximum (FWHM) area as a numerical estimate of spatial accuracy. Results: At rest, the FWHM area was determined as 1.3 mm 2.Ata device rate of 60 bpm, the FWHM area ranged from 1.51 mm 2 to 21.62 mm 2, depending on the time of image reconstruction. Mean reproducibility of the FWHM area measurements was determined as 0.05, whereas visual estimates of motion artifact were highly variable between different readers (kappa 0.19). Visually determined image quality correlated closely with the FWHM area metric (Spearman s rank correlation, P 0.0001, rho 0.841). At a device rate of 100 bpm, the minimum FWHM area was 2.00 mm 2 using a single-sector algorithm, 1.41 mm 2 using a 2-segment algorithm, and 1.37 mm 2 using a 4-segment algorithm. Conclusions: Use of a pulsatile cardiac assist device could serve as an in vitro test bed for cardiac CT imaging methods. Area FWHM of the PSF correlates well with visually determined image quality of a dynamic phantom, but provides better reproducibility than visual analysis. Received September 29, 2002 and accepted for publication, after revision, July 28, 2003. From the *Department of Radiology, Charité, Campus Virchow-Klinikum, Humboldt-University, Berlin, Germany; GE Global Research, Niskayuna, New York; GE Medical Systems, Milwaukee, Wisconsin; and the Department of Cardiothoracic and Vascular Surgery, German Heart Institute, Berlin, Germany. Reprints: Friedrich D. Knollmann, MD, Strahlenklinik, Charité, Campus Virchow-Klinikum, Augustenburger Platz 1, 13353 Berlin, Germany. E-mail: Friedrich.Knollmann@charite.de Copyright 2003 by Lippincott Williams & Wilkins ISSN: 0020-9996/04/3901-0013 DOI: 10.1097/01.rli.0000091848.80376.d5 Key Words: computed tomography (CT), artifact, computed tomography (CT), image quality, coronary vessels, CT, heart, CT, computed tomography (CT), experimental studies (Invest Radiol 2004;39: 13 19) The recent introduction of multirow detector computed tomography (MDCT) 1,2 has provided a potentially useful noninvasive cardiac imaging modality. 3 5 The most important potential application is noninvasive coronary angiography, but motion artifact remains an important limitation. In a direct comparison with conventional coronary angiography, all arteries were evaluable with 4-row multidetector CT in only 30% of the cases. 6 To overcome motion artifacts, temporal resolution of the CT technique can be improved by accelerating gantry speed or using segmented image reconstruction techniques. 7 To improve the diagnostic capabilities of cardiac CT, new image reconstruction algorithms need to be tested and objectively compared. Because the performance of such algorithms can vary with different heart rates, time points within the cardiac cycle, and other image acquisition parameters such as gantry speed, an in vitro test using a dynamic phantom would be desirable for a comprehensive preclinical assessment. Assessment of the full width at half maximum (FWHM) of the point spread function (PSF) using the image of an intensity point source is a well-established approach for the evaluation of imaging methods and forms the basis of acquiring the modulation transfer function (MTF) as a widely accepted way to define spatial resolution. However, the calculation of the MTF using currently applied algorithms depends on rotational symmetry of the imaging system. 8 In the case of a moving intensity point source, this precondition is violated, and the methods to describe spatial resolution are necessarily much more complex than with a stationary object. Without rotational symmetry of the PSF, it becomes necessary to take full account of the 2-dimensional properties of the PSF, which explains why the FWHM area of the 2-di- Investigative Radiology Volume 39, Number 1, January 2004 13

Knollmann et al Investigative Radiology Volume 39, Number 1, January 2004 mensional PSF curve should be used with the presented approach, whereas a 1-dimensional analysis sufficed with stationary intensity point sources. 9 According to transfer theory of imaging systems, the point spread function of an infinitely small object can define the system quality and help predict the system output to any given input. A 50- m tungsten wire comes close to being infinitely small in relation to CT pixel dimensions, which explains its use as a point source for the characterization of CT systems. We set out to validate the use of a dynamic phantom to compare the effective spatial resolution as the FWHM PSF of segmented cardiac CT reconstruction algorithms. MATERIALS AND METHODS A pulsatile cardiac assist device (Novacor N100 LVAS; Edwards Life Sciences, Oakland, CA) was used to simulate human ventricular wall motion at device rates of 0, 60, and 100 bpm. This device consists of 2 aluminum pusher plates that compress the artificial heart chamber and is steered by a portable microcomputer. This device was set into electrocardiogram (EKG) triggering mode using a synthetic EKG signal (Heart Simulator Model 3550; St. Jude Medical, Sylmar, CA). The device motion was mechanically transmitted to a 10-mm long piece of 50- m tungsten wire, which is commonly used to assess the in-plane spatial resolution of CT systems. The tungsten wire was placed parallel to the z-axis in the isocenter of the CT scanner and moved vertically. To synchronize the CT scanner with the dynamic phantom, we used the same simulated EKG signal that also steered the assist device. To determine the motion characteristics of our phantom, the tungsten probe was replaced by a felt tip pen, and the device motion recorded on paper using a modified EKG printer (Schwarzer Cardioscript CD 6000; Picker, Wiesbaden, Germany). To compare different cardiac CT protocols, we tested a subsecond helical CT scanner incorporating an 8-row detector (LightSpeed Ultra; GE Medical Systems, Milwaukee, WI) with our phantom. A slice thickness of 1.25 mm, 0.5 seconds gantry rotation time, and 3-mm table feed per rotation (pitch: 0.3) were selected according to the clinical protocol for coronary artery CT angiography (CTA). With this protocol, images were acquired at 120 kv tube voltage and 270 ma tube current. The scan field of view was set at its minimum of 21 cm, and the display field of view at its minimum of 96 mm. With the 512 512 image matrix, pixel size was 0.035 mm 2. All images were reconstructed with a standard kernel. Image reconstruction was performed using a partial scan algorithm ( snapshot segment algorithm) with a temporal resolution of 330 ms (one half rotation time plus transit 14 time of 1 fan beam width). Then, the same helical raw data set was reconstructed with a segmented 2 sector algorithm ( snapshot burst algorithm) that used image data from 2 consecutive pump cycles with a data acquisition window of 125 ms and a segmented 4-sector algorithm ( snapshot burst plus algorithm), which assembled each image from 4 consecutive device cycles with a data acquisition window of 63 ms. All data sets were repeatedly reconstructed at 0% to 90% of the RR-interval in 10% increments. For device rates of 0 and 60 bpm, the console allowed for using the snapshot segment algorithm only. Thus, comparisons of different reconstruction algorithms were performed for 100 bpm only. For visual analysis, all images were visually rated on a 6-point scale using a display window level of -780 HU and a window width of 650 HU. Images were rated by 1 person after measurement system analysis of his readings had been performed using a panel of 5 radiologists and a medical student. A set of 2 reference images was created to define visual scores of 0 and 5. Score zero was defined by the image of the tungsten wire at rest without any motion artifact. Score 1 indicated the presence of a blurred wire contour and deformation of its shape but with the image having still less than twice the diameter of the wire at rest. Score 2 indicated that the wire transection could still be located, but several streak artifacts were present in its vicinity. With a score of 3, the pin location could not be clearly determined, but emerging streak artifacts were confined to less than 3 times the wire width representation at rest. Score 4 denoted more extensive streak artifacts, and with a score of 5, the wire transsection could not be seen at all. The reproducibility of visual analysis was tested with data from a helical scan at a device rate of 60 bpm. Ten consecutive images from this data set were rated twice by an experienced radiologist, and then by a medical student and 4 radiology residents with variable experience in CT diagnosis. Observer agreement was compared using kappa statistics. 10 For numeric analysis, all image data were digitally transferred to a personal computer and processed by publicly available software (ImageJ, version 1.26t, Wayne Rasband, National Institutes of Health, Bethesda, MD). In a first step, a39 39-mm 2 region of interest (ROI) was centered on the wire transsection in a fashion that the ROI contained all or most of the visible image artifacts. The remaining portions were cropped away and the ROI saved as a text file. The text file was then loaded into an Excel spreadsheet (Microsoft, Redmond, WA) and numerically processed. In a first step, the point spread function of the wire transsection was graphically displayed. Then, the FWHM attenuation of the wire point spread function was determined. 2003 Lippincott Williams & Wilkins

Investigative Radiology Volume 39, Number 1, January 2004 Gauging Effective Spatial Resolution To assess the reproducibility of numeric measurements, the relative variability of repeated measurements was determined as Var Abs((FWHM 1 -FWHM 2 )/((FWHM 1 FWHM 2 )/2)). with FWHM1 being the FWHM of the PSF from the original measurement and FWHM2 the FWHM of the wire PSF from a repeated measurement. The reproducibility is reported as the mean variability (Var) between 2 repeated measurements for the same imaging parameters as with the assessment of visual reproducibility (device rate 60 bpm, images reconstructed at 10 times throughout the cardiac cycle with 10% of the RR-interval interscan spacing). The numeric surrogate of spatial resolution (FWHM of the wire PSF) was compared with visual analysis by Spearman s rank correlation test. The level of statistical significance was set to 5% in all instances. To compare visual and numeric measurements, measurement system analysis 11 was performed (Minitab Statistical Software, release 13 for Windows; Minitab Inc., State College, PA). With this approach, the better measurement system is defined by a lower contribution of the measurement system to overall data variability. It is important to note that the attribute data type of visual assessment alone severely limits all further analysis of visual assessments. RESULTS The wire dynamics remained almost unchanged in systole for all device rates, whereas the diastolic interval decreased with increasing device rates. Reviewing the paper tracings of probe dynamics, 4 different sections of the device cycle could be determined with rapid systolic motion, slow early diastolic filling, almost no motion in the late diastolic phase, and rapid end-diastolic filling (Fig. 1). According to these recordings, the maximum velocity occurred with rapid diastolic filling (1500 mm/s) and ejection (380 mm/s), whereas slow diastolic filling correlated with a pin speed of 51 mm/s. The late diastolic period lasted for 470 ms at a device rate of 60 bpm and for 70 ms at 100 bpm. Visual Analysis At rest, the tungsten wire was depicted as a point in the cross-sectional plane. On visual analysis of the moving phantom, the image of the tungsten wire degenerated into linear patterns as illustrated in Figure 2. At a device rate of 60 bpm, visual assessment of image artifact severity varied between 0 (no motion artifact) at 0% of the RR-interval to 5 (most severe) at 60% of the RRinterval. There was general agreement that the wire was reproduced most accurately at 0% and 90% of the RR- FIGURE 1. (A) Experimental setup of the dynamic phantom with the Novacor circulatory assist device seen from above, and mechanical transmission of device motion to a piece of tungsten wire. (B) Paper recording of pin motion at a device rate of 100 bpm. At this rate, the diastolic interval lasts only 70 ms. Device phases comprise slow diastolic filling, diastolic standstill, rapid end-diastolic filling, and rapid contraction (device systole). The markings on the upper edge of the paper recording indicate a 1-second period. Minor lines are separated by 1 mm. 2003 Lippincott Williams & Wilkins 15

Knollmann et al Investigative Radiology Volume 39, Number 1, January 2004 FIGURE 2. (A) Axial computed tomography (CT) image of the 50- m tungsten wire at rest with a display field of view of 96 mm. (B) Axial CT image of the tungsten wire at 100 bpm with moderate motion artifact at 60% of the RR-interval using the single-segment snapshot reconstruction algorithm. (C) Axial CT image of the tungsten wire at 100 bpm with severe motion artifact at 30% of the RR-interval using the single-segment snapshot reconstruction algorithm. (D) Point spread function of the tungsten wire CT image at rest. Attenuation is given as the difference above air attenuation. (E) Point spread function of the tungsten wire at 100 bpm with maximum motion artifact at 30% of the RR-interval using the single-segment snapshot reconstruction algorithm. Attenuation is given as the difference above air attenuation. Note that the attenuation scale has been adjusted to its maximum. 16 interval, whereas the timing of the most severe artifact was more variable (Fig. 3). However, intraobserver agreement using repeated readings was moderate with a kappa of 0.51, and interobserver agreement was poor with a kappa value of 0.19 on average (range, 0 0.68) in our 6 readers. At a device rate of 100 bpm, the 1-segment snapshot algorithm resulted in visual scores in the same observer between 2 at 60% of the RR-interval and 5 at 10 and 30% of the RR-interval (Fig. 3B). Using the 2-segment snapshot burst algorithm, the score varied between 1 at 10% of the RR-interval and 5 at 30% of the RR-interval (Fig. 3B). Using the 4-segment snapshot burst plus algorithm, the score varied between 0 at 10% of the RR-interval and 5 at 30% of the RR-interval. Comparing the visual ratings of the different reconstruction algorithms, the relative merit of each technique was highly dependent on the time point of image reconstruction within the cardiac cycle. However, the 2-segment snapshot burst algorithm provided a better representation of the wire phantom than the 1-segment snapshot algorithm at most times, and the 4-segment snapshot burst plus algorithm performed better than both the 1-segment snapshot algorithm at most times, and also better than the 2-segment snapshot burst algorithm at some times. Numeric Analysis On numeric analysis, the tungsten wire at rest was reproduced with a 2% MTF of 7.8 line pairs per mm and an FWHM area of the image point spread function (PSF) of 1.30 mm 2. The FWHM area at a device rate of 60 bpm ranged from 1.51 mm 2 at 10% of the RR-interval to 21.62 mm 2 at 60% of the RR-interval (Fig. 4). At a device rate of 100 bpm, the 1-segment snapshot algorithm resulted in FWHM areas between 2.00 mm 2 at 90% 2003 Lippincott Williams & Wilkins

Investigative Radiology Volume 39, Number 1, January 2004 Gauging Effective Spatial Resolution FIGURE 3. (A) Visual assessment of multirow detector computed tomography (MDCT) artifact severity at a device rate of 60 bpm. Image numbers 1 to 10 correspond to reconstruction times of zero to 90% of the RR-interval. Interobserver variability at all times indicates that visual assessment entails significant observer bias. (B) Visual assessment of MDCT artifact severity at a device rate of 100 bpm for single- (SSEG), 2- (SB2), and 4-segement (SB4) reconstruction algorithms. Image numbers 1 to 10 correspond to reconstruction times of zero to 90% of the RR-interval. of the RR-interval and 14.03 mm 2 at 80% of the RR-interval (Fig. 4B). Using the 2-segment snapshot burst algorithm, the FWHM varied between 1.41 mm 2 at 20% of the RR-interval and 66.16 mm 2 at 30% of the RR-interval. Using the 4-segment snapshot burst plus algorithm, FWHM varied between 1.37 mm 2 at 20% of the RR-interval and 63.46 mm 2 at 30% of the RR-interval. Comparing the FWHM values of the different reconstruction algorithms, the relative merit of each technique was highly dependent on the time point of image reconstruction within the cardiac cycle. The 2-segment snapshot burst algorithm provided a better representation of the wire phantom than the 1-segment snapshot algorithm at only 3 of the 10 time points tested, and the 4-segment snapshot burst plus algorithm performed better than both the 1-segment snapshot algorithm at 4 times, and also better than the 2-segment snapshot burst algorithm at 4 times. Both segmented algorithms resulted in much worse image quality than the single-segment snapshot technique at 30% of the RRinterval, however. FIGURE 4. (A) Effective spatial accuracy of multirow detector computed tomography as the full width at half maximum area of the point spread function at a device rate of 60 bpm. Image numbers 1 to 10 correspond to reconstruction times of 0 to 90% of the RR-interval. (B) Effective spatial resolution of helical computed tomography protocols at a device rate of 100 bpm for single- (SSEG), 2- (SB2), and 4-segment (SB4) reconstruction algorithms. Image numbers 1 to 10 correspond to reconstruction times of 0 to 90% of the RR-interval. The reproducibility of numerical effective spatial accuracy as per the FWHM area of the tungsten wire PSF was determined as a mean variability of 0.05 (range, 0.0 0.15). Correlation of Visual With Numeric Estimates of Image Quality On Spearman s rank correlation, the visual estimates closely correlated with the numeric surrogate FWHM (P 0.0001, rho 0.841), confirming that the numeric surrogate for image quality increases with visually more severe artifacts and decreases with better visual estimates of image quality (Fig. 5). On gauging repeatability and reproducibility (Gage R&R), it was found that total Gage R&R of the numeric analysis method contributed only 5.9% of overall study variability, which indicates excellent performance of the numeric measurement system. With this method, 24 distinct categories could have been discerned, whereas more than 5 distinct categories would indicate that the test is acceptable. For the visual analysis method, using Attribute Gage R&R analysis, it was found that only 3.3% of all assessments agreed with each other. 2003 Lippincott Williams & Wilkins 17

Knollmann et al Investigative Radiology Volume 39, Number 1, January 2004 FIGURE 5. Correlation of effective spatial resolution as expressed by full width at half maximum area of the wire point spread function and the visual estimate of image quality. DISCUSSION Assessment of cardiac CT image quality by calculating the effective spatial resolution as the FWHM area of a CT PSF correlated well with visual estimates of image quality, albeit at a higher level of reproducibility. Using this method, distinct advantages of multisegment image reconstruction algorithms for coronary artery CTA could be demonstrated. Because the time of image reconstruction can be adjusted retrospectively with EKG-gated helical acquisition techniques, the reconstruction time with the best possible image quality can be chosen even after the patient has completed his examination. Thus, we suggest citing the best effective spatial resolution achieved with a specific acquisition technique for comparison. For the reconstruction algorithms used in this investigation, the best effective spatial resolution was 1.51 mm 2 at a device rate of 60 bpm, compared with 2.00 mm 2 at 100 bpm with the single-segment snapshot algorithm. The consistency of the lowest artifact severity throughout the cardiac cycle is less significant for retrospectively gated helical acquisition methods, because the time of best image quality can be retrospectively determined form the initial data set. It is also clinically routine for helical cardiac examinations to perform tentative image reconstructions at many different time points throughout the cardiac cycle to determine which time point offers the optimal image quality. Like in clinical practice, the best image quality that can be derived from a scan is dependent on the reconstruction time with least image artifacts. We therefore compared the best image quality that could be possibly achieved with each algorithm, and thus came to the conclusion that the multisegment algorithms offer an image quality benefit. 18 Our experiments imply that increased temporal resolution could cause more severe motion artifacts at selected systolic time points, which can be explained by the inability of a low-resolution technique to capture the entire motion. This effect is illustrated by the deterioration of spatial resolution at 30% of the RR-interval at 100 bpm with the 2- and 4-segment snapshot burst algorithms. The single-segment snapshot algorithm yielded a clearly better spatial resolution under these conditions, although its nominal temporal resolution was only a half and one fourth of the other methods temporal resolution, respectively. Comparisons of the FWHM area with visual estimates of image quality reveal that severe artifacts are consistently noted once the FWHM area exceeds 3 mm 2.WithaFWHM area of less than 2 mm 2, image artifacts could be visible, but the location of the object remains well defined. Using multisegmented image reconstruction algorithms allows us to improve the FWHM area even at a device rate of 100 bpm from 2.0 mm 2 with the single-segment snapshot technique to 1.41 mm 2 with the 2-segment snapshot burst algorithm or to 1.37 mm 2 with the 4-segment snapshot burst plus algorithm. However, the success of multisegmented reconstruction algorithms depends on consistently stable RR-intervals and constant cardiac preload, because small differences of the reconstruction time could cause severe degradation of image quality. 12 With the present setup of our dynamic phantom, device motion was perfectly regular as a result of its microprocessor-controlled rhythm. Yet, the method can potentially reproduce any degree of sinus arrhythmia if the source EKG signal is modified accordingly. After clinical observations, the use of a 2-segment reconstruction algorithm could not prevent the degeneration of image quality with increasing heart rates. 13 This observation indicates that sinus arrhythmia and variations in cardiac preload play an important role when image data are collected from more than a single heartbeat. The validity of our dynamic phantom is confirmed by the multiphasic phantom dynamics, similar to human coronary artery dynamics. The device velocity during slow diastolic filling is in the range of what others have found as the mean velocity of human coronary arteries by electron beam CT, MRI, and cineangiography. 14 Similarly, the systolic velocity in our phantom is within the range of the maximum human coronary artery velocity. The late diastolic filling boost exceeded the normal velocity of human coronary arteries and provides an opportunity to investigate the effect of higher velocities with the same setup. However, this single 10-ms moment of supernatural velocity should not be used to predict the extent of motion artifact in human coronary imaging. Like in the human cardiac cycle, systolic dynamics are mostly unaffected by heart rate, and the shortening of the RR-interval is achieved mostly by decreasing the diastolic 2003 Lippincott Williams & Wilkins

Investigative Radiology Volume 39, Number 1, January 2004 Gauging Effective Spatial Resolution period. Although the device dynamics differ from human coronary artery motion patterns in some details, the range of velocities encountered in this setup should represent the clinical setting sufficiently to allow for its use as an in vitro test bed. However, the exact timing of the highest spatial accuracy relative to the cardiac cycle does not correspond to clinical observations, which is the result of the 150-ms start delay of the assist device after the trigger EKG R-wave. This delay can be freely selected in future investigations. To apply the exact timing sequence to clinical practice, clinical confirmation of the timing would be required. In earlier investigations, the ideal time for reconstruction was highly variable, and the authors recommended to perform image reconstructions at multiple times to determine the best time point on an individual basis. 15 The use of dynamic phantoms for the assessment of cardiac CT techniques has been described by several investigators before. 16,17 However, these studies relied on a visual assessment of image quality, which implies significant observer bias. In our experiments, interobserver agreement on visual assessment was generally poor, and the extent of interobserver disagreement was, in some cases, as large as the differences observed by a single reader between different time points or acquisition methods. Thus, the extent of observer bias has the potential to obscure resolution differences that can be detected by numeric analysis. Also, visual estimates remain semiquantitative by necessity, whereas the numeric surrogate marker presented here can be used as a true interval scale variable. In 1 other investigation, a cardiac assist device has been used as a dynamic phantom for the quantification of coronary calcium. 18 Although a comparison of visual estimates of artifact severity with the presented numeric method might seem challenging at first, understanding of measurement system analysis still allows for a valid comparison. With this approach, the observed variation of both visual and numeric indicators of image quality can be analyzed by determining the reproducibility of each indicator, which addresses that portion of the observed variability that is the result of the measurement system. Reviewing Figures 3 and 4, it is obvious that the portion of image quality variability ascribable to the measurement system is much larger for visual evaluation than for the presented numeric method with its mean variability of 5% on repeated testing. Thus, the numeric method is better suited to detect smaller differences of image quality, which would be obscured by the relatively poor reproducibility of the visual score method. We conclude that the use of moving tungsten wire to determine the FWHM area of the wire PSF could be used as an objective and valid measure of effective spatial accuracy for the evaluation of cardiac CT protocols. This approach extends the concept of using an intensity point source for the comparison of CT spatial resolution to moving objects. The method has the potential to compare different cardiac CT protocols in vitro to foster the development of new CT techniques for noninvasive coronary imaging. REFERENCES 1. Klingenbeck-Regn K, Schaller S, Flohr T, et al. Subsecond multi-slice computed tomography: basics and applications. Eur J Radiol. 1999;31: 110 124. 2. Hu H. Multi-slice helical CT: scan and reconstruction. Med Phys. 1999;26:5 18. 3. Achenbach S, Ulzheimer S, Baum U, et al. Noninvasive coronary angiography by retrospectively ECG-gated multislice CT. Circulation. 2000;102:2823 2828. 4. Ohnesorge B, Flohr T, Becker C, et al. Cardiac imaging by means of electrocardiographically gated multislice spiral CT: initial experiences. Radiology. 2000; 217:564 571. 5. Nieman K, Oudkerk M, Rensing BJ, et al. Coronary angiography with multi-slice computed tomography. Lancet. 2001;357:599 603. 6. Achenbach S, Giesler T, Ropers D, et al. Detection of coronary artery stenoses by contrast-enhanced, retrospectively electrocardiographicallygated, multislice spiral computed tomography. Circulation. 2001;103: 2535 2538. 7. Ohnesorge B, Flohr T, Fischbach R, et al. Reproducibility of coronary calcium quantification in repeat examinations with retrospectively ECGgated multisection spiral CT. Eur Radiol. 2002;12:1532 1540. 8. Rossmann K. Point spread function and modulation transfer function. Radiology. 1969;93:257 272. 9. Assimakopulos PA, Boyd DP, Jaschke W, et al. Spatial resolution of computed tomographic images. Invest Radiol. 1986;21:260 271. 10. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33:159 174. 11. Breyfogle FW III. Implementing Six Sigma Smarter Solutions Using Statistical Methods. New York: John Wiley and Sons; 1999:222 234. 12. Hong C, Becker CR, Huber A, et al. ECG-gated reconstructed multidetector row CT coronary angiography: effect of varying trigger delay on image quality. Radiology. 2001;220:712 717. 13. Vogl TJ, Abolmaali ND, Diebold T, et al. Techniques for the detection of coronary atherosclerosis: multi-detector row CT coronary angiography. Radiology. 2002;223:212 220. 14. Achenbach S, Ropers D, Holle J, et al. In-plane coronary arterial motion velocity: measurement with electron-beam CT. Radiology. 2000;216: 457 463. 15. Kopp AF, Schroeder S, Kuettner A, et al. Coronary arteries: retrospectively ECG-gated multi-detector row CT angiography with selective optimization of the image reconstruction window. Radiology. 2001;221: 683 688. 16. Knollmann F, Halfmann R, Regn J, et al. Motion artifacts in cardiac CT. Acta Radiol. 1999;40:569 577. 17. Wicky S, Rosol M, Hamberg LM, et al. Evaluation of retrospective multisector and half scan ECG-gated multi-detector cardiac CT protocols with moving phantom.j Comput Assist Tomogr. 2003;27:392 398. 18. Hopper KD, Strollo DC, Mauger DT. Comparison of electron-beam and ungated helical CT in detecting coronary arterial calcification by using a working heart phantom and artificial coronary arteries. Radiology. 2002; 222:474 482. 2003 Lippincott Williams & Wilkins 19