Assessment of calcium scoring performance in cardiac computed tomography

Transcription

1 Eur Radiol (2003) 13: DOI /s y CARDIAC Stefan Ulzheimer Willi A. Kalender Assessment of calcium scoring performance in cardiac computed tomography Received: 19 June 2002 Revised: 30 September 2002 Accepted: 10 October 2002 Published online: 4 December 2002 Springer-Verlag 2002 S. Ulzheimer ( ) W. A. Kalender Institute of Medical Physics, University of Erlangen-Nürnberg, Krankenhausstrasse 12, Erlangen, Germany stefan.ulzheimer@imp.uni-erlangen.de Tel.: Fax: Abstract Electron beam tomography (EBT) has been used for cardiac diagnosis and the quantitative assessment of coronary calcium since the late 1980s. The introduction of mechanical multi-slice spiral CT (MSCT) scanners with shorter rotation times opened new possibilities of cardiac imaging with conventional CT scanners. The purpose of this work was to qualitatively and quantitatively evaluate the performance for EBT and MSCT for the task of coronary artery calcium imaging as a function of acquisition protocol, heart rate, spiral reconstruction algorithm (where applicable) and calcium scoring method. A cardiac CT semi-anthropomorphic phantom was designed and manufactured for the investigation of all relevant image quality parameters in cardiac CT. This phantom includes various test objects, some of which can be moved within the anthropomorphic phantom in a manner that mimics realistic heart motion. These tools were used to qualitatively and quantitatively demonstrate the accuracy of coronary calcium imaging using typical protocols for an electron beam (Evolution C-150XP, Imatron, South San Francisco, Calif.) and a 0.5-s four-slice spiral CT scanner (Sensation 4, Siemens, Erlangen, Germany). A special focus was put on the method of quantifying coronary calcium, and three scoring systems were evaluated (Agatston, volume, and mass scoring). Good reproducibility in coronary calcium scoring is always the result of a combination of high temporal and spatial resolution; consequently, thin-slice protocols in combination with retrospective gating on MSCT scanners yielded the best results. The Agatston score was found to be the least reproducible scoring method. The hydroxyapatite mass, being better reproducible and comparable on different scanners and being a physical quantitative measure, appears to be the method of choice for future clinical studies. The hydroxyapatite mass is highly correlated to the Agatston score. The introduced phantoms can be used to quantitatively assess the performance characteristics of, for example, different scanners, reconstruction algorithms, and quantification methods in cardiac CT. This is especially important for quantitative tasks, such as the determination of the amount of calcium in the coronary arteries, to achieve high and constant quality in this field. Keywords Computed tomography Cardiac imaging Quality assurance Coronary calcium Heart Cardiac phantom

2 485 Introduction According to a recent World Health Report by the World Health Organization (WHO) the leading cause of mortality in the world, amounting to 13.7%, is coronary artery disease (CAD). Due to higher risk factors and a lower number of infectious diseases, the percentage in industrialized countries is dramatically higher, e.g., 25.5% in Europe [1]. Conventional X-ray coronary angiography and intravascular ultrasound (IVUS) are considered to be the gold standards for the assessment of artherosclerotic changes of the coronary arteries, the cause of CAD. As these quite expensive procedures are highly invasive having a small but not negligible risk of mortality (0.15%) and morbidity (1.5%) [2], alternative methods for the assessment of coronary artery wall irregularities are desirable. Moreover, since possibly as many as half of the coronary events occur in previously asymptomatic persons [3] an early and easy diagnosis of CAD possibly feasible as a screening procedure is urgently needed. Early changes in the walls of the coronary arteries that cannot be assessed with coronary angiographies determine the risk for coronary events [4, 5]. Since calcium in the form of hydroxyapatite (HA) {Ca [Ca 3 (PO 4 ) 2 ] 3 } 2+ 2OH in the coronary arteries is a known marker for the presence of artherosclerotic lesions of the coronary arteries, the screening for it has been suggested as an early indication for CAD. It has been shown that the risk for coronary events is associated with the amount of coronary calcium [6, 7]. The absence of coronary calcium does almost certainly imply the absence of CAD [8]. Calcium strongly attenuates X-rays due to its relatively high atomic number; therefore, X-ray techniques are suitable methods for detecting coronary calcifications. Plain chest radiographs and fluoroscopy were used to detect coronary calcifications, but they are both not very sensitive and it is not possible to quantify the amount of calcium [9]. The first group that used electron beam tomography (EBT) images for the quantification of coronary calcium were Agatston et al. in 1990 [10]. They introduced a scoring method later called Agatston score which has been used for more than a decade. The reproducibility in the quantification of coronary calcium with EBT and the Agatston score is not known exactly, although individual papers addressed this issue in patient studies [11, 12, 13, 14, 15, 16, 17]. Yoon et al. determined a reproducibility of around 30% [18]. This means that according to the 3 σ criterion, a measured change has to surpass 90% in order to be accepted as a statistically significant change. Consequently, Wang et al. even judged the method as being... not sufficiently reproducible to allow serial quantification of coronary calcium in individual patients over relatively short periods (<2 years) [17]. Modifications for the Agatston score were suggested to improve its reproducibility [19] or sensitivity [20]. Calibration phantoms for EBT were introduced to reduce variability in coronary calcium scoring [21] and some groups started to use mass scoring in parallel to the Agatston score but all efforts were not consistent and have not led to better, widely accepted standards. Despite their comparatively long scan times, even conventional CT scanners with rotation times of up to 1.5 s were used to quantify coronary calcium without making special arrangements to reduce motion artifacts [20, 22, 23, 24, 25, 26]. Compared with the known high interscan variability of calcium scores for EBT and the Agatston score, these conventional approaches yielded similar results. For example, the intermodality variability in a recent study which compared conventional 0.75-s sequential scans without ECG triggering to EBT was 42% [22]. The present technology offers new potential. The new widely available multi-slice spiral computed tomography (MSCT) scanners of the latest generation offer the possibility of cardiac imaging with similar or even better image quality than EBT. They offer rotation times of down to 420 ms and can acquire up to 16 slices with nearly isotropic resolution by using slice collimations of down to 0.5 mm. Since the rotation times of the present scanners can still be long compared with the duration of one complete cardiac cycle, special techniques for cardiac imaging with sequential (prospective triggering) and spiral CT (retrospective gating) [27] were developed at the Institute of Medical Physics (IMP) at the University of Erlangen Nürnberg. They use the recorded ECG signal for image acquisition and reconstruction and aim at reducing effective scan times. These techniques were already introduced in 1997 for single-slice scanners [28, 29, 30, 31] and showed impressive results; however, only their generalizations to multi-slice scanners which allow the depiction of the whole heart with high spatial and temporal resolution from one MSCT scan in a single breathhold led to a breakthrough for cardiac CT [32, 33, 34, 35, 36, 37]. Presently, all CT manufacturers implemented variations of these algorithms in their scanners. Especially for follow-up studies high reproducibility is mandatory to be able to detect changes in the observed parameters, i.e., the progress of the disease or the response to a certain medication. The accurate assessment of reproducibility and measurement errors which has been neglected up to now for coronary calcium scoring is very important for the determination of significant changes in the state of the disease. This article closely describes tools and concepts for quality assurance in cardiac CT especially in the field of quantifying the amount of calcium in the coronary arteries. These tools were first introduced in scientific exhibitions at the European Congress of Radiology in 1999 and 2000 [38, 39]. Their performance is demonstrated here in a systematic fashion for two exemplary systems.

3 486 Fig. 1 a Sketch and b photo of the anthropomorphic phantom body. It can hold different inserts at the position of the heart. c The sketch of a solid insert that allows the static assessment of calcium scoring performance and the determination of calibration factors Further details on the assessment of image-quality parameters for two scanners and several protocols in cardiac CT are presented in a recent PhD thesis [40]. Materials and methods Tools for quality assurance Anthropomorphic phantoms consisting of tissue-, water-, and bone-equivalent materials have been developed that allow the assessment of all important image quality parameters in cardiac CT in general and especially quality assurance in coronary calcium scoring. We thereby continued the concepts successfully established for bone mineral density (BMD) and lung density measurements. The semi-anthropomorphic European Spine Phantom [41] constitutes an internationally accepted calibration standard for X- ray based osteoporosis assessment. The Pulmo Phantom is established in the Siemens Pulmo CT option; its semi-anthropomorphic geometry was derived as an average of a reference population consisting of young male adults used for determination of normal lung density value in All phantom materials are based on epoxy resin. Several additives, such as calcium carbonate CaCO 3, magnesium oxide MgO, hydroxyapatite and microspheres were added to obtain solid water, soft tissue, and lung and bone equivalent materials for X-ray photons in an energy range from 30 to 150 kev [41, 42]. The manufacturer of the phantoms and phantom materials used here (QRM Quality Assurance in Radiology and Medicine GmbH, Möhrendorf, Germany) takes special care in the manufacturing and testing to ensure that materials and phantoms can be used as calibration standards for HA. Since coronary calcifications consist of HA [8], we can use this material directly for the simulation of calcifications. To be able to achieve maximum flexibility in the assessment of different image parameters a modular concept was chosen. Anthropomorphic environment To be able to offer a reproducible and realistic environment for the assessment of image quality, an anthropomorphic phantom body was developed consisting of a cross section of the human thorax with artificial lungs and a spine insert surrounded by tissue-equivalent material (Fig. 1). As a model we used the Pulmo Phantom. At the position of the heart we placed an empty hole that can hold different kinds of inserts for various purposes. Inserts for the assessment of image-quality parameters With the described phantom body, principally all objects that are suitable for the investigation of image-quality parameters in cardiac CT can be scanned in an anthropomorphic and reproducible environment. This can be achieved by putting the respective objects (e.g., wires, discs, hole or bar patterns, water-equivalent materials as well as low- and high-contrast structures), which are well known from the determination of image-quality parameters in non-anthropomorphic environments [27], in a water-filled tank that fits into the hole of the cardiac phantom body. In this paper we concentrate on the assessment of calcium scoring performance in cardiac CT. For this purpose, small, exactly defined cylinders of different hydroxyapatite densities were manufactured and immersed in water-equivalent rods to simulate varying sizes and densities of coronary calcifications. Cylinders with diam-

4 487 Fig. 2a c The motion phantom. A robot setup was developed for moving arbitrary 3D objects inside a water tank on realistic trajectories to simulate cardiac motion. The setup also produces a corresponding ECG signal. c A pseudo-4d plot of a realistic motion function obtained from angiography data in correspondence to the ECG signal. An animated sequence of this plot can be accessed on eters of 1, 2, 3, and 5 mm were produced, whereby the length of the cylinders always equals its diameter. Each calcification exists in three different HA densities of 200, 400, and 800 mg/cm 3. In addition, to these 12 artificial calcified inserts, we obtained a calcified coronary artery specimen which was also immersed in waterequivalent plastic. All these inserts were manufactured in a way so that they can be moved dynamically in the phantom. We also composed solid, stationary inserts for the quick and easy assessment of image-quality parameters. One of these inserts is shown in Fig. 1c). This solid cylinder that fits in the hole of the phantom body contains nine small calcifications of varying size and HA density. Moreover, we included two larger calibration inserts (solid water and 200 mg HA/cm 3 ) for the determination of calibration factors. Setup for the simulation of cardiac motion Since the influence of cardiac motion is of decisive interest in cardiac CT, a primary focus of the phantom developments was the creation of assemblies that allow the examination of the effect of realistic cardiac motion on image quality; therefore, a setup was constructed to allow the movement of different objects for quality assurance in a water tank inside the anthropomorphic environment. Realistic motion functions were obtained from cineangiography. For all investigations, we used the typical motion of a left anterior descending (LAD) coronary artery obtained from a patient with a heart rate of 58 bpm. The maximum contraction and therefore a maximum of the motion function is at 29% (systole), and the minimum at 96% of R-R interval (end diastole). The amplitude from the maximum to the minimum of this motion curve is approximately 7.5 mm. During diastole from 55 to 85% of R-R is a phase of relatively little movement (Fig. 2c).

5 488 It is clear that this motion function can only be an example for the movement of structures in the heart, as there are big differences for varying positions on the heart walls, different heart rates, and above all between different individuals. Especially for the right coronary artery, the amplitudes and velocities are much higher. This specific function was chosen as it includes all details needed for comparing different imaging methods in a sufficient manner, above all realistic phases of fast and slow motion. This motion function was scaled for different heart rates by simply decreasing the time for the R-R cycle. This is not in accordance with the physiologic behavior as the motion function significantly changes with increasing heart rate. Nevertheless, this method of scaling to other heart rates appears sufficient for our purpose: the time of slow motion decreases and the velocities increase which corresponds to the physiologic tendency. To really find the motion function for higher heart rates for the specific individual was not the goal in this proof of concept, as the variation between individuals are very high anyway. Nevertheless, for future studies any desired motion function can be used in this setup. A commercially available set of linear translation tables (iselautomation, Eiterfeld, Germany) were used which were mounted perpendicular to each other in order to reach all points in 3D space. A system of metal rods that reaches into a water tank was attached to one of the translation tables. Water-equivalent rods containing the respective test objects were attached to the end of the metal rod. Each linear table can be accessed independently with a control unit and a software library. This library was used in the implementation of a C++ program that allows the comfortable input and execution of arbitrary 3D motion functions with a precision of 1/100 mm. The only limitations in the present implementation are the achievable velocity of approximately 150 mm/s and the cruising radius of approximately 200 mm. The control unit for the linear tables can be programmed to generate 12-V signals that correspond to the desired motion. These signals are converted to a corresponding ECG signal of approximately 2 mv with a selfbuilt voltage divider. This signal can be fed directly into the ECG monitor used for triggering and gating techniques. Figure 2 shows a drawing of the setup and a photo of the complete setup inside the gantry of a CT scanner. Image acquisition CT scanners For all measurements concerning EBT, the EBT Scanner Evolution C150XP (Imatron, South San Francisco, Calif.) with software version 12.4 was used. As an example of a modern MSCT scanner, we used the Sensation 4 (Siemens, Erlangen, Germany). It provides the simultaneous acquisition of up to four slices and rotation times of down to 0.5 s. For cardiac applications the slice collimations of 4 1 and mm are relevant. Scan protocols Scan protocols have in common that they allow the coverage of the complete heart (typically approximately 120 mm) during one breathhold (typically <20 40 s) with the respective modality. For coronary calcium screening with EBT, usually a singleslice mode with 3-mm collimation, a scan time of 100 ms per slice, a table increment d of 3 mm per slice, and prospective triggering are used because this allows the coverage of the whole heart (N=40 slices) during one breathhold and keeps noise within a reasonable range. The tube current and voltage are fixed to I=630 ma and U=130 kv on this EBT scanner. This protocol is commonly accepted [43]. For calcium scoring with MSCT scanners, two principally different image acquisition methods are possible: prospective triggering in sequential mode and retrospective gating in spiral mode. We used the protocol suggested by the manufacturer for coronary calcium scoring with prospective triggering (S=4 2.5 mm, d=10 mm, N=10, I=100 ma, U=120 kv and rotation time t rot =500 ms) and a consensus protocol proposed by German Sensation 4 users for coronary calcium scoring with spiral CT (S=4 2.5 mm, pitch p=0.375, number of rotations N=32, I=100 ma, U=120 kv and t rot =500 ms). Moreover, some MSCT scans with thinner slice collimations (4 1 mm) were carried out to investigate the influence of thinner slice collimations on the quantification of coronary calcium. Image reconstruction For EBT image reconstructions were carried out only with the scanner s software with the prescribed sharp reconstruction kernel and a field of view of 250 mm. For MSCT two options were used to reconstruct images from the scanner s attenuation raw data: Firstly, we used the Siemens software Somaris/5 (version VA20Q) with a special option for cardiac imaging which allows prospective ECG triggering and retrospective ECG gating. The exact implementation of the retrospective gating algorithm on the scanner has not been published. Nevertheless, publications by Siemens engineers give some hints [44, 45, 46]: above a certain heart rate, the algorithm switches from a partial scan approach with an effective scan time of 250 ms to an interpolation approach called multiphasic or biphasic reconstruction with a fixed effective scan time of 125 ms. Secondly, to be able to reconstruct images with our own reconstruction software and the well-described and evaluated cardiac reconstruction algorithms 180 multi-slice cardiac delta (MCD) and 180 multislice cardiac interpolation (MCI) [36, 37] the raw data and the ECG information was exported to a special image reconstruction station (ImpactIR, VAMP, Möhrendorf, Germany). The software runs on any standard PC. Currently, a reconstruction time of below 1.5 s per image is achieved with a 1-GHz dual Pentium III processor and 1 GB of memory. For comparison purposes we also reconstructed images with standard reconstruction algorithms 180 multi-slice filtered interpolation (MFI) and 180 multi-slice linear interpolation (MLI) [27] that do not utilize ECG information for image reconstruction. In all cases MSCT images were reconstructed with a field of view of 220 mm and the recommended Siemens kernel B35f. All reconstructed images were exported in Digital Imaging and Communications in Medicine (DICOM) format to CD or network drives and evaluated with self-developed software tools. Measurements and data evaluation All 12 artificial calcifications and the specimen were scanned three times with the three different protocols at rest and for at least two different heart rates to allow the assessment of measurement errors. Additionally, to investigate the influence of the heart rate more closely one of the artificial calcifications (3 mm, 400 mg HA/cm 3 ) was scanned three times for seven more heart rates with all protocols. For prospective triggering techniques data sets for at least five different positions in the R-R interval were acquired to be able to determine the influence of heart phase on image quality. These measurements resulted in more than 450 raw data sets. For the retrospective gating technique images were reconstructed for ten different positions in the R-R cycle and evaluated with the three quantification methods defined below. The images of each method which contained the least motion artifacts were taken for further evaluation. Shaded-surface dis-

6 489 plays (SSDs) of the calcifications were rendered with a threshold of 130 HU to allow an easy 3D inspection. To be able to estimate the detection limits for the different systems and protocols the mean CT value above the threshold of the calcification is plotted against the size (see Fig. 4). The detection limit, i.e., the size where the maximum CT value falls below the threshold of 130 HU, can be determined by a linear extrapolation of the curves to zero. This yields rough estimates for which sizes of calcifications are just visible in CT images. For all quantitative measurements the mean absolute error or the relative error of the determined mean value of a quantity x is shown either as an error bar in diagrams or in the form of a table. Most test series conducted contain only a small number of measured values and so itself is subject to a significant error. For such small measurement series has tobe calculated from the standard deviation s, the number of measured values N, and a so-called t -factor that corrects for the small number of measured values [47]. t depends on the number of measured values N and the probability p that a measured value lies within the interval. We used p=95% (significance) and mostly N=3 which yields t=4.3 and. We determined the measurement errors for all artificial and real calcifications for different scan protocols and scoring methods. The implementation of the different quantification methods is based on the definitions given below. Quantification methods Agatston score. The Agatston score in its original form is determined from 20 contiguous EBT slices of the heart (3-mm slice thickness, no interslice gaps, 100-ms data acquisition time per slice during breath holding). It was implemented as described in the original publication [10]. In each of the 20 images, calcifications are determined by setting a threshold of 130 HU and ignoring structures with sizes below 1 mm 2 to exclude noise from evaluation. Then, one region of interest (ROI) is placed around lesion j of coronary artery i in each of the 20 images and the area A ij in mm 2 and the maximum CT numbers in HU of the ROI are determined. The score S ij of lesion j in coronary artery i is calculated by multiplying the area of the lesion with a weighting factor w ij that depends on the maximum CT number in the lesions for each image: (1) with (2) The total calcium score (S tot or TCS) is determined by summing up the scores of the lesions for all arteries in all images: The Agatston score was designed for a special modality and protocol; therefore, the score is not invariant with respect to image parameters such as slice thickness, slice spacing, absolute CT numbers, and reconstruction kernels. For example the images used for calcium scoring with new scanners and protocols will not necessarily have a slice thickness of S 0 =3 mm and a spacing of RI 0 =3 mm such as in the protocol (3) used by Agatston. If arbitrary spacing of the images with, for instance, reconstruction increments RI or table feeds smaller than the slice thickness, i.e., RI<S, are allowed and the score in each image is determined as before, RI 0 /RI more slices per volume are evaluated. Then the determined Agatston score had to be multiplied with the factor RI/RI 0 to keep it invariant with respect to these parameters and to be able to compare the obtained scores with the ones obtained with EBT. For 2.5-mm slices the determined score was multiplied with to obtain a comparable Agatston score. Moreover, the fact that different systems and protocols yield different absolute CT numbers for the same structure does not only influence the segmentation threshold but also the threshold of the weighting function for the Agatston score. The reconstruction kernels also severely affect CT numbers of small calcifications and consequently the Agatston score, an effect that cannot be easily corrected for. Here we corrected the Agatston scores only for the different slice thicknesses. All other parameters were neglected. Volume scores. Volume scores are simply calculated as the number of voxels N voxel in the volume data set that belong to the calcification multiplied by the volume of one voxel V voxel (4) One has to be aware that V does not necessarily represent the real volume of the calcification as it strongly depends on the threshold used. This is obviously a drawback of this scoring method as it also does not correspond to a real physical measure. The observed size of a segmented homogeneous structure strongly depends on the position of the threshold compared with the maximum CT value of the structure [40]. Generally, if the threshold is set to CT# max /2, V approaches the real volume of the calcification. At a fixed threshold of 130 HU, V overestimates the volume of very dense calcifications and underestimates the volume of less dense calcifications. Only the volume of calcifications with mean CT numbers of approximately 260 HU is close to the real volume. This is the reason why these scores are called volume scores and not simply the volume of the calcification. Additionally, when using slice thicknesses that are relatively large compared with the size of the calcification, this quantification method can lead to large deviations from the true diameters of the calcification due to the fact that objects much smaller than one voxel but with high density nevertheless contribute to the score with the complete voxel volume (linear partial-volume effect). Hydroxyapatite mass. The best and easiest quantification method might be to use the calcium mass as a measure for the amount of calcium as it corresponds to a real physical measure, automatically corrects for linear partial-volume effects, and is invariant with respect to most scan parameters when using appropriate calibration. The density of a homogeneous calcification is defined as ρ HA =m/v. This means that the mass is given as m =ρ HA V or for the general case of calcifications with heterogeneous density If we assume that the HA density is directly proportional to the CT numbers (6) with c HA as calibration constant, and if we consider the case of discrete voxels, we get If we now take into account that we get (5) (7) (8)

7 490 i.e., the mean CT number of each calcification multiplied by the volume score V of the calcification is directly proportional to the calcium massm. To obtain absolute values for the calcium mass, a calibration measurement of a sufficiently large calcification with known HA density ρ HA has to be carried out and the calibration factor c HA has to be determined. The calibration factor c HA is calculated as (9) Equation 9 assumes that the CT number of water is 0 according to definitions. If an exact CT value for water is available from the same calibration measurement, and if it is not equal to 0, this can be taken into account by subtracting the CT value for water from the CT number of the calcification (baseline correction): (10) In practical cases the CT number of water always has to be checked and taken into account. As the CT number of all materials, except water, depends on the X-ray spectrum used, the calibration factor has to be determined for all scanners and protocols. Of course, the determined HA mass can only be the mass above the threshold used for segmentation. The lower the threshold can be chosen, the more exactly the HA mass can be determined. Nevertheless, it must be ensured that the calcification is safely separated from the background with the threshold chosen; therefore, if no other errors are involved, it can be expected that the real mass is always underestimated by a certain amount. Apart from effects that are due to the fact that calcifications may contain non-calcium components, the calcium mass automatically corrects for linear partial-volume effects, however, as objects smaller than the slice thickness are displayed with accordingly decreased mean CT numbers. Results Determination of calibration factors for the HA mass Table 1 shows the result of a calibration for the C 150XP and the Sensation 4 for all tube voltages based on the 0- and 200-mg HA/cm 3 regions of the calibration insert (Fig. 1c). Apparently, a voltage setting of 120 kv with the Sensation 4 yields approximately similar absolute CT values as the EBT scanner. This is important when comparing the Agatston score which uses absolute CT values for different scanners. Using tube voltages other than 120 kv would yield significantly different Agatston scores without additional corrections. Such corrections would affect both the weighting function and the threshold for segmentation. Performance assessment in calcium scoring Table 1 Calibration factors for determination of the hydroxyapatite (HA) mass for all tube voltages of the Siemens Sensation 4 and the Imatron Evolution C 150XP Scanner Evolution C 150XP 130 kv Sensation 4 80 kv kv kv Figure 3 shows a qualitative comparison of the different methods at different heart rates for an artificial calcification and the calcified specimen. The standard reconstruction algorithms 180 MFI and 180 MLI without ECG correlation are only able to depict the calcifications with strong motion artifacts. A 180 MFI tends to smear the calcification due to its very low temporal resolution, and 180 MLI tends to provide sharper displays but ruptures the calcification apart due to its better temporal resolution. For the low heart rate all other methods yield good results, although the higher spatial resolution of the 4 1-mm collimation is clearly visible. For the high heart rate, especially methods with low temporal resolution run into more or less severe problems. For the small and less dense HA cylinder the interpolation algorithm 180 MCI performs still very well, whereas the partial scan algorithm 180 MCD and the prospective triggering technique with MSCT clearly show artifacts. For the large and quite dense calcification (Fig. 3b) we get a different picture. Here, also the interpolation algorithm 180 MCI increasingly runs into problems with the depiction of the calcification. Although it performs still better than 180 MCD, small calcium pieces turn up in the vicinity of the calcification. The reason for this effect is star-shaped artifacts in the axial images for high heart rates that are above the threshold for dense and large calcifications but below the threshold for smaller and less dense calcifications. Detection limits When the HA mass was introduced in the previous section we assumed that the HA density of a calcification is directly proportional to its CT number (ρ HA CT#). If it were that easy, the determination of detection limits would not be a big problem. It would be enough to determine the HA concentration which leads to values above the threshold. For the protocols and scanners examined here this would be a HA density of approximately 100 mg/cm 3 (Table 1) which corresponds to a threshold of 130 HU. This assumption is only true for large objects where partial-volume effects do not occur. For small objects the maximum CT value and with it the mean CT value of the calcification are decreased. For calcifications their maximum CT value in the image is important

8 491 Fig. 3a, b The artificial calcification (diameter: 3 mm; length: 3 mm; hydroxypapatite (HA) density: 400 mg HA/cm 3 ) and the calcified specimen (diameter 5 mm, length 15 mm) scanned at rest (0 bpm) and at different heart rates depicted as shaded-surface display. The top row indicates the different reconstruction or scan methods and the slice collimations used for scanning. Seq stands for sequential conventional CT with prospective triggering. Always the data sets that contained least motion artifacts are shown. MFI multi-slice filtered interpolation, MLI multislice linear interpolation, MCD multi-slice cardiac delta, MCI multi-slice cardiac interpolation, EBT electron beam tomography as this parameter determines if the calcification is detected above a given threshold, in our case 130 HU; thus, the detectability of calcifications also depends on their size and not only on their density. Figure 4 exemplarily illustrates this effect for resting calcifications consisting of different HA density for the three different protocols. Table 2 shows an overview of the detection limits determined for the three protocols, different HA densities of calcifications, and heart rates. The different protocols and scanners yield different results mainly due to their different effective slice thicknesses and in-plane resolution. Thinner slices yield a lower detection limit of calcifications over the threshold of 130 HU due to decreased linear partial-volume effects; therefore, the detection limit for calcifications at rest is best with MSCT and prospective triggering (effective slice width S eff =2.5 mm) and worst with retrospectively gated MSCT with four times 2.5-mm slice collimation due to its relatively large effective slice width of S eff =3.3 mm. For resting objects EBT (S eff =3.0 mm) shows a slightly worse detection limit compared with prospective triggering due to its slightly larger effective slice width. The large effective slice width for retrospective gating is due to the fact that we also used cardiac reconstruction algorithms with a simulated ECG signal for resting calcifications. The detection limits for moving objects are affected by motion according to the temporal resolution of the scanning method.

9 492 Measurement errors and reproducibility Fig. 4 Determination of detection limits for calcifications in coronary calcium scoring at rest. The mean CT number of the calcification above the threshold is plotted against the calcification s size. An extrapolation to zero yields an estimate for the size where the maximum CT number of the calcification falls below the threshold. Calcifications with smaller sizes cannot be detected using the given threshold. MSCT multi-slice spiral computed tomography Table 2 Detection limits at a threshold of 130 HU in coronary calcium scoring. EBT electron beam tomography, MSCT multislice spiral CT, HA hydroxyapatite mg HA/cm 3 0 bpm 60 bpm 100 bpm (mm) (mm) (mm) EBT MSCT, prospective triggering MSCT, retrospective gating The measurement errors for all three protocols and all calcifications have been assessed for all resting and moving calcifications. Additionally, the influence of the heart frequency was examined more closely for a typical medium size calcification with average density (length: 3 mm; density: 400 mg HA/cm 3 ). Figure 5a shows the Agatston score s dependence on the heart rate for different protocols. We also examined the influence of different slice thicknesses and reconstruction methods; therefore, in addition to the standard calcium scoring protocols, retrospectively gated MSCT with a thin-slice collimation of 4 1 mm was used and the acquired data for the mm collimation were reconstructed with the manufacturer s image reconstruction algorithm and the IMP algorithm 180 MCI. Table 3 shows the evaluation of quantitative measures for the different methods from Fig. 5. The mean values, the standard deviation σ, and the coefficient of variation CV=σ/mean of the different methods for varying heart rate, were calculated. Firstly we can note that the 4 1-mm collimation in combination with retrospective gating on the MSCT scanner definitely yields the best results with respect to reproducibility (CV=5.9%). As expected, the methods are influenced by motion to a degree according to their temporal resolution. Retrospective gating techniques show improved reproducibility because overlapping slices are reconstructed. Nevertheless, the scores decrease with increasing heart rate due to increased blurring. The IMP implementation of the reconstruction algorithm 180 MCI appears more robust as the score decreases to a lesser extent than for the manufacturer s implementation. The same applies when calculating the calcium mass (Fig. 5b), but generally, compared with the Agatston score, the measurement errors are clearly decreased, which can be seen when looking at the error bars in the diagrams. The quality of the scoring method is not ex- Table 3 Variation of Agatston scores and HA mass for the different methods and varying heart rate shown in Fig. 5. The mean values for the different heart rates, the standard deviation σ, and the coefficient of variation CV=σ/Mean are calculated for the different methods. IMP Institute of Medical Physics EBT MSCT, MSCT, MSCT, MSCT, prospective prospective retrospective retrospective retrospective triggering, triggering, gating, gating, gating, 3 mm, mm, mm, mm, mm, Imatron Siemens Siemens IMP IMP Agatston score Mean σ CV (%) HA mass Mean (mg) σ (mg) CV (%)

10 493 By evaluating the measurements for all calcifications from 1 to 5 mm and for different densities, we found that the volume score and the HA mass are clearly advantageous compared with the Agatston score in terms of reproducibility. The calcification s mass above the threshold is another key factor for good reproducibility. For high HA densities and large calcifications measurement errors are low. Even large calcifications but with low densities are highly affected by motion. Accuracy and comparability Fig. 5a, b Reproducibility of the Agatston score and the HA mass for a moving calcification (length: 3 mm; diameter: 3 mm; density: 400 mg HA/cm 3 ) for different scanning methods. Retrospective gating with thin slice collimation yields best, prospective triggering with MSCT worst reproducibility. Generally, b the calcium mass is better reproducible than a the Agatston score pected to be mainly a function of heart rate and therefore the calculated CVs are not dramatically decreased for the calcium mass compared with the Agatston score. Nevertheless, also here the lower CVs indicate that mass scoring is more robust. As for the Agatston score, the thin-slice collimation yields the best results and the mean values for the manufacturer s reconstruction algorithm decrease more strongly than those determined from the images reconstructed with 180 MCI. Especially, the IMP algorithm 180 MCI can clearly deal better with heart rates over 70 compared with the manufacturer s implementation (p< , Wilcoxon test). In this section the accuracy of the methods, i.e., how good the measured values correspond to the true ones, is evaluated. For the Agatston score no physically true values exist; therefore, we here use the Agatston score determined with EBT as reference to which we compare the Agatston scores determined with the other scanner. In Fig. 6 the Agatston scores of all calcifications measured with MSCT are plotted against the ones determined with EBT. The comparison yields a reasonable correlation, although especially low scores seem to be underestimated with MSCT compared with EBT. For prospectively triggered MSCT and scores below 50, deviations of approximately 50% occur. Retrospectively gated MSCT with a slice collimation of mm underestimates the scores more severely due to its larger effective slice width of 3.3 mm; therefore, many of the small and low-density calcifications are missed even though they are depicted above the threshold with EBT. It has been pointed out before that the Agatston score has many disadvantages; nevertheless, the extensive data acquired with it should not be lost when switching to a new score. The Agatston score does not increase linearly with the density but at least the steps of the weighting function are in equal distance; therefore, it is of interest to investigate if it is possible to estimate the calcium mass from the Agatston score, and vice versa, with a reasonable error. This was tried for EBT and all measured values in Fig. 7. Apparently, the calcium mass is closely correlated to the Agatston score (R 2 =0.9733). The calcium mass can be calculated quite accurately from the total calcium score (Agatston) by multiplying with a factor τ: (11) where in this case τ= mg. For prospectively triggered MSCT and retrospectively gated MSCT the correlation of HA mass and Agatston score is quite good also. The respective values are given in Table 4. Nine measured values in Fig. 7 with HA masses above 80 mg have been excluded from the regression. They are too dense and therefore yield too high CT numbers to still yield an approximately linear relationship with the Agatston score. These are the measured values

11 494 Fig. 7 Relation of the HA mass vs the Agatston score. Nine measured values with HA masses above 80 mg were not included in the linear regression. They were too dense and therefore yield too high CT numbers to still yield an approximately linear relationship with the Agatston score (see text) Discussion and conclusion Fig. 6 Comparison of Agatston scores determined with a prospectively triggered MSCT and b retrospectively gated MSCT with EBT. Due to its relatively large effective slice width of 3.3 mm with the given protocol, retrospective gating yields zero scores for small calcifications that can be detected with EBT using the threshold of 130 HU Table 4 Conversion factors from Agatston score to HA mass for different techniques and protocols Protocol τ/mg R 2 EBT Prospective triggering Retrospective gating for the 5-mm, 800-HA calcification at rest and for two different heart rates. The 5-mm, 800-HA calcification is probably not physiologically relevant due to its large size combined with very high density. Using a custom-designed anthropomorphic cardiac phantom, we have qualitatively and quantitatively evaluated the performance of coronary calcium imaging as a function of scanner type, heart rate, reconstruction algorithm, and quantification system. The new MSCT scanners offer the possibility of highquality cardiac imaging with good reproducibility in quantitative studies. Prospective triggering with MSCT scanners is most severely affected by motion, and EBT is least severely affected by motion. Although retrospective gating is affected by motion to a lesser extent, due to shorter effective scan times and the possibility of reconstructing overlapping images, its relatively high effective slice thickness of 3.3 mm with the mm slice collimation yields the worst detection limits. This also leads to the worst reproducibility in many cases as the amount of calcium above the threshold is a key factor for good reproducibility. Good performance in the assessment of quantitative parameters is always the result of high temporal and high spatial resolution. Thin-slice protocols in combination with retrospective gating have the potential to improve the performance of MSCT scanners significantly. These protocols are already used in CT angiography applications for coronary imaging, but they are avoided for screening procedures such as calcium scoring due to the higher patient dose of 4 8 msv applied [40, 48]. Patient dose values of the present standard protocols for the quantification of coronary calcium are in the range of approximately 0.5 to 2 msv [40]. Probably methods for

12 495 dose reduction in spiral CT [27, 40, 46] will help to decrease patient dose values to an extent that they can be used for the quantification of coronary calcium. Not only the scanners have a big influence on the reproducibility of scoring results, but also the quantification methods themselves. The Agatston score with its many limitations has been used since The most obvious drawbacks are its complexity, its dependence on CT number levels and on the number of slices per scan length, its non-linearity, and its strong dependence on small variations due to noise and motion as it uses maximum CT numbers. Furthermore, as the score does not correspond to a physical measure, it cannot be easily compared with true values for exactly defined calcifications and cannot be used in an intuitive manner. The large amount of data acquired with the Agatston score and the risk tables derived therefrom for coronary events [6, 7] are the only reason to keep it. Epidemiologic data has been derived from these measurements and many physicians are used to this score. Nevertheless, it appears mandatory to switch to a new quantification method that can be compared for different scanners and that is robust with respect to modifications of scanners, protocols, and segmentation approaches. Here the HA mass measured in milligrams of hydroxyapatite above a certain threshold is clearly superior. It can be determined easily with a simple calibration measurement that has to be carried out only once for each scanner and protocol used, and it is a physical measure which can be validated. It offers significantly improved reproducibility and even correlates with certain limitations to the Agatston score. This might be important for those who thus far have doubted the clinical relevance of the HA mass compared with the Agatston score. The cutoff points for risk groups determined in epidemiologic studies can be converted in good approximation from Agatston score to HA mass. Thereby these data will not be lost and all investigations can be continued with the HA mass instead of the Agatston score. Also the segmentation methods leave much room for improvement. They can be considered to lower the threshold for modern scanners as they allow to reduce noise significantly by increasing the tube current. The only reason for the traditionally relatively high threshold of 130 HU for EBT is the high noise in EBT images. We suggest to switch from a CT number threshold in HU to an HA density threshold (e.g., 80 mg HA/cm 3 ) that is independent of the scanner in future studies. Moreover, non-threshold-based segmentation approaches have to be considered. Progression studies that aim at already detecting small changes of calcium in a short time should always be carried out with the same scanner, with the same scanning technique, and with the same evaluation method because only this ensures highest reproducibility. There is a clear analogy to other quantitative methods, such as BMD measurements, in osteoporosis [27]. Although it is possible to correct for certain effects with the introduced quality assurance tools, some errors will still remain for which there is no compensation. For scanners that are not as stable as the current MSCT scanners, it can be considered to scan a reference phantom with each scan to be able to correct individually for variations in tube voltage, etc. Generally, future clinical studies will require standardized methods for image evaluation that have to be compiled to assure high quality of the assessment of quantitative parameters with different scanners. Performance characteristics of scanners have to be evaluated and scan protocols and evaluation procedures have to be optimized in order to achieve high reproducibility and comparability of data acquired at different clinical sites. Initiatives in Europe and in the U.S. are already working on this ambitious subject. Only then will it be possible to obtain reliable results in the field of coronary calcium scoring. Modifications to the described phantoms already implemented by the authors offer the possibility of inserting ionization chambers at different positions for the assessment of dose values, and extension rings allow simulation of different patient sizes with the introduced quality-assurance tools. Respective measurements and the assessment of physical parameters, such as slice-sensitivity profiles, modulation transfer functions, contrast and noise, and patient dose, including the new generation of 8- and 16-slice scanners, are presently in progress. Acknowledgements We thank our colleagues from the Department of Cardiology, especially S. Achenbach and D. Ropers, and from the Department of Diagnostic Radiology, for a good and pleasant collaboration. S. Achenbach supplied us with the cardiac 4D motion functions determined from angiography data. Parts of this work were supported by grants from the Bayerische Forschungsstiftung (AZ 262/98 and AZ 322/99). We also thank the reviewers for their constructive comments which helped to improve the manuscript.

13 496 References 1. World Health Organization (1999) The World Health Report 1999: making a difference. World Health Organization, Geneva 2. Kennedy J (1982) Complications associated with cardiac catheterization and angiography. Catheter Cardiovasc Diagn 8: O Rourke R, Brundage B, Froelicher V, Greenland P, Grundy S, Hachamovitch R, Pohost G, Shaw L, Weintraub W, Winters JWL (2000) American College of Cardiology/American Heart Association Expert Consensus Document on electronbeam computed tomography for the diagnosis and prognosis of coronary artery disease. J Am Coll Cardiol 36: Little W, Constantinescu M, Applegate R, Kutcher M, Burrows M, Kahl F, Santamore W (1988) Can coronary angiography predict the site of a subsequent myocardial infarction in patients with mild-to-moderate coronary artery disease? Circulation 78: Ambrose J, Fuster V (1998) The risk of coronary occlusion is not proportional to the prior severity of coronary stenoses. Heart 79:34 6. Arad Y, Spadaro L, Goodman K, Newstein D, Guerci A (2000) Prediction of coronary events with electron beam computed tomography. J Am Coll Cardiol 36: Raggi P, Callister T, Cooil B, He Z, Lippolis N, Russo D, Zelinger A, Mahmarian J (2000) Identification of patients at increased risk of first unheralded acute myocardial infarction by electron-beam computed tomography. Circulation 101: Wexler L, Brundage B, Crouse J, Detrano R, Fuster V, Maddahi J, Rumberger J, Stanford W, White R, Taubert K (1996) Coronary artery calcification: pathophysiology, epidemiology, imaging methods, and clinical implications. A statement for health professionals from the American Heart Association. Writing Group. Circulation 94: Stanford W, Thompson B, Weiss R (1993) Coronary artery calcification: clinical significance and current methods of detection. Am J Roentgenol 161: Agatston A, Janowitz W, Hildner F, Zusmer N, Viamonte M, Detrano R (1990) Quantification of coronary artery calcium using ultrafast computed tomography. J Am Coll Cardiol 15: Bielak L, Sheedy n PF, Peyser P (2001) Coronary artery calcification measured at electron-beam CT: agreement in dual scan runs and change over time. Radiology 218: Achenbach S, Ropers D, Mohlenkamp S, Schmermund A, Muschiol G, Groth J, Kusus M, Regenfus M, Daniel W, Erbel R, Moshage W (2001) Variability of repeated coronary artery calcium measurements by electron beam tomography. Am J Cardiol 87: Becker C, Knez A, Ohnesorge B, Schöpf U, Flohr T, Bruening R, Haberl R, Reiser M (2000) Visualization and quantification of coronary calcifications with electron beam and spiral computed tomography. Eur Radiol 10: Devries S, Wolfkiel C, Shah V, Chomka E, Rich S (1995) Reproducibility of the measurement of coronary calcium with ultrafast computed tomography. Am J Cardiol 75: Hernigou A, Challande P, Boudeville J, Séné V, Grataloup C, Plainfossé M (1996) Reproducibility of coronary calcification detection with electronbeam computed tomography. Eur Radiol 6: Shields J, Mielke C, Rockwood T, Short R, Viren F (1995) Reliability of electron beam computed tomography to detect coronary artery calcification. Am J Card Imaging 9: Wang S, Detrano R, Secci A, Tang W, Doherty T, Puentes G, Wong N, Brundage B (1996) Detection of coronary calcification with electron-beam computed tomography: evaluation of interexamination reproducibility and comparison of three image-acquisition protocols. Am Heart J 132: Yoon H, Greaser L, Mather R, Sinha S, McNitt-Gray M, Goldin J (1997) Coronary artery calcium: alternate methods for accurate and reproducible quantitation. Acad Radiol 4: Shemesh J, Tenenbaum A, Kopecky K, Apter S, Rozenman J, Itzchak Y, Motro M (1997) Coronary calcium measurements by double helical computed tomography. Using the average instead of peak density algorithm improves reproducibility. Invest Radiol 32: Broderick L, Shemesh J, Wilensky R, Eckert G, Zhou X, Torres W, Balk M, Rogers W, Conces J DJ, Kopecky K (1996) Measurement of coronary artery calcium with dual-slice helical CT compared with coronary angiography: evaluation of CT scoring methods, interobserver variations, and reproducibility. Am J Roentgenol 167: McCollough C, Kaufmann R, Cameron B, Katz D, Sheedy P II, Peyser P (1995) Electron-beam CT: use of a calibration phantom to reduce variability in calcium quantification. Radiology 196: Becker C, Knez A, Jakobs T, Aydemir S, Becker A, Schöpf U, Bruening R, Haberl R, Reiser M (1999) Detection and quantification of coronary artery calcification with electron-beam and conventional CT. Eur Radiol 9: Shemesh J (1994) Spiral methods quantify coronary calcification. Computed Tomography : Shemesh J, Apter S, Rozenman J, Lusky A, Rath S, Itzchak Y, Motro M (1995) Calcification of coronary arteries: detection and quantification with double-helix CT. Radiology 197: Shemesh J, Fisman E, Tenenbaum A, Apter S, Leibovitch L, Rath S, Itzchak Y, Motro M (1997) Coronary artery calcification in women with syndrome X: usefulness of double-helical CT for detection. Radiology 205: Shemesh J, Tenenbaum A, Fisman E, Apter S, Rath S, Rozenman J, Itzchak Y, Motro M (1996) Absence of coronary calcification on double-helical CT scans: predictor of angiographically normal coronary arteries in elderly women? Radiology 199: Kalender W (2001) Computed tomography. Wiley, New York 28. Kachelriess M, Kalender W (1997) ECG-based phase-oriented reconstruction from subsecond spiral CT scans of the heart. Radiology 205: Kachelriess M, Kalender W (1998) ECG-correlated image reconstruction from subsecond spiral CT scans of the heart. Med Phys 25: Kachelriess M, Kalender W (1998) Imaging of the heart by ECG-oriented reconstruction from subsecond spiral multi-row detector CT scans. Radiology 209: Kachelriess M, Kalender W, Karakaya M, Achenbach S, Nossen J, Moshage W, Bautz W (1998) Imaging of the heart by ECG-oriented reconstruction from subsecond spiral CT scans. In: Glazer G, Krestin G (eds) Advances in CT, vol IV. Springer, Berlin Heidelberg New York, pp Kachelriess M, Kalender W (1999) ECG-correlated image reconstruction from subsecond multirow spiral CT scans of the heart: theoretical considerations, phantom measurements and patient studies. Radiology 213:401

14 Kachelriess M, Kalender W (1999) ECG-based phase oriented image reconstruction from subsecond spiral multi-row CT scans of the heart. Eur Radiol 9: Kachelriess M, Ulzheimer S, Kalender W (1999) EKG-basierte phasenorientierte Bildrekonstruktion aus Subsekunden-Mehrzeilen-Spiral- CT-Aufnahmen des Herzens. Rofo Fortschr Geb Röntgenstr Neuen Bildgeb Verfahr (Suppl) 170: Kachelriess M, Ulzheimer S, Kalender W (2001) Neue Entwicklungen in der Computertomographie: EKG korrelierte Subsekunden- Mehrschicht-Spiral-CT. Röntgenpraxis 36. Kachelriess M, Ulzheimer S, Kalender W (2000) ECG-correlated image reconstruction from subsecond multislice spiral CT scans of the heart. Med Phys 27: Kachelriess M, Ulzheimer S, Kalender W (2000) ECG-correlated imaging of the heart with subsecond multislice spiral CT. IEEE Trans Med Imaging 19: Ulzheimer S, Kachelriess M, Kalender W (1999) Improvements of cardiac CT using ECG-oriented image reconstruction in subsecond spiral multirow scanning. Eur Radiol 9 (Suppl 1): Ulzheimer S, Decker R, Kachelriess M, Kalender W (2000) Quality assurance in cardiac CT: phantoms for scanner calibration and comparison of different imaging parameters. Eur Radiol 10 (Suppl 1): Ulzheimer S (2001) Cardiac imaging with X-ray computed tomography: new approaches to image acquisition and quality assurance. Reports from the Institute of Medical Physics, University of Erlangen-Nürnberg. Shaker, Aachen 41. Kalender W, Felsenberg D, Genant H, Fischer M, Dequeker J, Reeve J (1995) The European Spine Phantom: a tool for standardization and quality control in spinal bone mineral measurements by DXA and QCT. Eur J Radiol 20: Kalender W, Süss C, Faust U (1988) Polyethylene-based water- and boneequivalent materials for calibration phantoms in quantitative computed tomography. Biomed Tech Berl 33: Schmermund A, Baumgart D, Gorge G, Seibel R, Gronemeyer D, Ge J, Haude M, Rumberger J, Erbel R (1997) Coronary artery calcium in acute coronary syndromes: a comparative study of electron-beam computed tomography, coronary angiography, and intracoronary ultrasound in survivors of acute myocardial infarction and unstable angina. Circulation 96: Ohnesorge B, Flohr T, Becker C, Knez A, Kopp A, Fukuda K, Reiser M (2000) Cardiac imaging with rapid, retrospective ECG synchronized multilevel spiral CT. Radiologe 40: Ohnesorge B, Flohr T, Becker C, Kopp A, Schöpf U, Baum U, Knez A, Klingenbeck-Regn K, Reiser M (2000) Cardiac imaging by means of electrocardiographically gated multisection spiral CT: initial experience. Radiology 217: Ohnesorge B, Becker C, Flohr T, Reiser M (2002) Multi-slice CT in cardiac imaging. Springer, Berlin Heidelberg New York 47. Hoffmann J (2000) Taschenbuch der Messtechnik, 2nd edn. Fachbuchverlag, Leipzig 48. Achenbach S, Ulzheimer S, Baum U, Kachelriess M, Ropers D, Giesler T, Bautz W, Daniel W, Kalender W, Moshage W (2000) Non-invasive coronary angiography by retrospectively ECG-gated multislice spiral CT. Circulation 102: