- Email: [email protected]
CT number variations due to different image acquisition and reconstruction parameters: a thorax phantom study
Computerized Medical Imaging and Graphics PERGAMON
Computerized Medical Imaging and Graphics 24 (2000) 53–58 www.elsevier.com/locate/compmedimag
CT number variations due to different image acquisition and reconstruction parameters: a thorax phantom study R. Groell*, R. Rienmueller, G.J. Schaffler, H.R. Portugaller, E. Graif, P. Willfurth Department of Radiology, University Hospital Graz, Auenbruggerplatz 9, A-8036, Graz, Austria Received 2 June 1999; accepted 9 December 1999
Abstract A humanoid thorax phantom containing six compartments was scanned with two different computed tomography (CT) scanners using various image acquisition and reconstruction parameters. The differences of CT numbers were statistically significant between the two CT scanners for each compartment p ⬍ 0:001 except for the “air” compartment. The variabilities of the CT numbers are described for the different parameters. The mean CT numbers of the “water” compartment, for instance, ranged from 1 to 15 HU (Hounsfield Units), those of the “air” compartment varied from ⫺962 to ⫺990 HU. Knowledge of these CT number variabilities is necessary when CT numbers are used for tissue characterization. 䉷 2000 Elsevier Science Ltd. All rights reserved. Keywords: Computed tomography—Physics, Spiral, Electron-beam; Attenuation values—Lung
1. Introduction
2. Materials and methods
In ideal computed tomography (CT) the CT numbers reflect the average attenuation coefficient of the corresponding voxel. Based on this assumption it is possible to characterize certain tissues or lesions by their CT number. This is particularly important for example in determining fatcomponents or calcifications within solid masses, in evaluating diffuse lung diseases, in determining bone densities, and in analyzing the contents of cysts or effusion [1–4]. However, it is well known that CT numbers are also influenced by different factors such as beam hardening, scattered radiation, reconstruction artifacts, or object orientation and inhomogeneities which may cause considerable intra- and inter-scanner variabilities of the measured CT numbers [5–10]. For the proper interpretation of CT numbers the radiologist should be aware of the origin and range of these variabilities particularly when these numbers are used for tissue characterization. The purpose of this study was to determine how slice thickness, exposure dose, and reconstruction algorithm influence the CT numbers and the signal-to-noise ratio (SNR) in two different CT scanners with conventionaland electron-beam technology. A commercially available humanoid thorax phantom was used for this purpose.
The phantom (pulmo CT, Siemens, Erlangen, Germany) was 300 × 200 × 30 mm3 in size (in the x-, y- and z-axis, respectively). It contained six compartments consisting of resin equivalent in density to water, soft tissue, cortical- and cancellous-bone, as well as a compartment of cork representing lung tissue, and a compartment of air (Fig. 1). The CT numbers or nominal densities of these compartments as given by the producer are listed in Table 1. The phantom was investigated using two different CT scanners: conventional CT (capable of spiral image acquisition) (Somatom Plus 4, Siemens, Erlangen, Germany) and electron-beam CT (Evolution, Siemens, Erlangen, Germany). In both CT scanners the examinations were performed after the daily routine calibration process according to the user manual and after placing the phantom into the center of rotation.
* Corresponding author. Tel.: ⫹ 43-316-385-2151; fax: ⫹ 43-316-3853266. E-mail address: [email protected] (R. Groell).
2.1. Image acquisition parameters The phantom was scanned with four different slice thicknesses and with two different exposure dosages for each slice thickness. Each CT image was processed by four different reconstruction algorithms, which were routinely used in patient scanning at our institution. The scanning parameters for both scanners are listed in Table 2. The tube voltages available in both scanners were 130 kV for the electron-beam CT scanner and either 120 or 140 kV for the conventional CT scanner. That is why these
0895-6111/00/$ - see front matter 䉷 2000 Elsevier Science Ltd. All rights reserved. PII: S0895-611 1(99)00043-9
54
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58 Table 2 Image acquisition and reconstruction parameters for fast- and electronbeam CT (Al, aluminium)
kV mA Slice thickness (mm) Scan time (sec) mA/s-product Filter (pre-patient) Field-of-view (mm) Object size Reconstruction algorithm
Fig. 1. A humanoid thorax phantom was used in this study. It contained six compartments equivalent in density to water (W), soft tissue (ST), cortical(Co) and cancellous-bone (Ca), lung (L), and air (A).
parameters differed between the two scanners. The incremental “sequence mode” was used for the conventional CT examination. In electron-beam CT the images were performed using the corresponding “single slice mode”. In conventional CT the exposure time was kept constant (1 s), the dosage was changed by varying the tube current (50 vs. 320 mA). In electron-beam CT the images were performed with fixed 625 mA, the exposure dosage was regulated by altering the exposure time (100 vs. 500 ms). In electronbeam CT the 62.5 mA/s scan was not available for the 10 mm slices. The 0.8 mm steel pre-patient filter in electron-beam CT equaled that in conventional CT (10 mm Al equivalent). Each CT scan was repeated three times consecutively. 2.2. CT number evaluation The measurements of the CT numbers were performed on a digital image workstation (DRC 104, Siemens, Erlangen, Germany). In each image regions-of-interest (ROIs) were manually defined in the six compartments using circular areas of 300 mm 2 (for water, soft tissue, air and for cancellous bone), and 1000 mm 2 (for the lung), and a rectangular area of 100 mm 2 (for cortical bone). These regions were drawn with sufficient distance to the contour borders to Table 1 Density specifications for six different compartments inside the phantom as given by the manufacturer (SD, standard deviation; HU, Hounsfield Units; CaHA, Calcium-hydroxyl-apatite) Compartment
Density specifications (Mean ^ SD)
Water Soft tissue Air Lung Cancellous bone Cortical bone
0 ^ 3 HU 40 ^ 20 HU ⫺1000 HU ⫺800 ^ 50 HU 200 ^ 0.5% mgCaHA/ml 400 ^ 0.5% mgCaHA/ml
Conventional CT
Electron-beam CT
120 50, 320 1, 3, 5, 10 1 50, 320 0.8 mm steel 300 Medium AB 10 AB 40 AB 70 AB 90
130 625 1.5, 3, 6, 10 a 0.1, 0.5 62.5, 312.5 10 mm Al-equivalent 300 Normal Smooth Normal Sharp Very sharp
a In electron-beam CT the 100 ms scan is not available for images performed with a slice thickness of 10 mm.
minimize possible partial volume effects. The mean CT numbers and the corresponding standard deviations (SDs) of the CT numbers in the ROIs were determined for each ROI. The measured CT numbers were averaged over the three consecutive scans with identical acquisition and reconstruction parameters and these mean values were used for the further calculation. Signal-to-noise ratio (SNR) in a given ROI was calculated as the ratio of the mean CT number and the standard deviation of the CT numbers. For statistical comparison of the CT numbers in both scanners and to determine the influence of acquisition and reconstruction parameters we performed multivariate analysis using a general linear model (GLM) with a 5% level of statistical security. 3. Results A total of 180 images were obtained and 1080 ROIs were analyzed. For both scanners the overall range of the mean CT numbers and the range of the SD of the CT numbers within the ROIs are listed in Table 3. 3.1. Differences between the two CT scanners (Fig. 2c) The differences of CT numbers were statistically significant between the two CT scanners for each compartment p ⬍ 0:001 except for the “air” compartment in which the CT numbers showed only weak differences between the scanners p 0:065 (Table 3). 3.2. Variation of reconstruction algorithm (Fig. 2a) Amalgamating the results of both scanners the reconstruction algorithms did not significantly effect the mean CT numbers except for the “air” compartment p ⬍ 0:001: With rising sharpness of the reconstruction
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
55
Table 3 Mean CT numbers in six compartments using 32 different image acquisition parameters for two CT scanners (SE, standard error; HU, Hounsfield Units; differences between the two CT scanners were statistically significant for each compartment) Compartment
Water Soft tissue Air Lung Cancellous bone Cortical bone
Conventional CT
Electron-beam CT
Mean ^ SE (HU)
Range (HU)
Mean ^ SE (HU)
Range (HU)
4^1 37 ^ 1 ⫺987 ^ 1 ⫺778 ^ 2 251 ^ 1 465 ^ 1
1–8 32–39 ⫺(962–990) ⫺(760–792) 245–256 457–472
13 ^ 1 43 ^ 1 ⫺980 ^ 1 ⫺757 ^ 3 245 ^ 1 460 ^ 1
10–15 39–46 ⫺(976–987) ⫺(747–762) 243–248 453–463
algorithm the SNR considerably decreased. In conventional CT, the highest SNR (using the “AB10” algorithm) was approximately six times higher than the lowest SNR (using the “AB90” algorithm). In electron-beam CT, the highest SNR (using the “smooth” algorithm) was approximately four times higher than the lowest SNR (using the “very sharp” algorithm). 3.3. Variation of slice thickness (Fig. 2b) The slice thickness influenced the CT numbers of the “air” and of the “cortical bone” compartment significantly p ⬍ 0:001: However, the differences of the mean CT numbers showed no obvious or linear association to the slice thickness on both CT scanners. With increasing slice thickness the SNR gradually decreased in both scanners. The SNR of a 1 or 1.5 mm scan was approximately three times higher than the SNR of a 10 mm scan. 3.4. Variation of dosage (Fig. 2c) The differences of the mean CT numbers between the “high dose scans” and the “low dose scans” were significant for each compartment p ⬍ 0:003 except for the “lung compartment”. The SNR of the “high dose scans” (300 or 312.5 mA/s) was approximately twice as high when compared with the “low dose scans” (50 or 62.5 mA/s). This relation was comparable for each slice thickness and algorithm in both CT-scanners. 4. Discussion Both CT scanners used in this study were calibrated before performing the phantom study. The complex data acquisition and processing systems of modern CT scanners do not allow exact calibrations in a predictable way, but only empirically by using calibration procedures and phantoms. For reason of practicability in the routine work of a hospital this calibration process is kept simple and is, therefore, based on representative calibration scans. For this calibration process an empty scan field (for the conventional CT scanner) or a large homogenous region of water (for electron-beam CT) were used according to the user manual.
Since the quantitative accuracy of a CT scanner is only as good as its calibration, these particular calibration processes may be a major cause of the CT number discrepancies and variabilities in this study. However, this resembles the situation in virtually every routine CT unit. The humanoid thorax phantom consisted of several compartments in a heterogeneous composition. A comparable situation is true for real scanning of living objects, which is usually associated with an even higher degree of heterogeneity. As previous studies have demonstrated CT numbers of the same structure are not at all absolute values but they depend on a variety of factors, which additionally influence themselves [6–9]. These factors may either be machine-related or object-related. Former reports [7–9] have carefully evaluated the influence of object-related factors on CT number variations, with special emphasis on object-orientation in the scan field and object composition. In this study we only used one kind of phantom without changing its position or orientation within the scan field during the examinations (constant object). That is why the differences of the CT numbers as determined in this study were considered to be caused by the alteration of the scanning parameters. The dimension of the phantom was 30 mm in the z-axis, which considerably exceeded the maximum slice thickness (10 mm) used in this study. Moreover, the regions-of-interest were defined in sufficient distance to the contour borders. That is why the measurements were not disturbed by partial volume effects. However, such partial volume effects have to be considered when smaller objects are scanned and when motion artifacts occur, as it may be the case in patient studies. A direct comparison between conventional- and electronbeam CT is difficult, particularly with respect to the SNR as the image acquisition and reconstruction parameters offered by the manufacturer are not identical in the two CT scanners. For instance, the conventional CT scanner offers a wider range of possible reconstruction algorithms, whereas in electron-beam CT only those algorithms investigated in this study are available. The “smooth” algorithm appeared to be smoother in conventional CT when compared to the electron-beam CT scanner, while the “very sharp” algorithm was sharper in conventional CT.
56
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
Fig. 2. Signal-to-noise ratio as a function: (a) of the reconstruction algorithm; (b) of the slice thickness; and (c) of the exposure dosage in conventional- and electron-beam CT.
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
In electron-beam CT a 10 mm scan is generated by scanning during the movement of a 6 mm collimation over a distance of 10 mm, which takes at least 400 ms. That is why the 10 mm scan cannot be performed with an exposure time of 100 ms. We consider this identical collimation to be the reason for the similarities of the measured SNR in 6 and 10 mm scans with electron-beam CT. In electron-beam CT with its 210⬚ geometry the edge artifacts are different from the conventional CT scanner with a 360⬚ geometry. We believe that these artifacts, mainly arising from dense vertebral bone and running through the “water” and “air” compartment of the phantom in electron-beam CT, represent one of the major causes of CT number discrepancies between the two types of CT scanners, particularly with respect to the “water” and “air” compartment. Levi et al. [6] have reported that CT numbers vary not only between different but also between similar scanners produced by the same manufacturer. In general, the data determined in CT-densitometric studies should not be extrapolated to other CT scanners with different soft- and hardware releases. Additionally, possible drift with time of CT scanner normalization has also be taken into account, a factor that was not investigated in this study. Apart from these “technical” problems in CT-densitometry, there is another one that is “biological” in nature: what range of attenuation values may be characteristic for a particular tissue? This was not addressed in this study but has to be considered in any attempt at tissue characterization; therefore further studies using morphological–pathological correlation are necessary to answer this question. Nevertheless, we believe that such quantitative analysis of CT number variabilities as performed in the present study may be helpful for the proper interpretation of the measured CT numbers. In conclusion, CT number variabilities seem to be an immanent problem associated with CT, even in the latest generations of conventional- and electron-beam CT scanners. The radiologist should know the range of these CT number variabilities in his own CT scanner. In order to interpret CT numbers properly one should carefully weigh their possible variabilities against the ranges of attenuation values for the tissue being investigated.
5. Summary CT numbers are often used for tissue characterization for instance to distinguish simple from complex cysts or to detect calcified or fatty components within solid masses.
57
This study was performed to determine the effects of various image acquisition and reconstruction parameters on the measured CT numbers in two different CT scanners. A humanoid thorax phantom was scanned with conventionaland electron-beam CT after calibration of the scanners. The phantom contained six compartments resembling water, soft tissue, cortical- and cancellous-bone, lung, and air. In both CT scanners, images were performed with four different slice thicknesses (1–10 mm), two exposure dosages (50– 312.5 mA/s), and four reconstruction algorithms (“smooth”, “normal”, “sharp”, “very sharp”). On all images the mean CT number and the signal-to-noise ratio were determined for each compartment. The differences of CT numbers were statistically significant between the two CT scanners for each compartment p ⬍ 0:001 except for the “air” compartment. Additionally, CT number variations were observed in both scanners. The mean CT numbers, e.g. of the “water” compartment ranged from 1 to15 HU, those of the “air” compartment varied from ⫺962 to ⫺990 HU and those of the “soft tissue” compartment from 32 to 46 HU. The radiologist has to be aware of these variabilities in his own CT scanner, particularly when CT numbers are used for tissue characterization. References [1] Korobkin M, Brodeur FJ, Yutzy GG, et al. Differentiation of adrenal adenomas from nonadenomas using CT attenuation values. Am J Roentgenol 1996;166:531–6. [2] Rienmuller RK, Behr J, Kalender WA, Schatzl M, Altmann I, Merin M, Beinert T. Standardized quantitative high resolution CT in lung diseases. J Comput Assist Tomogr 1991;15:742–9. [3] Faulkner KG, Gluer CC, Majumdar S, Lang P, Engelke K, Genant HK. Noninvasive measurement of bone mass, structure, and strength: current methods and experimental techniques. Am J Roentgenol 1991;157:1229–37. [4] Bosniak MA. The current radiological approach to renal cysts. Radiology 1986;158:1–10. [5] Kemerink GJ, Lamers RJS, Thelissen GRP, Van Engelshoven JMA. Scanner conformity in CT densitometry of the lungs. Radiology 1995;197:749–52. [6] Levi C, Gray JE, McCullough EC, Hattery RR. The unreliability of CT numbers as absolute values. Am J Roentgenol 1982;139:443–7. [7] McCullough E, Morin RL. CT-number variability in thoracic geometry. Am J Roentgenol 1983;141:135–40. [8] Zerhouni EA, Spivey JF, Morgan RH, Leo FP, Stitik FP, Siegelman SS. Factors influencing quantitative CT measurements of solitary pulmonary nodules. J Comput Assist Tomogr 1982;6(6):1075–87. [9] Baxter BS, Sorenson JA. Factors affecting the measurement of size and CT number in computed tomography. Investigative Radiol 1981;16:337–41. [10] McCollough CH, Kaufmann RB, Cameron BM, Katz DJ, Sheedy PF, Peyser PA. Electron-beam CT: use of a calibration phantom to reduce variability in calcium quantification. Radiology 1995;196:159–65.
58
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
Dr Reinhard Groell is a staff radiologist at the Department of Radiology at the Karl-Franzens-University of Graz. He is currently undergoing a fellowship in nuclear medicine. His areas of interest include body-computed tomography and quantitative image analysis.
Dr Rainer Rienmueller is Professor of Radiology at the KarlFranzens-University of Graz and Chairman of the Department of General Radiology. He is particularly interested in cardiac and thoracic radiology as well as in quantitative image analysis.
Dr Gottfried J. Schaffler is a fellow in radiology and nuclear medicine at the Department of Radiology at the Karl-Franzens-University of Graz.
Dr Horst R. Portugaller is a fellow at the Department of Radiology at the Karl-Franzens-University of Graz specializing in interventional techniques.
Dipl.-Ing. Ewald Graif worked as a Medical Engineer at the Department of General Radiology at the Karl-Franzens-University of Graz. He is currently teaching Electrical Engineering at the Fachhochschule Joanneum Graz.
Dr Peter Willfurth is a resident at the Department of Radiology at the Karl-Franzens-University Graz. His areas of interest are radiation therapy and medical statistics.
Computerized Medical Imaging and Graphics 24 (2000) 53–58 www.elsevier.com/locate/compmedimag
CT number variations due to different image acquisition and reconstruction parameters: a thorax phantom study R. Groell*, R. Rienmueller, G.J. Schaffler, H.R. Portugaller, E. Graif, P. Willfurth Department of Radiology, University Hospital Graz, Auenbruggerplatz 9, A-8036, Graz, Austria Received 2 June 1999; accepted 9 December 1999
Abstract A humanoid thorax phantom containing six compartments was scanned with two different computed tomography (CT) scanners using various image acquisition and reconstruction parameters. The differences of CT numbers were statistically significant between the two CT scanners for each compartment p ⬍ 0:001 except for the “air” compartment. The variabilities of the CT numbers are described for the different parameters. The mean CT numbers of the “water” compartment, for instance, ranged from 1 to 15 HU (Hounsfield Units), those of the “air” compartment varied from ⫺962 to ⫺990 HU. Knowledge of these CT number variabilities is necessary when CT numbers are used for tissue characterization. 䉷 2000 Elsevier Science Ltd. All rights reserved. Keywords: Computed tomography—Physics, Spiral, Electron-beam; Attenuation values—Lung
1. Introduction
2. Materials and methods
In ideal computed tomography (CT) the CT numbers reflect the average attenuation coefficient of the corresponding voxel. Based on this assumption it is possible to characterize certain tissues or lesions by their CT number. This is particularly important for example in determining fatcomponents or calcifications within solid masses, in evaluating diffuse lung diseases, in determining bone densities, and in analyzing the contents of cysts or effusion [1–4]. However, it is well known that CT numbers are also influenced by different factors such as beam hardening, scattered radiation, reconstruction artifacts, or object orientation and inhomogeneities which may cause considerable intra- and inter-scanner variabilities of the measured CT numbers [5–10]. For the proper interpretation of CT numbers the radiologist should be aware of the origin and range of these variabilities particularly when these numbers are used for tissue characterization. The purpose of this study was to determine how slice thickness, exposure dose, and reconstruction algorithm influence the CT numbers and the signal-to-noise ratio (SNR) in two different CT scanners with conventionaland electron-beam technology. A commercially available humanoid thorax phantom was used for this purpose.
The phantom (pulmo CT, Siemens, Erlangen, Germany) was 300 × 200 × 30 mm3 in size (in the x-, y- and z-axis, respectively). It contained six compartments consisting of resin equivalent in density to water, soft tissue, cortical- and cancellous-bone, as well as a compartment of cork representing lung tissue, and a compartment of air (Fig. 1). The CT numbers or nominal densities of these compartments as given by the producer are listed in Table 1. The phantom was investigated using two different CT scanners: conventional CT (capable of spiral image acquisition) (Somatom Plus 4, Siemens, Erlangen, Germany) and electron-beam CT (Evolution, Siemens, Erlangen, Germany). In both CT scanners the examinations were performed after the daily routine calibration process according to the user manual and after placing the phantom into the center of rotation.
* Corresponding author. Tel.: ⫹ 43-316-385-2151; fax: ⫹ 43-316-3853266. E-mail address: [email protected] (R. Groell).
2.1. Image acquisition parameters The phantom was scanned with four different slice thicknesses and with two different exposure dosages for each slice thickness. Each CT image was processed by four different reconstruction algorithms, which were routinely used in patient scanning at our institution. The scanning parameters for both scanners are listed in Table 2. The tube voltages available in both scanners were 130 kV for the electron-beam CT scanner and either 120 or 140 kV for the conventional CT scanner. That is why these
0895-6111/00/$ - see front matter 䉷 2000 Elsevier Science Ltd. All rights reserved. PII: S0895-611 1(99)00043-9
54
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58 Table 2 Image acquisition and reconstruction parameters for fast- and electronbeam CT (Al, aluminium)
kV mA Slice thickness (mm) Scan time (sec) mA/s-product Filter (pre-patient) Field-of-view (mm) Object size Reconstruction algorithm
Fig. 1. A humanoid thorax phantom was used in this study. It contained six compartments equivalent in density to water (W), soft tissue (ST), cortical(Co) and cancellous-bone (Ca), lung (L), and air (A).
parameters differed between the two scanners. The incremental “sequence mode” was used for the conventional CT examination. In electron-beam CT the images were performed using the corresponding “single slice mode”. In conventional CT the exposure time was kept constant (1 s), the dosage was changed by varying the tube current (50 vs. 320 mA). In electron-beam CT the images were performed with fixed 625 mA, the exposure dosage was regulated by altering the exposure time (100 vs. 500 ms). In electronbeam CT the 62.5 mA/s scan was not available for the 10 mm slices. The 0.8 mm steel pre-patient filter in electron-beam CT equaled that in conventional CT (10 mm Al equivalent). Each CT scan was repeated three times consecutively. 2.2. CT number evaluation The measurements of the CT numbers were performed on a digital image workstation (DRC 104, Siemens, Erlangen, Germany). In each image regions-of-interest (ROIs) were manually defined in the six compartments using circular areas of 300 mm 2 (for water, soft tissue, air and for cancellous bone), and 1000 mm 2 (for the lung), and a rectangular area of 100 mm 2 (for cortical bone). These regions were drawn with sufficient distance to the contour borders to Table 1 Density specifications for six different compartments inside the phantom as given by the manufacturer (SD, standard deviation; HU, Hounsfield Units; CaHA, Calcium-hydroxyl-apatite) Compartment
Density specifications (Mean ^ SD)
Water Soft tissue Air Lung Cancellous bone Cortical bone
0 ^ 3 HU 40 ^ 20 HU ⫺1000 HU ⫺800 ^ 50 HU 200 ^ 0.5% mgCaHA/ml 400 ^ 0.5% mgCaHA/ml
Conventional CT
Electron-beam CT
120 50, 320 1, 3, 5, 10 1 50, 320 0.8 mm steel 300 Medium AB 10 AB 40 AB 70 AB 90
130 625 1.5, 3, 6, 10 a 0.1, 0.5 62.5, 312.5 10 mm Al-equivalent 300 Normal Smooth Normal Sharp Very sharp
a In electron-beam CT the 100 ms scan is not available for images performed with a slice thickness of 10 mm.
minimize possible partial volume effects. The mean CT numbers and the corresponding standard deviations (SDs) of the CT numbers in the ROIs were determined for each ROI. The measured CT numbers were averaged over the three consecutive scans with identical acquisition and reconstruction parameters and these mean values were used for the further calculation. Signal-to-noise ratio (SNR) in a given ROI was calculated as the ratio of the mean CT number and the standard deviation of the CT numbers. For statistical comparison of the CT numbers in both scanners and to determine the influence of acquisition and reconstruction parameters we performed multivariate analysis using a general linear model (GLM) with a 5% level of statistical security. 3. Results A total of 180 images were obtained and 1080 ROIs were analyzed. For both scanners the overall range of the mean CT numbers and the range of the SD of the CT numbers within the ROIs are listed in Table 3. 3.1. Differences between the two CT scanners (Fig. 2c) The differences of CT numbers were statistically significant between the two CT scanners for each compartment p ⬍ 0:001 except for the “air” compartment in which the CT numbers showed only weak differences between the scanners p 0:065 (Table 3). 3.2. Variation of reconstruction algorithm (Fig. 2a) Amalgamating the results of both scanners the reconstruction algorithms did not significantly effect the mean CT numbers except for the “air” compartment p ⬍ 0:001: With rising sharpness of the reconstruction
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
55
Table 3 Mean CT numbers in six compartments using 32 different image acquisition parameters for two CT scanners (SE, standard error; HU, Hounsfield Units; differences between the two CT scanners were statistically significant for each compartment) Compartment
Water Soft tissue Air Lung Cancellous bone Cortical bone
Conventional CT
Electron-beam CT
Mean ^ SE (HU)
Range (HU)
Mean ^ SE (HU)
Range (HU)
4^1 37 ^ 1 ⫺987 ^ 1 ⫺778 ^ 2 251 ^ 1 465 ^ 1
1–8 32–39 ⫺(962–990) ⫺(760–792) 245–256 457–472
13 ^ 1 43 ^ 1 ⫺980 ^ 1 ⫺757 ^ 3 245 ^ 1 460 ^ 1
10–15 39–46 ⫺(976–987) ⫺(747–762) 243–248 453–463
algorithm the SNR considerably decreased. In conventional CT, the highest SNR (using the “AB10” algorithm) was approximately six times higher than the lowest SNR (using the “AB90” algorithm). In electron-beam CT, the highest SNR (using the “smooth” algorithm) was approximately four times higher than the lowest SNR (using the “very sharp” algorithm). 3.3. Variation of slice thickness (Fig. 2b) The slice thickness influenced the CT numbers of the “air” and of the “cortical bone” compartment significantly p ⬍ 0:001: However, the differences of the mean CT numbers showed no obvious or linear association to the slice thickness on both CT scanners. With increasing slice thickness the SNR gradually decreased in both scanners. The SNR of a 1 or 1.5 mm scan was approximately three times higher than the SNR of a 10 mm scan. 3.4. Variation of dosage (Fig. 2c) The differences of the mean CT numbers between the “high dose scans” and the “low dose scans” were significant for each compartment p ⬍ 0:003 except for the “lung compartment”. The SNR of the “high dose scans” (300 or 312.5 mA/s) was approximately twice as high when compared with the “low dose scans” (50 or 62.5 mA/s). This relation was comparable for each slice thickness and algorithm in both CT-scanners. 4. Discussion Both CT scanners used in this study were calibrated before performing the phantom study. The complex data acquisition and processing systems of modern CT scanners do not allow exact calibrations in a predictable way, but only empirically by using calibration procedures and phantoms. For reason of practicability in the routine work of a hospital this calibration process is kept simple and is, therefore, based on representative calibration scans. For this calibration process an empty scan field (for the conventional CT scanner) or a large homogenous region of water (for electron-beam CT) were used according to the user manual.
Since the quantitative accuracy of a CT scanner is only as good as its calibration, these particular calibration processes may be a major cause of the CT number discrepancies and variabilities in this study. However, this resembles the situation in virtually every routine CT unit. The humanoid thorax phantom consisted of several compartments in a heterogeneous composition. A comparable situation is true for real scanning of living objects, which is usually associated with an even higher degree of heterogeneity. As previous studies have demonstrated CT numbers of the same structure are not at all absolute values but they depend on a variety of factors, which additionally influence themselves [6–9]. These factors may either be machine-related or object-related. Former reports [7–9] have carefully evaluated the influence of object-related factors on CT number variations, with special emphasis on object-orientation in the scan field and object composition. In this study we only used one kind of phantom without changing its position or orientation within the scan field during the examinations (constant object). That is why the differences of the CT numbers as determined in this study were considered to be caused by the alteration of the scanning parameters. The dimension of the phantom was 30 mm in the z-axis, which considerably exceeded the maximum slice thickness (10 mm) used in this study. Moreover, the regions-of-interest were defined in sufficient distance to the contour borders. That is why the measurements were not disturbed by partial volume effects. However, such partial volume effects have to be considered when smaller objects are scanned and when motion artifacts occur, as it may be the case in patient studies. A direct comparison between conventional- and electronbeam CT is difficult, particularly with respect to the SNR as the image acquisition and reconstruction parameters offered by the manufacturer are not identical in the two CT scanners. For instance, the conventional CT scanner offers a wider range of possible reconstruction algorithms, whereas in electron-beam CT only those algorithms investigated in this study are available. The “smooth” algorithm appeared to be smoother in conventional CT when compared to the electron-beam CT scanner, while the “very sharp” algorithm was sharper in conventional CT.
56
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
Fig. 2. Signal-to-noise ratio as a function: (a) of the reconstruction algorithm; (b) of the slice thickness; and (c) of the exposure dosage in conventional- and electron-beam CT.
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
In electron-beam CT a 10 mm scan is generated by scanning during the movement of a 6 mm collimation over a distance of 10 mm, which takes at least 400 ms. That is why the 10 mm scan cannot be performed with an exposure time of 100 ms. We consider this identical collimation to be the reason for the similarities of the measured SNR in 6 and 10 mm scans with electron-beam CT. In electron-beam CT with its 210⬚ geometry the edge artifacts are different from the conventional CT scanner with a 360⬚ geometry. We believe that these artifacts, mainly arising from dense vertebral bone and running through the “water” and “air” compartment of the phantom in electron-beam CT, represent one of the major causes of CT number discrepancies between the two types of CT scanners, particularly with respect to the “water” and “air” compartment. Levi et al. [6] have reported that CT numbers vary not only between different but also between similar scanners produced by the same manufacturer. In general, the data determined in CT-densitometric studies should not be extrapolated to other CT scanners with different soft- and hardware releases. Additionally, possible drift with time of CT scanner normalization has also be taken into account, a factor that was not investigated in this study. Apart from these “technical” problems in CT-densitometry, there is another one that is “biological” in nature: what range of attenuation values may be characteristic for a particular tissue? This was not addressed in this study but has to be considered in any attempt at tissue characterization; therefore further studies using morphological–pathological correlation are necessary to answer this question. Nevertheless, we believe that such quantitative analysis of CT number variabilities as performed in the present study may be helpful for the proper interpretation of the measured CT numbers. In conclusion, CT number variabilities seem to be an immanent problem associated with CT, even in the latest generations of conventional- and electron-beam CT scanners. The radiologist should know the range of these CT number variabilities in his own CT scanner. In order to interpret CT numbers properly one should carefully weigh their possible variabilities against the ranges of attenuation values for the tissue being investigated.
5. Summary CT numbers are often used for tissue characterization for instance to distinguish simple from complex cysts or to detect calcified or fatty components within solid masses.
57
This study was performed to determine the effects of various image acquisition and reconstruction parameters on the measured CT numbers in two different CT scanners. A humanoid thorax phantom was scanned with conventionaland electron-beam CT after calibration of the scanners. The phantom contained six compartments resembling water, soft tissue, cortical- and cancellous-bone, lung, and air. In both CT scanners, images were performed with four different slice thicknesses (1–10 mm), two exposure dosages (50– 312.5 mA/s), and four reconstruction algorithms (“smooth”, “normal”, “sharp”, “very sharp”). On all images the mean CT number and the signal-to-noise ratio were determined for each compartment. The differences of CT numbers were statistically significant between the two CT scanners for each compartment p ⬍ 0:001 except for the “air” compartment. Additionally, CT number variations were observed in both scanners. The mean CT numbers, e.g. of the “water” compartment ranged from 1 to15 HU, those of the “air” compartment varied from ⫺962 to ⫺990 HU and those of the “soft tissue” compartment from 32 to 46 HU. The radiologist has to be aware of these variabilities in his own CT scanner, particularly when CT numbers are used for tissue characterization. References [1] Korobkin M, Brodeur FJ, Yutzy GG, et al. Differentiation of adrenal adenomas from nonadenomas using CT attenuation values. Am J Roentgenol 1996;166:531–6. [2] Rienmuller RK, Behr J, Kalender WA, Schatzl M, Altmann I, Merin M, Beinert T. Standardized quantitative high resolution CT in lung diseases. J Comput Assist Tomogr 1991;15:742–9. [3] Faulkner KG, Gluer CC, Majumdar S, Lang P, Engelke K, Genant HK. Noninvasive measurement of bone mass, structure, and strength: current methods and experimental techniques. Am J Roentgenol 1991;157:1229–37. [4] Bosniak MA. The current radiological approach to renal cysts. Radiology 1986;158:1–10. [5] Kemerink GJ, Lamers RJS, Thelissen GRP, Van Engelshoven JMA. Scanner conformity in CT densitometry of the lungs. Radiology 1995;197:749–52. [6] Levi C, Gray JE, McCullough EC, Hattery RR. The unreliability of CT numbers as absolute values. Am J Roentgenol 1982;139:443–7. [7] McCullough E, Morin RL. CT-number variability in thoracic geometry. Am J Roentgenol 1983;141:135–40. [8] Zerhouni EA, Spivey JF, Morgan RH, Leo FP, Stitik FP, Siegelman SS. Factors influencing quantitative CT measurements of solitary pulmonary nodules. J Comput Assist Tomogr 1982;6(6):1075–87. [9] Baxter BS, Sorenson JA. Factors affecting the measurement of size and CT number in computed tomography. Investigative Radiol 1981;16:337–41. [10] McCollough CH, Kaufmann RB, Cameron BM, Katz DJ, Sheedy PF, Peyser PA. Electron-beam CT: use of a calibration phantom to reduce variability in calcium quantification. Radiology 1995;196:159–65.
58
R. Groell et al. / Computerized Medical Imaging and Graphics 24 (2000) 53–58
Dr Reinhard Groell is a staff radiologist at the Department of Radiology at the Karl-Franzens-University of Graz. He is currently undergoing a fellowship in nuclear medicine. His areas of interest include body-computed tomography and quantitative image analysis.
Dr Rainer Rienmueller is Professor of Radiology at the KarlFranzens-University of Graz and Chairman of the Department of General Radiology. He is particularly interested in cardiac and thoracic radiology as well as in quantitative image analysis.
Dr Gottfried J. Schaffler is a fellow in radiology and nuclear medicine at the Department of Radiology at the Karl-Franzens-University of Graz.
Dr Horst R. Portugaller is a fellow at the Department of Radiology at the Karl-Franzens-University of Graz specializing in interventional techniques.
Dipl.-Ing. Ewald Graif worked as a Medical Engineer at the Department of General Radiology at the Karl-Franzens-University of Graz. He is currently teaching Electrical Engineering at the Fachhochschule Joanneum Graz.
Dr Peter Willfurth is a resident at the Department of Radiology at the Karl-Franzens-University Graz. His areas of interest are radiation therapy and medical statistics.