- Email: [email protected]
Comparison of pure and hybrid iterative reconstruction techniques with conventional filtered back projection: Image quality assessment in the cervicothoracic region
European Journal of Radiology 82 (2013) 356–360
Contents lists available at SciVerse ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Comparison of pure and hybrid iterative reconstruction techniques with conventional filtered back projection: Image quality assessment in the cervicothoracic region Masaki Katsura ∗ , Jiro Sato, Masaaki Akahane, Izuru Matsuda, Masanori Ishida, Koichiro Yasaka, Akira Kunimatsu, Kuni Ohtomo Department of Radiology, Graduate School of Medicine, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8655, Japan
a r t i c l e
i n f o
Article history: Received 14 August 2012 Received in revised form 25 October 2012 Accepted 2 November 2012 Keywords: Model-based iterative reconstruction Adaptive statistical iterative reconstruction Filtered back projection Streak artifact Image noise Cervicothoracic region
a b s t r a c t Objectives: To evaluate the impact on image quality of three different image reconstruction techniques in the cervicothoracic region: model-based iterative reconstruction (MBIR), adaptive statistical iterative reconstruction (ASIR), and filtered back projection (FBP). Methods: Forty-four patients underwent unenhanced standard-of-care clinical computed tomography (CT) examinations which included the cervicothoracic region with a 64-row multidetector CT scanner. Images were reconstructed with FBP, 50% ASIR-FBP blending (ASIR50), and MBIR. Two radiologists assessed the cervicothoracic region in a blinded manner for streak artifacts, pixilated blotchy appearances, critical reproduction of visually sharp anatomical structures (thyroid gland, common carotid artery, and esophagus), and overall diagnostic acceptability. Objective image noise was measured in the internal jugular vein. Data were analyzed using the sign test and pair-wise Student’s t-test. Results: MBIR images had significant lower quantitative image noise (8.88 ± 1.32) compared to ASIR images (18.63 ± 4.19, P < 0.01) and FBP images (26.52 ± 5.8, P < 0.01). Significant improvements in streak artifacts of the cervicothoracic region were observed with the use of MBIR (P < 0.001 each for MBIR vs. the other two image data sets for both readers), while no significant difference was observed between ASIR and FBP (P > 0.9 for ASIR vs. FBP for both readers). MBIR images were all diagnostically acceptable. Unique features of MBIR images included pixilated blotchy appearances, which did not adversely affect diagnostic acceptability. Conclusions: MBIR significantly improves image noise and streak artifacts of the cervicothoracic region over ASIR and FBP. MBIR is expected to enhance the value of CT examinations for areas where image noise and streak artifacts are problematic. © 2012 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Image reconstruction in computed tomography (CT) is a mathematical process that generates images from the acquired X-ray projection data. Image reconstruction has a fundamental impact on image quality. For a given radiation dose, it is desirable to reconstruct images with the lowest possible noise without sacrificing image accuracy and spatial resolution. Two major categories of methods exist, analytical reconstruction and iterative reconstruction (IR). Methods based on filtered back projection (FBP) are one type of analytical reconstruction that is currently used on most clinical CT systems. In FBP, the reconstruction kernel, also referred to as “filter” or “algorithm” by some CT vendors, is one of the most
∗ Corresponding author. Tel.: +81 3 5800 8666; fax: +81 3 5800 8935. E-mail address: [email protected] (M. Katsura). 0720-048X/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejrad.2012.11.004
important parameters that affect image quality. Generally, there is a trade off between image noise and spatial resolution. In the cervicothoracic region, image noise and streak artifact from the shoulders are problematic and interfere with adequate visualization of anatomical structures and lesions. In such areas, low-pass filter algorithms that decrease noise are usually used for image reconstruction, but these algorithms also degrade spatial resolution. High-pass filter algorithms preserve spatial resolution, however, there is usually too much image noise. IR has recently received much attention in CT because it has many advantages compared with conventional FBP techniques. IR generates a set of synthesized projections by accurately modeling the data collection process in CT. The model incorporates statistical information of the CT system (including photon statistics and electronic acquisition noise), and details of the system optics (including the size of each detector cell, dimensions of the focal spot, and the shape and size of each image voxel), yielding lower image noise and higher spatial resolution compared with FBP.
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360 Table 1 Patient characteristics and CT parameters. Men/women Age (years) Acquisition mode Noise index Tube voltage (kVp) Field of view (mm) Gantry rotation time (s) Table speed (mm per gantry rotation) Detector configuration (mm) Pitch
357
Table 2 Radiation dose descriptors. 26/18 65.5 ± 15.5 Helical 39.6 (at 0.625 mm) 120 350a 0.5 39.37 64 × 0.625 0.984:1
Data are mean ± standard deviation for each value unless indicated otherwise. a A field of view of 350 mm was typically set; however, it was adjusted according to patient size.
One of the first IR algorithms released for clinical use was the adaptive statistical iterative reconstruction (ASIR) algorithm (GE Healthcare, Waukesha, WI, USA). ASIR is a hybrid IR that involves blending with FBP images, and it models just the photons and electronic noise statistics that primarily affect image noise. Prior phantom and clinical studies have already shown that ASIR provides diagnostically acceptable images with a reduction in image noise compared to the FBP algorithm [1–10]. The recently developed model-based iterative reconstruction (MBIR) technique is a pure IR technique that does not involve blending with FBP images (i.e. no reconstruction kernel), and is mathematically more complex and accurate than ASIR. MBIR not only incorporates modeling of photon and noise statistics like ASIR, it also involves modeling of system optics. Phantom experiments have shown that MBIR provides a significant reduction in image noise and streak artifacts, and a significant improvement in spatial resolution [11–13]. However, clinical studies that have directly compared MBIR with ASIR or FBP are limited [14,15]. The purpose of this study was to evaluate the impact on image quality of three different image reconstruction techniques (MBIR, ASIR and FBP) in one of the most common areas that image noise and streak artifacts are problematic: the cervicothoracic region.
CTDIvol (mGy) DLP (mGy-cm)
Lower neck to chest (n = 21)
Lower neck to abdomen (n = 12)
Lower neck to pelvis (n = 14)
6.59 ± 5.59 239.6 ± 200.3
5.76 ± 2.23 282.5 ± 125.5
6.75 ± 2.87 471.6 ± 221.5
Data are mean ± standard deviation of each value. CTDIvol = CT dose index volume; DLP = dose–length products.
cartilage) to chest (n = 19), from lower neck to abdomen (n = 12), and from lower neck to pelvis (n = 13). The main clinical indications for CT were as follows: staging or restaging of known or suspected malignancy (n = 22), follow-up for a pulmonary nodule (n = 9), pneumonia (n = 7), interstitial lung disease (n = 3), and nontuberculous mycobacterial disease (n = 3). 2.2. CT data acquisition CT data were acquired with a 64-row multidetector CT system (Discovery CT750 HD; GE Healthcare). Imaging parameters are summarized in Table 1. CT acquisition involved the use of automatic tube current modulation (ATCM; Auto mA 3D; GE Healthcare) with a fixed noise index (NI) of 35.6 at 0.625 mm, according to our institutional protocol. The operator-selected NI level modulates the tube current during gantry rotation to achieve a predicted average statistical noise level. All images were reconstructed with 0.625 mm thick axial slices, and then images with increased slice thickness of 2.5 mm were created and used for image analysis. Coronal/sagittal reformats were not used for evaluation in this study (discussed later) and only axial slices were used in the present study. The estimated CT dose index volume (CTDIvol) and dose–length product (DLP) were recorded for each image data set following completion of the CT examination, according to the dose report. Radiation dose descriptors are summarized in Table 2.
2. Methods
2.3. Image reconstruction
This retrospective study was compliant with Health Insurance Portability and Accountability Act guidelines and was approved by the Human Research Committee of our Institutional Review Board. The requirement for informed patient consent was waived.
For each patient, images were reconstructed with MBIR, blending of 50% filtered back projection (FBP) and 50% ASIR image data (ASIR50), and FBP, at the same workstation. The blending factor of 50% for ASIR was chosen based on previous literature [8,9] and vendor recommendations. Blending with FBP does not apply to MBIR, as it is a pure IR technique. Thus, three image datasets (MBIR, ASIR50 and FBP) were generated for each patient (Fig. 1). Each image dataset was coded, patient information was removed, and the datasets were randomized before blinded evaluation. For reconstructing FBP (and subsequently ASIR) images, we used the STANDARD kernel (a proprietary kernel of GE Healthcare), according to our institutional protocol for evaluating soft tissue structures in the neck and mediastinum. Reconstruction kernel does not apply to MBIR, since it is a pure IR technique.
2.1. Patients Between March 21, 2011, and March 25, 2011, 47 consecutive patients underwent unenhanced standard-of-care clinical CT examinations which included the cervicothoracic region at a single tertiary care center. Three patients were selected from the 47 patients using a random number table, and to understand the evaluation system, two thoracic radiologists (M.I. and J.S., with 4 and 13 years of experience, respectively) were trained in the subjective grading of image quality. Images of these three patients were subsequently eliminated from the rest of the analysis. Therefore, 44 patients were included in the final analysis. Details of patient demographic information are summarized in Table 1. All patients were age ≥18 years, were able to undergo CT in the supine position with both arms elevated, and were able to remain still for the duration of CT acquisition. Patients underwent CT without intravenous contrast, as instructed by their attending physicians for any reason (e.g. no clinical indication for using contrast, history of a previous adverse reaction to iodine contrast media, or impairment in renal function). Scan range of CT examinations were as follows: from lower neck (the level of thyroid
2.4. Objective image quality Objective measurements were performed for the three image datasets of the 44 patients (for a total of 132 image sets) on a diagnostic workstation (Centricity RA1000; GE Yokogawa Medical Systems) by a radiologist (MK) with 4 years of imaging experience. Circular regions of interest (ROI) approximately 10 mm in diameter were drawn in the homogenous part of the right internal jugular vein and the posterior paravertebral neck muscle at the level of cricoid cartilage. Calcifications and areas with prominent streak artifacts were carefully avoided, and the standard deviation (i.e.
358
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360 Table 3 Objective image quality (image noise measured within each region of interest).
Internal jugular vein (image noise) Posterior paravertebral neck muscles (image noise)
MBIR
ASIR
FBP
8.88 ± 1.32
18.63 ± 4.19
26.52 ± 5.81
9.68 ± 1.29
24.99 ± 6.04
33.63 ± 7.93
Data are mean ± standard deviation (in Hounsfield units). Image noise was expressed as the standard deviation of the values within the region of interest. The differences in image noise were significant among all pairs (P < 0.001, Student’s paired t-test).
structures and overall diagnostic acceptability were graded on a 5point scale (1 = unacceptable; 2 = suboptimal; 3 = average; 4 = above average; 5 = excellent). Image quality characteristics assessed in the present study have been described in the European Guidelines on Quality Criteria for Computerized Tomography [16] and have been used in multiple previous studies in the radiology literature [5,6,8,9]. 2.6. Statistical analysis
Fig. 1. CT of a 42-year-old woman on follow-up for ovarian cancer, reconstructed with (a) FBP, (b) ASIR, and (c) MBIR. Streak artifacts from the shoulders are markedly reduced with the use of MBIR compared to FBP and ASIR.
objective image noise) of Hounsfield units (HU) within the ROI were recorded.
The data were analyzed using JMP 9.0.0 software (SAS Institute, Cary, NC, USA). Whenever possible, results were expressed as the mean ± standard deviation. Interobserver agreement for the two radiologists was estimated for the subjective image quality parameters using Cohen’s weighted kappa (Ä) analysis. The following Ä values were used to indicate agreement: 0.00–0.20, poor agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, good agreement; and 0.81–1.00, excellent agreement. A sign test was used for each reader to compare subjective image quality assessments between image pairs. A Student’s paired t-test was used to determine the significance of differences in objective image quality data between image pairs. To reduce the possibility of significance due to chance because of multiple statistical testing, a Bonferroni correction was applied to the P value, and significance was assumed only when the P value was <0.016.
2.5. Subjective image quality 3. Results Two radiologists (M.I. and J.S.) independently assessed the image datasets of 44 patients (132 image sets) for image quality on the same diagnostic workstation (Centricity RA1000). Images with increased slice thickness of 2.5 mm were used for image interpretation, which is standard in our institution for reading CT of the neck and mediastinum. Images were presented in a randomized order, and both radiologists were blinded to patient data, clinical information, and image reconstruction technique. Both radiologists already had 2 years of experience with ASIR images at the time of the present study, which were introduced to our department in January 2009. They had little experience with MBIR images at the time of the present study, although they became familiar with MBIR images through the training session. In addition to the default preselected soft tissue window settings [window width (WW), 400 HU; window level (WL), 40 HU], radiologists were allowed to change the WW and WL for ease of assessment. For each image dataset, each radiologist specifically focused on the evaluation of the cervicothoracic region, and graded the following characteristics: streak artifacts, pixilated blotchy appearances, critical reproduction of visually sharp anatomical structures (namely, the thyroid gland, the common carotid artery, and the esophagus) and overall diagnostic acceptability. Streak artifacts and pixilated blotchy appearances were graded on a 4-point scale (1 = substantial artifacts affecting diagnostic information; 2 = major artifacts recognized, but diagnosis still possible under limited conditions; 3 = minor artifacts not interfering with diagnostic decision making; 4 = artifacts unapparent or only minimally recognizable). Visual sharpness of anatomical
3.1. Objective image quality MBIR images had significantly lower quantitative image noise in the internal jugular vein (8.88 ± 1.32) compared to ASIR images (18.63 ± 4.19, P < 0.01) and FBP images (26.52 ± 5.8, P < 0.01) (Table 3). Image noise in the posterior paravertebral neck muscles was significantly lower with MBIR images (9.68 ± 1.29) compared to ASIR images (24.99 ± 6.04, P < 0.01) and FBP images (33.63 ± 7.93, P < 0.01). 3.2. Subjective image quality Interobserver agreement between the two radiologists was moderate (Ä = 0.66–0.71) for streak artifacts and pixilated blotchy appearance, whereas it was fair (Ä = 0.52–0.60) for visual sharpness of the thyroid gland, common carotid artery, and esophagus, and was fair (Ä = 0.52–0.60) for overall diagnostic acceptability. Subjective image quality scores for the two radiologists with the MBIR, ASIR, and FBP techniques are summarized in Table 4. For streak artifacts, MBIR images were typically graded with “minor artifacts not interfering with diagnostic decision making” (P < 0.001 each for MBIR vs. the other two image data sets for both readers). ASIR images were typically graded with “major artifacts recognized, but diagnosis still possible under limited conditions” and did not significantly differ from FBP images (P > 0.9 for ASIR vs. FBP for both readers). MBIR images were more often associated with a pixilated
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360
359
Table 4 Subjective image quality scores for the two radiologists. Image quality
Streak artifact (1/2/3/4) Pixilated blotchy appearance (1/2/3/4) Visualization of the thyroid gland (1/2/3/4/5) Visualization of the common carotid artery (1/2/3/4/5) Visualization of the esophagus (1/2/3/4/5) Overall diagnostic acceptability (1/2/3/4/5)
Reader 1
Reader 2
MBIR
ASIR
FBP
MBIR
ASIR
FBP
2/9/33/0† 1/15/28/0† 0/1/13/30/0* 0/0/14/23/7** 0/1/15/24/4** 0/1/13/25/5**
4/38/2/0† 0/0/44/0† 1/9/26/8/0* 0/12/19/13/0** 0/10/25/9/0** 0/7/28/9/0**
4/37/3/0† 0/0/44/0† 1/21/22/0/0* 1/28/15/0/0** 2/26/16/0/0** 1/16/27/0/0**
1/3/40/0† 2/7/35/0† 0/0/15/29/0* 0/1/13/30/0†† 0/2/15/27/0†† 0/0/8/33/3††
3/41/0/0† 0/3/41/0† 0/11/32/1/0* 0/12/32/0/0†† 0/17/23/4/0†† 0/7/34/3/0††
8/36/0/0† 0/0/44/0† 0/27/17/0/0* 1/14/29/0/0†† 0/20/22/2/0†† 0/9/33/2/0††
Data show the frequency of numerical scores given in each category. Using a sign test, the subjective score differences were statistically significant between all pairs for visualization of the thyroid gland for both readers (*P < 0.001), and for visualization of the common carotid artery and the esophagus, and for overall diagnostic acceptability for reader 1 (**P < 0.001). The subjective scores of MBIR were significantly different from ASIR and FBP for streak artifacts and pixilated blotchy appearance for both readers († P < 0.001). The subjective scores of MBIR were significantly different from ASIR and FBP for visualization of the common carotid artery and the esophagus, and for overall diagnostic acceptability for reader 2 (†† P < 0.001).
blotchy appearance compared to the ASIR and FBP images (P < 0.001 each for MBIR vs. the other two image data sets for both readers). The subjective score differences were statistically significant between all image pairs in visualization of the thyroid gland for both readers (P < 0.001). Subjective MBIR scores were significantly different from ASIR and FBP in visualization of the common carotid artery and the esophagus for reader 2 (P < 0.001). For overall diagnostic acceptability, modal scores were “above average” for MBIR, “average” for ASIR, and “suboptimal” or “average” for FBP (P < 0.001 among all pairs for reader 1, and P < 0.001 each for MBIR vs. the other two image data sets for reader 2). There were no “diagnostically unacceptable” MBIR images.
4. Discussion In the present study, image quality characteristics of CT reconstructed with MBIR, ASIR, and FBP were compared. Significant improvements in image noise and streak artifacts in the cervicothoracic region were observed with the use of MBIR compared to ASIR and FBP. MBIR images were all diagnostically acceptable. To the best of our knowledge, this is the first clinical study to directly compare image quality characteristics of the cervicothoracic region in the same patients with three different reconstruction methods, MBIR, ASIR, and FBP. The data processing steps of IR help to improve image quality from the noise and resolution perspectives compared to the conventional FBP, which is based on simpler mathematical assumptions of the tomographic imaging system. The ASIR technique only models photons and electronic noise statistics that primarily affect image noise, which are not as computationally intensive or timeconsuming. This enables near real-time display of images at the time of imaging. On the other hand, an idealized set of system optics (as does FBP) is used in ASIR, which limits its ability to suppress artifacts and to improve spatial resolution. In the present study, ASIR had limited effect on streak artifact suppression and subjective scores did not significantly differ from FBP. As for MBIR, it not only incorporates modeling of photon and noise statistics like ASIR, it also involves modeling system optics. The intensive computations required for incorporating system optics information substantially prolong the reconstruction duration of MBIR compared with FBP as well as other IR techniques. With incorporation of system optics information and therefore a more accurate account of voxel and focal spot size and geometry, one can expect more reduction in image noise and artifacts and improvements in spatial resolution. The present results are consistent with previous reports of phantom experiments that evaluated image quality improvement with MBIR [11,12]. Coronal/sagittal reformats were not used for evaluation in this study. Our preliminary results of phantom experiments indicate that MBIR, ASIR and FBP behave differently in terms of image noise
when reformatted into coronal and sagittal slices (unpublished data). Therefore, to directly compare image quality characteristics, only axial slices were used in the present study. Unique image features noted in the MBIR-reconstructed images include a pixilated blotchy appearance. MBIR images were also described by the readers as being “artificial” or “waxy.” The exact reasons for these MBIR-unique appearances remain unknown, and may be due to inherent differences in image reconstruction. A pixilated blotchy appearance has been described in many initial ASIR reports [1,6–9]. However, this appearance was not prominently seen on ASIR images in the present study, which, according to the vendor, can be attributed to the advancements of the ASIR algorithm that have been made following the earlier studies. In the present study, although the readers had little experience with MBIR images, they became familiar with them through the training session. Overall, these MBIR-unique image features were not overly distracting in the present study, and had little effect on diagnostic acceptability. MBIR is expected to enhance the value of CT examinations for areas where image noise and streak artifacts are problematic, such as the cervicothoracic region. This enables lower dose settings in CT examinations. Previous studies in the chest [14] and the abdomen [15] indicate that MBIR creates high-quality low-dose CT images and has a greater potential than ASIR or FBP to provide diagnostically acceptable low-dose CT images without severely compromising image quality. MBIR should also be assessed for image quality in different body regions that are associated with prominent noise and streak artifacts, such as the pelvis. Another issue to consider for MBIR in a practical setting is the long reconstruction time (about 1 h per case). MBIR may not yet be suitable for evaluation in acute clinical settings (the reconstruction time in the present report was not recorded as it was not a feature in the application software). Several limitations of this study must be considered. First, all of the CT examinations were performed without intravenous contrast medium administration. The contrast enhancement effect itself has not yet been thoroughly assessed in MBIR, and the presence of streak artifacts from the superior vena cava is another important issue that needs to be considered in contrast-enhanced chest CT. A second limitation is that, the ability of MBIR to detect and localize lesions has not been sufficiently assessed in the present study, since none of the patients had pathologic findings in the cervicothoracic region. This needs to be investigated in future clinical studies. A third limitation is the relatively small number of patients in the study. A fourth limitation of the study is that, owing to the difference in image appearance, blinding of the radiologists between the MBIR and ASIR/FBP image sets during subjective image analysis was difficult; however, the image sets acquired with different reconstruction techniques were randomized. Fifth, the results of the present may not apply to other similar IR methods available from other vendors.
360
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360
In conclusion, MBIR significantly improves image noise and streak artifacts in the cervicothoracic region over a hybrid IR algorithm such as ASIR, or conventional analytical reconstruction algorithms such as FBP. Pure IR algorithm such as MBIR is expected to enhance the value of CT examination for areas where image noise and streak artifacts are problematic. Disclosure Authors state no financial relationship to disclose. References [1] Hara AK, Paden RG, Silva AC, et al. Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study. American Journal of Roentgenology 2009;193(3):764–71. [2] Flicek KT, Hara AK, Silva AC, et al. Reducing the radiation dose for CT colonography using adaptive statistical iterative reconstruction: a pilot study. American Journal of Roentgenology 2010;195(1):126–31. [3] Leipsic J, Labounty TM, Heilbron B, et al. Adaptive statistical iterative reconstruction: assessment of image noise and image quality in coronary CT angiography. American Journal of Roentgenology 2010;195(3):649–54. [4] Leipsic J, Nguyen G, Brown J, et al. A prospective evaluation of dose reduction and image quality in chest CT using adaptive statistical iterative reconstruction. American Journal of Roentgenology 2010;195(5):1095–9. [5] Prakash P, Kalra MK, Digumarthy SR, et al. Radiation dose reduction with chest computed tomography using adaptive statistical iterative reconstruction technique: initial experience. Journal of Computer Assisted Tomography 2010;34(1):40–5. [6] Prakash P, Kalra MK, Kambadakone AK, et al. Reducing abdominal CT radiation dose with adaptive statistical iterative reconstruction technique. Investigative Radiology 2010;45(4):202–10.
[7] Sagara Y, Hara AK, Pavlicek W, et al. Abdominal CT: comparison of low-dose CT with adaptive statistical iterative reconstruction and routine-dose CT with filtered back projection in 53 patients. American Journal of Roentgenology 2010;195(3):713–9. [8] Singh S, Kalra MK, Gilman MD, et al. Adaptive statistical iterative reconstruction technique for radiation dose reduction in chest CT: a pilot study. Radiology 2011;259(2):565–73. [9] Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010;257(2):373–83. [10] Cornfeld D, Israel G, Detroy E, et al. Impact of Adaptive Statistical Iterative Reconstruction (ASIR) on radiation dose and image quality in aortic dissection studies: a qualitative and quantitative analysis. American Journal of Roentgenology 2011;196(3):W336–40. [11] Thibault JB, Sauer KD, Bouman CA, et al. A three-dimensional statistical approach to improved image quality for multislice helical CT. Medical Physics 2007;34(11):4526–44. [12] Yu Z, Thibault JB, Bouman CA, et al. Fast model-based X-ray CT reconstruction using spatially nonhomogeneous ICD optimization. IEEE Trans Image Process 2011;20(1):161–75. [13] Suzuki S, Machida H, Tanaka I, et al. Measurement of vascular wall attenuation: comparison of CT angiography using model-based iterative reconstruction with standard filtered back-projection algorithm CT in vitro. European Journal of Radiology 2012;81(11):3348–53. [14] Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. European Radiology 2012;22(8):1613–23. [15] Singh S, Kalra MK, Do S, et al. Comparison of hybrid and pure iterative reconstruction techniques with conventional filtered back projection: dose reduction potential in the abdomen. Journal of Computer Assisted Tomography 2012;36(3):347–53. [16] EUR 16262. European guidelines on quality criteria for computed tomography. www.drs.dk/guidelines/ct/quality/ [accessed 05.08.12].
Contents lists available at SciVerse ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Comparison of pure and hybrid iterative reconstruction techniques with conventional filtered back projection: Image quality assessment in the cervicothoracic region Masaki Katsura ∗ , Jiro Sato, Masaaki Akahane, Izuru Matsuda, Masanori Ishida, Koichiro Yasaka, Akira Kunimatsu, Kuni Ohtomo Department of Radiology, Graduate School of Medicine, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8655, Japan
a r t i c l e
i n f o
Article history: Received 14 August 2012 Received in revised form 25 October 2012 Accepted 2 November 2012 Keywords: Model-based iterative reconstruction Adaptive statistical iterative reconstruction Filtered back projection Streak artifact Image noise Cervicothoracic region
a b s t r a c t Objectives: To evaluate the impact on image quality of three different image reconstruction techniques in the cervicothoracic region: model-based iterative reconstruction (MBIR), adaptive statistical iterative reconstruction (ASIR), and filtered back projection (FBP). Methods: Forty-four patients underwent unenhanced standard-of-care clinical computed tomography (CT) examinations which included the cervicothoracic region with a 64-row multidetector CT scanner. Images were reconstructed with FBP, 50% ASIR-FBP blending (ASIR50), and MBIR. Two radiologists assessed the cervicothoracic region in a blinded manner for streak artifacts, pixilated blotchy appearances, critical reproduction of visually sharp anatomical structures (thyroid gland, common carotid artery, and esophagus), and overall diagnostic acceptability. Objective image noise was measured in the internal jugular vein. Data were analyzed using the sign test and pair-wise Student’s t-test. Results: MBIR images had significant lower quantitative image noise (8.88 ± 1.32) compared to ASIR images (18.63 ± 4.19, P < 0.01) and FBP images (26.52 ± 5.8, P < 0.01). Significant improvements in streak artifacts of the cervicothoracic region were observed with the use of MBIR (P < 0.001 each for MBIR vs. the other two image data sets for both readers), while no significant difference was observed between ASIR and FBP (P > 0.9 for ASIR vs. FBP for both readers). MBIR images were all diagnostically acceptable. Unique features of MBIR images included pixilated blotchy appearances, which did not adversely affect diagnostic acceptability. Conclusions: MBIR significantly improves image noise and streak artifacts of the cervicothoracic region over ASIR and FBP. MBIR is expected to enhance the value of CT examinations for areas where image noise and streak artifacts are problematic. © 2012 Elsevier Ireland Ltd. All rights reserved.
1. Introduction Image reconstruction in computed tomography (CT) is a mathematical process that generates images from the acquired X-ray projection data. Image reconstruction has a fundamental impact on image quality. For a given radiation dose, it is desirable to reconstruct images with the lowest possible noise without sacrificing image accuracy and spatial resolution. Two major categories of methods exist, analytical reconstruction and iterative reconstruction (IR). Methods based on filtered back projection (FBP) are one type of analytical reconstruction that is currently used on most clinical CT systems. In FBP, the reconstruction kernel, also referred to as “filter” or “algorithm” by some CT vendors, is one of the most
∗ Corresponding author. Tel.: +81 3 5800 8666; fax: +81 3 5800 8935. E-mail address: [email protected] (M. Katsura). 0720-048X/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejrad.2012.11.004
important parameters that affect image quality. Generally, there is a trade off between image noise and spatial resolution. In the cervicothoracic region, image noise and streak artifact from the shoulders are problematic and interfere with adequate visualization of anatomical structures and lesions. In such areas, low-pass filter algorithms that decrease noise are usually used for image reconstruction, but these algorithms also degrade spatial resolution. High-pass filter algorithms preserve spatial resolution, however, there is usually too much image noise. IR has recently received much attention in CT because it has many advantages compared with conventional FBP techniques. IR generates a set of synthesized projections by accurately modeling the data collection process in CT. The model incorporates statistical information of the CT system (including photon statistics and electronic acquisition noise), and details of the system optics (including the size of each detector cell, dimensions of the focal spot, and the shape and size of each image voxel), yielding lower image noise and higher spatial resolution compared with FBP.
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360 Table 1 Patient characteristics and CT parameters. Men/women Age (years) Acquisition mode Noise index Tube voltage (kVp) Field of view (mm) Gantry rotation time (s) Table speed (mm per gantry rotation) Detector configuration (mm) Pitch
357
Table 2 Radiation dose descriptors. 26/18 65.5 ± 15.5 Helical 39.6 (at 0.625 mm) 120 350a 0.5 39.37 64 × 0.625 0.984:1
Data are mean ± standard deviation for each value unless indicated otherwise. a A field of view of 350 mm was typically set; however, it was adjusted according to patient size.
One of the first IR algorithms released for clinical use was the adaptive statistical iterative reconstruction (ASIR) algorithm (GE Healthcare, Waukesha, WI, USA). ASIR is a hybrid IR that involves blending with FBP images, and it models just the photons and electronic noise statistics that primarily affect image noise. Prior phantom and clinical studies have already shown that ASIR provides diagnostically acceptable images with a reduction in image noise compared to the FBP algorithm [1–10]. The recently developed model-based iterative reconstruction (MBIR) technique is a pure IR technique that does not involve blending with FBP images (i.e. no reconstruction kernel), and is mathematically more complex and accurate than ASIR. MBIR not only incorporates modeling of photon and noise statistics like ASIR, it also involves modeling of system optics. Phantom experiments have shown that MBIR provides a significant reduction in image noise and streak artifacts, and a significant improvement in spatial resolution [11–13]. However, clinical studies that have directly compared MBIR with ASIR or FBP are limited [14,15]. The purpose of this study was to evaluate the impact on image quality of three different image reconstruction techniques (MBIR, ASIR and FBP) in one of the most common areas that image noise and streak artifacts are problematic: the cervicothoracic region.
CTDIvol (mGy) DLP (mGy-cm)
Lower neck to chest (n = 21)
Lower neck to abdomen (n = 12)
Lower neck to pelvis (n = 14)
6.59 ± 5.59 239.6 ± 200.3
5.76 ± 2.23 282.5 ± 125.5
6.75 ± 2.87 471.6 ± 221.5
Data are mean ± standard deviation of each value. CTDIvol = CT dose index volume; DLP = dose–length products.
cartilage) to chest (n = 19), from lower neck to abdomen (n = 12), and from lower neck to pelvis (n = 13). The main clinical indications for CT were as follows: staging or restaging of known or suspected malignancy (n = 22), follow-up for a pulmonary nodule (n = 9), pneumonia (n = 7), interstitial lung disease (n = 3), and nontuberculous mycobacterial disease (n = 3). 2.2. CT data acquisition CT data were acquired with a 64-row multidetector CT system (Discovery CT750 HD; GE Healthcare). Imaging parameters are summarized in Table 1. CT acquisition involved the use of automatic tube current modulation (ATCM; Auto mA 3D; GE Healthcare) with a fixed noise index (NI) of 35.6 at 0.625 mm, according to our institutional protocol. The operator-selected NI level modulates the tube current during gantry rotation to achieve a predicted average statistical noise level. All images were reconstructed with 0.625 mm thick axial slices, and then images with increased slice thickness of 2.5 mm were created and used for image analysis. Coronal/sagittal reformats were not used for evaluation in this study (discussed later) and only axial slices were used in the present study. The estimated CT dose index volume (CTDIvol) and dose–length product (DLP) were recorded for each image data set following completion of the CT examination, according to the dose report. Radiation dose descriptors are summarized in Table 2.
2. Methods
2.3. Image reconstruction
This retrospective study was compliant with Health Insurance Portability and Accountability Act guidelines and was approved by the Human Research Committee of our Institutional Review Board. The requirement for informed patient consent was waived.
For each patient, images were reconstructed with MBIR, blending of 50% filtered back projection (FBP) and 50% ASIR image data (ASIR50), and FBP, at the same workstation. The blending factor of 50% for ASIR was chosen based on previous literature [8,9] and vendor recommendations. Blending with FBP does not apply to MBIR, as it is a pure IR technique. Thus, three image datasets (MBIR, ASIR50 and FBP) were generated for each patient (Fig. 1). Each image dataset was coded, patient information was removed, and the datasets were randomized before blinded evaluation. For reconstructing FBP (and subsequently ASIR) images, we used the STANDARD kernel (a proprietary kernel of GE Healthcare), according to our institutional protocol for evaluating soft tissue structures in the neck and mediastinum. Reconstruction kernel does not apply to MBIR, since it is a pure IR technique.
2.1. Patients Between March 21, 2011, and March 25, 2011, 47 consecutive patients underwent unenhanced standard-of-care clinical CT examinations which included the cervicothoracic region at a single tertiary care center. Three patients were selected from the 47 patients using a random number table, and to understand the evaluation system, two thoracic radiologists (M.I. and J.S., with 4 and 13 years of experience, respectively) were trained in the subjective grading of image quality. Images of these three patients were subsequently eliminated from the rest of the analysis. Therefore, 44 patients were included in the final analysis. Details of patient demographic information are summarized in Table 1. All patients were age ≥18 years, were able to undergo CT in the supine position with both arms elevated, and were able to remain still for the duration of CT acquisition. Patients underwent CT without intravenous contrast, as instructed by their attending physicians for any reason (e.g. no clinical indication for using contrast, history of a previous adverse reaction to iodine contrast media, or impairment in renal function). Scan range of CT examinations were as follows: from lower neck (the level of thyroid
2.4. Objective image quality Objective measurements were performed for the three image datasets of the 44 patients (for a total of 132 image sets) on a diagnostic workstation (Centricity RA1000; GE Yokogawa Medical Systems) by a radiologist (MK) with 4 years of imaging experience. Circular regions of interest (ROI) approximately 10 mm in diameter were drawn in the homogenous part of the right internal jugular vein and the posterior paravertebral neck muscle at the level of cricoid cartilage. Calcifications and areas with prominent streak artifacts were carefully avoided, and the standard deviation (i.e.
358
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360 Table 3 Objective image quality (image noise measured within each region of interest).
Internal jugular vein (image noise) Posterior paravertebral neck muscles (image noise)
MBIR
ASIR
FBP
8.88 ± 1.32
18.63 ± 4.19
26.52 ± 5.81
9.68 ± 1.29
24.99 ± 6.04
33.63 ± 7.93
Data are mean ± standard deviation (in Hounsfield units). Image noise was expressed as the standard deviation of the values within the region of interest. The differences in image noise were significant among all pairs (P < 0.001, Student’s paired t-test).
structures and overall diagnostic acceptability were graded on a 5point scale (1 = unacceptable; 2 = suboptimal; 3 = average; 4 = above average; 5 = excellent). Image quality characteristics assessed in the present study have been described in the European Guidelines on Quality Criteria for Computerized Tomography [16] and have been used in multiple previous studies in the radiology literature [5,6,8,9]. 2.6. Statistical analysis
Fig. 1. CT of a 42-year-old woman on follow-up for ovarian cancer, reconstructed with (a) FBP, (b) ASIR, and (c) MBIR. Streak artifacts from the shoulders are markedly reduced with the use of MBIR compared to FBP and ASIR.
objective image noise) of Hounsfield units (HU) within the ROI were recorded.
The data were analyzed using JMP 9.0.0 software (SAS Institute, Cary, NC, USA). Whenever possible, results were expressed as the mean ± standard deviation. Interobserver agreement for the two radiologists was estimated for the subjective image quality parameters using Cohen’s weighted kappa (Ä) analysis. The following Ä values were used to indicate agreement: 0.00–0.20, poor agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, good agreement; and 0.81–1.00, excellent agreement. A sign test was used for each reader to compare subjective image quality assessments between image pairs. A Student’s paired t-test was used to determine the significance of differences in objective image quality data between image pairs. To reduce the possibility of significance due to chance because of multiple statistical testing, a Bonferroni correction was applied to the P value, and significance was assumed only when the P value was <0.016.
2.5. Subjective image quality 3. Results Two radiologists (M.I. and J.S.) independently assessed the image datasets of 44 patients (132 image sets) for image quality on the same diagnostic workstation (Centricity RA1000). Images with increased slice thickness of 2.5 mm were used for image interpretation, which is standard in our institution for reading CT of the neck and mediastinum. Images were presented in a randomized order, and both radiologists were blinded to patient data, clinical information, and image reconstruction technique. Both radiologists already had 2 years of experience with ASIR images at the time of the present study, which were introduced to our department in January 2009. They had little experience with MBIR images at the time of the present study, although they became familiar with MBIR images through the training session. In addition to the default preselected soft tissue window settings [window width (WW), 400 HU; window level (WL), 40 HU], radiologists were allowed to change the WW and WL for ease of assessment. For each image dataset, each radiologist specifically focused on the evaluation of the cervicothoracic region, and graded the following characteristics: streak artifacts, pixilated blotchy appearances, critical reproduction of visually sharp anatomical structures (namely, the thyroid gland, the common carotid artery, and the esophagus) and overall diagnostic acceptability. Streak artifacts and pixilated blotchy appearances were graded on a 4-point scale (1 = substantial artifacts affecting diagnostic information; 2 = major artifacts recognized, but diagnosis still possible under limited conditions; 3 = minor artifacts not interfering with diagnostic decision making; 4 = artifacts unapparent or only minimally recognizable). Visual sharpness of anatomical
3.1. Objective image quality MBIR images had significantly lower quantitative image noise in the internal jugular vein (8.88 ± 1.32) compared to ASIR images (18.63 ± 4.19, P < 0.01) and FBP images (26.52 ± 5.8, P < 0.01) (Table 3). Image noise in the posterior paravertebral neck muscles was significantly lower with MBIR images (9.68 ± 1.29) compared to ASIR images (24.99 ± 6.04, P < 0.01) and FBP images (33.63 ± 7.93, P < 0.01). 3.2. Subjective image quality Interobserver agreement between the two radiologists was moderate (Ä = 0.66–0.71) for streak artifacts and pixilated blotchy appearance, whereas it was fair (Ä = 0.52–0.60) for visual sharpness of the thyroid gland, common carotid artery, and esophagus, and was fair (Ä = 0.52–0.60) for overall diagnostic acceptability. Subjective image quality scores for the two radiologists with the MBIR, ASIR, and FBP techniques are summarized in Table 4. For streak artifacts, MBIR images were typically graded with “minor artifacts not interfering with diagnostic decision making” (P < 0.001 each for MBIR vs. the other two image data sets for both readers). ASIR images were typically graded with “major artifacts recognized, but diagnosis still possible under limited conditions” and did not significantly differ from FBP images (P > 0.9 for ASIR vs. FBP for both readers). MBIR images were more often associated with a pixilated
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360
359
Table 4 Subjective image quality scores for the two radiologists. Image quality
Streak artifact (1/2/3/4) Pixilated blotchy appearance (1/2/3/4) Visualization of the thyroid gland (1/2/3/4/5) Visualization of the common carotid artery (1/2/3/4/5) Visualization of the esophagus (1/2/3/4/5) Overall diagnostic acceptability (1/2/3/4/5)
Reader 1
Reader 2
MBIR
ASIR
FBP
MBIR
ASIR
FBP
2/9/33/0† 1/15/28/0† 0/1/13/30/0* 0/0/14/23/7** 0/1/15/24/4** 0/1/13/25/5**
4/38/2/0† 0/0/44/0† 1/9/26/8/0* 0/12/19/13/0** 0/10/25/9/0** 0/7/28/9/0**
4/37/3/0† 0/0/44/0† 1/21/22/0/0* 1/28/15/0/0** 2/26/16/0/0** 1/16/27/0/0**
1/3/40/0† 2/7/35/0† 0/0/15/29/0* 0/1/13/30/0†† 0/2/15/27/0†† 0/0/8/33/3††
3/41/0/0† 0/3/41/0† 0/11/32/1/0* 0/12/32/0/0†† 0/17/23/4/0†† 0/7/34/3/0††
8/36/0/0† 0/0/44/0† 0/27/17/0/0* 1/14/29/0/0†† 0/20/22/2/0†† 0/9/33/2/0††
Data show the frequency of numerical scores given in each category. Using a sign test, the subjective score differences were statistically significant between all pairs for visualization of the thyroid gland for both readers (*P < 0.001), and for visualization of the common carotid artery and the esophagus, and for overall diagnostic acceptability for reader 1 (**P < 0.001). The subjective scores of MBIR were significantly different from ASIR and FBP for streak artifacts and pixilated blotchy appearance for both readers († P < 0.001). The subjective scores of MBIR were significantly different from ASIR and FBP for visualization of the common carotid artery and the esophagus, and for overall diagnostic acceptability for reader 2 (†† P < 0.001).
blotchy appearance compared to the ASIR and FBP images (P < 0.001 each for MBIR vs. the other two image data sets for both readers). The subjective score differences were statistically significant between all image pairs in visualization of the thyroid gland for both readers (P < 0.001). Subjective MBIR scores were significantly different from ASIR and FBP in visualization of the common carotid artery and the esophagus for reader 2 (P < 0.001). For overall diagnostic acceptability, modal scores were “above average” for MBIR, “average” for ASIR, and “suboptimal” or “average” for FBP (P < 0.001 among all pairs for reader 1, and P < 0.001 each for MBIR vs. the other two image data sets for reader 2). There were no “diagnostically unacceptable” MBIR images.
4. Discussion In the present study, image quality characteristics of CT reconstructed with MBIR, ASIR, and FBP were compared. Significant improvements in image noise and streak artifacts in the cervicothoracic region were observed with the use of MBIR compared to ASIR and FBP. MBIR images were all diagnostically acceptable. To the best of our knowledge, this is the first clinical study to directly compare image quality characteristics of the cervicothoracic region in the same patients with three different reconstruction methods, MBIR, ASIR, and FBP. The data processing steps of IR help to improve image quality from the noise and resolution perspectives compared to the conventional FBP, which is based on simpler mathematical assumptions of the tomographic imaging system. The ASIR technique only models photons and electronic noise statistics that primarily affect image noise, which are not as computationally intensive or timeconsuming. This enables near real-time display of images at the time of imaging. On the other hand, an idealized set of system optics (as does FBP) is used in ASIR, which limits its ability to suppress artifacts and to improve spatial resolution. In the present study, ASIR had limited effect on streak artifact suppression and subjective scores did not significantly differ from FBP. As for MBIR, it not only incorporates modeling of photon and noise statistics like ASIR, it also involves modeling system optics. The intensive computations required for incorporating system optics information substantially prolong the reconstruction duration of MBIR compared with FBP as well as other IR techniques. With incorporation of system optics information and therefore a more accurate account of voxel and focal spot size and geometry, one can expect more reduction in image noise and artifacts and improvements in spatial resolution. The present results are consistent with previous reports of phantom experiments that evaluated image quality improvement with MBIR [11,12]. Coronal/sagittal reformats were not used for evaluation in this study. Our preliminary results of phantom experiments indicate that MBIR, ASIR and FBP behave differently in terms of image noise
when reformatted into coronal and sagittal slices (unpublished data). Therefore, to directly compare image quality characteristics, only axial slices were used in the present study. Unique image features noted in the MBIR-reconstructed images include a pixilated blotchy appearance. MBIR images were also described by the readers as being “artificial” or “waxy.” The exact reasons for these MBIR-unique appearances remain unknown, and may be due to inherent differences in image reconstruction. A pixilated blotchy appearance has been described in many initial ASIR reports [1,6–9]. However, this appearance was not prominently seen on ASIR images in the present study, which, according to the vendor, can be attributed to the advancements of the ASIR algorithm that have been made following the earlier studies. In the present study, although the readers had little experience with MBIR images, they became familiar with them through the training session. Overall, these MBIR-unique image features were not overly distracting in the present study, and had little effect on diagnostic acceptability. MBIR is expected to enhance the value of CT examinations for areas where image noise and streak artifacts are problematic, such as the cervicothoracic region. This enables lower dose settings in CT examinations. Previous studies in the chest [14] and the abdomen [15] indicate that MBIR creates high-quality low-dose CT images and has a greater potential than ASIR or FBP to provide diagnostically acceptable low-dose CT images without severely compromising image quality. MBIR should also be assessed for image quality in different body regions that are associated with prominent noise and streak artifacts, such as the pelvis. Another issue to consider for MBIR in a practical setting is the long reconstruction time (about 1 h per case). MBIR may not yet be suitable for evaluation in acute clinical settings (the reconstruction time in the present report was not recorded as it was not a feature in the application software). Several limitations of this study must be considered. First, all of the CT examinations were performed without intravenous contrast medium administration. The contrast enhancement effect itself has not yet been thoroughly assessed in MBIR, and the presence of streak artifacts from the superior vena cava is another important issue that needs to be considered in contrast-enhanced chest CT. A second limitation is that, the ability of MBIR to detect and localize lesions has not been sufficiently assessed in the present study, since none of the patients had pathologic findings in the cervicothoracic region. This needs to be investigated in future clinical studies. A third limitation is the relatively small number of patients in the study. A fourth limitation of the study is that, owing to the difference in image appearance, blinding of the radiologists between the MBIR and ASIR/FBP image sets during subjective image analysis was difficult; however, the image sets acquired with different reconstruction techniques were randomized. Fifth, the results of the present may not apply to other similar IR methods available from other vendors.
360
M. Katsura et al. / European Journal of Radiology 82 (2013) 356–360
In conclusion, MBIR significantly improves image noise and streak artifacts in the cervicothoracic region over a hybrid IR algorithm such as ASIR, or conventional analytical reconstruction algorithms such as FBP. Pure IR algorithm such as MBIR is expected to enhance the value of CT examination for areas where image noise and streak artifacts are problematic. Disclosure Authors state no financial relationship to disclose. References [1] Hara AK, Paden RG, Silva AC, et al. Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study. American Journal of Roentgenology 2009;193(3):764–71. [2] Flicek KT, Hara AK, Silva AC, et al. Reducing the radiation dose for CT colonography using adaptive statistical iterative reconstruction: a pilot study. American Journal of Roentgenology 2010;195(1):126–31. [3] Leipsic J, Labounty TM, Heilbron B, et al. Adaptive statistical iterative reconstruction: assessment of image noise and image quality in coronary CT angiography. American Journal of Roentgenology 2010;195(3):649–54. [4] Leipsic J, Nguyen G, Brown J, et al. A prospective evaluation of dose reduction and image quality in chest CT using adaptive statistical iterative reconstruction. American Journal of Roentgenology 2010;195(5):1095–9. [5] Prakash P, Kalra MK, Digumarthy SR, et al. Radiation dose reduction with chest computed tomography using adaptive statistical iterative reconstruction technique: initial experience. Journal of Computer Assisted Tomography 2010;34(1):40–5. [6] Prakash P, Kalra MK, Kambadakone AK, et al. Reducing abdominal CT radiation dose with adaptive statistical iterative reconstruction technique. Investigative Radiology 2010;45(4):202–10.
[7] Sagara Y, Hara AK, Pavlicek W, et al. Abdominal CT: comparison of low-dose CT with adaptive statistical iterative reconstruction and routine-dose CT with filtered back projection in 53 patients. American Journal of Roentgenology 2010;195(3):713–9. [8] Singh S, Kalra MK, Gilman MD, et al. Adaptive statistical iterative reconstruction technique for radiation dose reduction in chest CT: a pilot study. Radiology 2011;259(2):565–73. [9] Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010;257(2):373–83. [10] Cornfeld D, Israel G, Detroy E, et al. Impact of Adaptive Statistical Iterative Reconstruction (ASIR) on radiation dose and image quality in aortic dissection studies: a qualitative and quantitative analysis. American Journal of Roentgenology 2011;196(3):W336–40. [11] Thibault JB, Sauer KD, Bouman CA, et al. A three-dimensional statistical approach to improved image quality for multislice helical CT. Medical Physics 2007;34(11):4526–44. [12] Yu Z, Thibault JB, Bouman CA, et al. Fast model-based X-ray CT reconstruction using spatially nonhomogeneous ICD optimization. IEEE Trans Image Process 2011;20(1):161–75. [13] Suzuki S, Machida H, Tanaka I, et al. Measurement of vascular wall attenuation: comparison of CT angiography using model-based iterative reconstruction with standard filtered back-projection algorithm CT in vitro. European Journal of Radiology 2012;81(11):3348–53. [14] Katsura M, Matsuda I, Akahane M, et al. Model-based iterative reconstruction technique for radiation dose reduction in chest CT: comparison with the adaptive statistical iterative reconstruction technique. European Radiology 2012;22(8):1613–23. [15] Singh S, Kalra MK, Do S, et al. Comparison of hybrid and pure iterative reconstruction techniques with conventional filtered back projection: dose reduction potential in the abdomen. Journal of Computer Assisted Tomography 2012;36(3):347–53. [16] EUR 16262. European guidelines on quality criteria for computed tomography. www.drs.dk/guidelines/ct/quality/ [accessed 05.08.12].