M e d i c a l P hy s i c s a n d I n f o r m a t i c s • O r i g i n a l R e s e a r c h
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
Patino et al. IRT Performance for Noise Reduction Medical Physics and Informatics Original Research
A Quantitative Comparison of Noise Reduction Across Five Commercial (Hybrid and ModelBased) Iterative Reconstruction Techniques: An Anthropomorphic Phantom Study
Manuel Patino1 Jorge M. Fuentes Koichi Hayano Avinash R. Kambadakone Jennifer W. Uyeda Dushyant V. Sahani
OBJECTIVE. The objective of our study was to compare the performance of three hybrid iterative reconstruction techniques (IRTs) (ASiR, iDose4, SAFIRE) and their respective strengths for image noise reduction on low-dose CT examinations using filtered back projection (FBP) as the standard reference. Also, we compared the performance of these three hybrid IRTs with two model-based IRTs (Veo and IMR) for image noise reduction on low-dose examinations. MATERIALS AND METHODS. An anthropomorphic abdomen phantom was scanned at 100 and 120 kVp and different tube current–exposure time products (25–100 mAs) on three CT systems (for ASiR and Veo, Discovery CT750 HD; for iDose4 and IMR, Brilliance iCT; and for SAFIRE, Somatom Definition Flash). Images were reconstructed using FBP and using IRTs at various strengths. Nine noise measurements (mean ROI size, 423 mm2) on extracolonic fat for the different strengths of IRTs were recorded and compared with FBP using ANOVA. Radiation dose, which was measured as the volume CT dose index and dose-length product, was also compared. RESULTS. There were no significant differences in radiation dose and image noise among the scanners when FBP was used (p > 0.05). Gradual image noise reduction was observed with each increasing increment of hybrid IRT strength, with a maximum noise suppression of approximately 50% (48.2–53.9%). Similar noise reduction was achieved on the scanners by applying specific hybrid IRT strengths. Maximum noise reduction was higher on model-based IRTs (68.3–81.1%) than hybrid IRTs (48.2–53.9%) (p < 0.05). CONCLUSION. When constant scanning parameters are used, radiation dose and image noise on FBP are similar for CT scanners made by different manufacturers. Significant image noise reduction is achieved on low-dose CT examinations rendered with IRTs. The image noise on various scanners can be matched by applying specific hybrid IRT strengths. Model-based IRTs attain substantially higher noise reduction than hybrid IRTs irrespective of the radiation dose.
Patino M, Fuentes JM, Hayano K, Kambadakone AR, Uyeda JW, Sahani DV
Keywords: CT, filtered back projection, iterative reconstruction technique, phantom studies, radiation dose DOI:10.2214/AJR.14.12519 Received January 9, 2014; accepted after revision May 24, 2014. 1
All authors: Department of Radiology, Division of Abdominal Imaging and Intervention, Massachusetts General Hospital, Harvard Medical School, 55 Fruit St, Boston, MA 02114-2696. Address correspondence to D. V. Sahani (
[email protected]).
WEB This is a web exclusive article. AJR 2015; 204:W176–W183 0361–803X/15/2042–W176 © American Roentgen Ray Society
W176
B
ecause of technical advances and its wide availability, CT plays an essential role in patient management, as shown by its rapid and continuous expansion in the field of diagnostic and interventional imaging [1, 2]. The increasing use of CT has posed new challenges regarding radiation overexposure and its potential harmful effects [3, 4]. Therefore, CT users have made substantial efforts to reduce radiation dose, particularly in pediatric patients and small patients in whom these effects are greater [4, 5]. A commonly used approach to decrease radiation dose is reducing the tube current–exposure time product, which has a predictable linear relationship with radiation dose [6–8]. When the conventional reconstruction technique filtered back projection (FBP), a fast and efficient algorithm, is used, excessive dose re-
duction results in increased noise and artifacts that degrade image quality and render images suboptimal for diagnostic interpretation [6, 9–11]. To overcome these limitations and improve image quality at low-dose settings in different body regions, including the abdomen, CT manufacturers have introduced new reconstruction algorithms including hybrid iterative reconstruction techniques (IRTs) (ASiR, GE Healthcare; iDose4, Philips Healthcare; AIDR 3D, Toshiba Medical Systems; IRIS and SAFIRE, Siemens Healthcare) and modelbased IRTs (Veo, GE Healthcare; IMR, Philips Healthcare) [12–16]. Each algorithm uses a unique technical approach for image noise reduction and possesses different applicable strengths that influence image noise and texture. The ability of IRTs to reduce noise on low-dose CT studies has been proven for al-
AJR:204, February 2015
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
IRT Performance for Noise Reduction most all the algorithms independently. However, an objective comparison of the noise reduction potential of these IRTs at their different strengths is lacking. This knowledge could lay a foundation to attain comparable image noise among the various CT systems, thus facilitating the adoption of diverse CT technologies in clinical practice and preserving workflow. The purpose of this investigation was twofold: first, to quantitatively compare the performance of three commercially available hybrid IRTs (ASiR, iDose4, SAFIRE) at their respective strengths for image noise reduction on low-dose CT examinations using FBP as the reference standard; and, second, to quantitatively compare image noise reduction on low-dose examinations achieved by these three hybrid IRTs and by two commercial model-based IRTs (Veo and IMR). To our knowledge, there are no similar reports in the literature to date. Materials and Methods Phantom Previous investigations have shown the feasibility of using either geometric or anthropomor-
TABLE 1: Radiation Doses on Three Different Scanners Acquired at 120 kVp and Various Tube Current–Exposure Time Products CTDIvol (mGy)
Tube Current–Exposure Time Product
Scanner A (64-MDCT)a
25 mAs
1.46
Scanner B (256-MDCT)b Scanner C (128-MDCT)c 1.80
1.73
50 mAs
2.92
3.37
3.42
75 mAs
4.38
4.40
5.07
100 mAs
5.83
5.90
6.76
Note—Radiation output on scanner C was slightly higher than that on scanners A and B when comparable scanning parameters were used. CTDIvol = volume CT dose index. aDiscovery CT750 HD, GE Healthcare. bBrilliance iCT, Philips Healthcare. cSomatom Definition Flash, Siemens Healthcare.
phic phantoms to optimize clinical protocols with various IRTs [13, 17–20]. Based on the results of those studies and because of the considerable radiation exposure derived from multiple examinations using various CT systems, an anthropomorphic 35-cm lower torso phantom (SK250, Phantom Laboratory) was used as a surrogate for a small adult (body weight < 135 lb [61 kg]). The phantom was built using a polymer with material that simulates abdominal contents—including the colon, pelvic bones, and extracolonic fat tis-
sue—in terms of both morphology and attenuation. The anteroposterior and lateral diameters of the phantom were 21 and 28 cm, respectively. The attenuation (mean ± SD) of the colonic wall and extracolonic fat were 10 ± 4 HU and –103 ± 8 HU, respectively. This phantom has been described extensively in a prior report [21].
Scanners This study was performed on the following MDCT units: scanner A, a 64-MDCT unit (Discovery CT750 HD, GE Healthcare); scanner B, a 256-MDCT unit (Brilliance iCT, Philips Healthcare); and scanner C, a 128-MDCT unit (Somatom Definition Flash, Siemens Healthcare).
Rationale for Scanning Technique
A
B
Our protocols for routine abdominal CT examinations have been modified over the past few years to reduce dose. These efforts have focused preferentially on adjustments of the tube current– exposure time product (mAs) on scanners without IRTs. Small adults (body weight < 60 kg) are scanned on scanners not equipped with IRTs using a tube voltage of 120 kVp and tube current–exposure time products of approximately 60–100 mAs [22]. The baseline phantom CT scan matched our current clinical protocol in small adults and therefore served as the reference standard for comparisons of both image noise and dose. Dose-modified examinations were performed for quantitative comparison of image noise using various strengths of influence of IRTs.
Scanning Technique
C Fig. 1—CT images of phantom show positions of ROIs at different levels for noise measurements. A, Scout image shows levels of noise measurements (dashed lines). B–D, Axial images show ROIs (circles) in region of soft-tissue density.
D
The phantom was placed supine on the CT table with the abdominal structures at the isocenter to the gantry in a craniocaudal orientation to simulate the technique used for a standard abdominal CT examination. The scanning length was selected to cover the entire phantom. Based on the adult size represented with the phantom, two tube voltage settings were selected: 100 and 120 kVp.
AJR:204, February 2015 W177
Noise (SD of CT Number) (HU)
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
35
Scanner B Scanner C Scanner A
30 25 20 15 10 5 0
0
1
2
3
4
5
Noise (SD of CT Number) (HU)
Patino et al.
20 15 10 5 0
0
1
2
3
4
5
A
Image Reconstructions Images of 5 mm thickness were reconstructed in the axial plane using FBP on each scanner. The three scanners generated a total of 24 FBP datasets (n = 8 image datasets for each scanner). Similarly, 5-mm-thick axial images were reconstructed using the five reconstruction algorithms. All the images were reconstructed on the CT system console except the two model-based IRTs (Veo and IMR), which were processed on a standalone workstation. All the images were rendered using the soft-tissue filter. Hybrid iterative reconstruction techniques— All the hybrid IRT images were reconstructed on the scanner console using five predetermined levels per IRT to simplify comparison and statistical analysis. Levels 10%, 30%, 50%, 60%, and 80% were applied for ASiR (algorithm A) on the basis of previous clinical reports [14, 23–25]. Levels 1, 3, 4, 5, and 6 were applied for iDose4 (algorithm B), and levels 1, 2, 3, 4, and 5 were applied for SAFIRE (algorithm C). A total of 120 image datasets were generated using hybrid IRTs (40 hybrid IRT image datasets × 3 scanners). Model-based iterative reconstruction techniques—Images were also reconstructed using two model-based IRTs: Veo (algorithm D) and IMR (algorithm E). Eight datasets were obtained
W178
6
7
8
CTDIvol (mGy)
CTDIvol (mGy)
For assessments of image noise reduction on lowdose studies obtained using IRTs compared with our current protocol (120 kVp, 60–100 mAs), four tube current–exposure time product settings (25, 50, 75, and 100 mAs) were matched for each tube voltage selection; thus, eight acquisitions were performed on each CT system. A total of 24 phantom CT scans were performed (8 acquisitions per CT system × 3 CT systems). Other CT parameters were kept the same for each acquisition: gantry rotation speed, 0.5 second; pitch, 0.993; and slice thickness and interval, 5 mm. The automatic tube current modulation tool was applied as practiced routinely for our clinical protocols.
Fig. 2—Filtered back projection (FBP) datasets for CT examinations performed on three different scanners: Scanner A was 64-MDCT unit (Discovery CT750 HD, GE Healthcare); scanner B, 256-MDCT unit (Brilliance iCT, Philips Healthcare); and scanner C, 128-MDCT unit (Somatom Definition Flash, Siemens Healthcare). A and B, Graphs show image noise and volume CT dose index (CTDIvol) measurements for soft-tissue regions obtained on three scanners using 100 kVp (A) and 120 kVp (B).
Scanner B Scanner C Scanner A
25
B
using Veo, and all three available predetermined level selections (L1, L2, and L3) of IMR were used to obtain 24 datasets. Thus, a total of 32 datasets were generated from the model-based IRTs. The total number of datasets was 176 (24 FBP datasets + 120 hybrid IRT datasets + 32 modelbased IRT datasets). The datasets were transferred to a digital PACS diagnostic workstation (Impax RS 3000 review station, AGFA Technical Imaging Systems) for quantitative analysis.
Radiation Dose The volume CT dose index (CTDIvol) and doselength product (DLP) were recorded from the dose reports generated at the end of the examinations. Image noise is inversely related to the square root of the tube current–exposure time product (mAs) as follows [26]: 1 / (mAs). To verify a match between the doses on different scanners and attribute the noise changes only to the applied reconstruction technique, we compared the radiation dose among the CT systems for each one of the scanning acquisitions.
Image Analysis The magnitude of image noise (SD of the CT number) served as the reference for our study, given the variability of noise texture on images generated using the different reconstruction algorithms, including FBP, as reported by Solomon et al. [27]. Because of limitations of the phantom in terms of adequately representing the solid organs, multiple datasets derived from various acquisitions, and the potential for a recall bias of phantom anatomy on each subsequent image dataset, reliable subjective analysis of image quality would be difficult. Therefore, only objective evaluation of image noise was performed independently by one reader with 7 years of experience in interpreting abdominal CT and by a trainee with 1 year of experience using PACS and IRT soft-
ware. Noise measurements and attenuation values (average of CT number) were obtained manually by drawing a circular ROI averaging 423 mm 2 (range, 420–430 mm 2) in the extracolonic region showing soft-tissue density. Three measurements were taken per image level at different locations in the extracolonic simulated fat (Fig. 1). In total, three image levels representative of the proximal, middle, and distal portions of the phantom were objectively assessed; consequently, nine measurements were acquired and recorded for each CT phantom dataset (objective noise, n = 9 measurements; attenuation, n = 9 measurements). In total, 1584 image noise measurements (9 noise measurements × 176 datasets) and 1584 attenuation measurements (9 attenuation measurements × 176 datasets) were recorded. Image noise was compared on the FBP datasets to assess for similarities among the scanners and to have a reference for further noise reduction comparisons among the various algorithms.
Statistical Analysis The data analysis was performed using statistical software (JMP Pro, version 10.0 2012, SAS Institute). Averages and SDs of objective noise, attenuation values (measured in Hounsfield units), and CTDIvol and DLP values were obtained for images reconstructed using FBP and IRTs. Image noise in the datasets processed using various hybrid IRT strengths from a single acquisition obtained using 120 kVp and 50 mAs was compared using ANOVA. The Tukey test was used if ANOVA showed a significant difference. The level of statistical significance for the different analyses was p < 0.05.
Results Radiation Dose Comparisons Among the Scanners Similar radiation doses were found among the three scanners for each phantom acquisition (Table 1). However, the dose estimates from scanner C were slightly higher when compared
AJR:204, February 2015
IRT Performance for Noise Reduction TABLE 2: Equivalent Noise Reduction Achieved by Three Different Hybrid Iterative Reconstruction Techniques (IRTs) at Different Strengths Using Low-Dose Settingsa
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
Algorithm Ab Noise Reduction (%)
Strength of Influence (%)
Algorithm Bc
Noise Reduction (%)
Strength of Influence (Level)
Algorithm Cd
Noise Reduction (%)
Strength of Influence (Level)
Noise Reduction (%)
p
10–20
10
17.0
1
14.3
1
10.0
> 0.05
21–25
30
22.7
3
25.0
2
21.2
> 0.05
26–30
50
30.0
4
30.0
3
28.8
> 0.05
31–40
60
38.8
5
37.3
4
37.7
> 0.05
41–50
80
48.2
6
42.2
5
48.6
0.01 and > 0.05
Note—Algorithm A at 80%, algorithm B at level 6, and algorithm C at level 5 showed noise reductions in the range of 40–50%. Although there were statistical differences among these algorithms with regard to noise reduction (p = 0.01), no significant difference in noise reduction was found between algorithm A at 80% and algorithm C at level 5 (p > 0.05). a120 kVp and 50 mAs. bASiR, GE Healthcare. ciDose4, Philips Healthcare. dSAFIRE, Siemens Healthcare.
with those from scanners A and B. The CTDIvol values for the 120-kVp acquisitions ranged from 1.46 to 6.76 mGy among the CT systems. Quantitative Noise Measurements Filtered back projection—Image noise on FBP images from the three scanners showed an inverse-squared relationship with radiation dose. Higher noise (SDs, 23.7–30.4 HU) was observed at low-dose settings (100 kVp and 25 mAs), and lower noise (8.9–11.7 HU) was observed at the opposite end of the spectrum (120 kVp and 100 mAs) (Fig. 2). No significant difference was found in noise measurements and the magnitude of change in the image noise at various dose parameters on FBP images obtained on scanners A and B (p > 0.05). Scanner C yielded statistically significant lower noise values at 25 and 100 mAs with 120-kVp acquisitions when compared with scanners A and B (p < 0.05). Hybrid iterative reconstruction techniques—The objective noise measurements were significantly lower in hybrid IRT–processed images when compared with FBP images (p < 0.05). Noise measurements on hybrid IRT images followed a trend similar to FBP images along the spectrum of dose settings: Image noise was lower as the applied strength increased for each hybrid IRT (Fig. 3). Along the spectrum of dose settings and strengths, the noise reduction with hybrid IRTs ranged from 3.5% to 48.2% for algorithm A, from 10.6% to 53.9% for algorithm B, and from 9.5% to 51.4% for algorithm C. To compare the various hybrid IRTs with regard to performance for noise reduction, two image datasets obtained at 120 kVp were selected: 100 and 50 mAs (50% reduction). The intent of this ap-
proach was to compare moderate strengths of influence of hybrid IRTs, which were applied to preserve diagnostic quality without introducing significant changes to image texture on the hybrid IRT images. With these settings, the percentage of image noise reduction was compared among the different algorithms using the previously selected strength levels. Similar image noise reduction across the scanners was achieved by applying various strengths of hybrid IRTs as shown in Table 2. Model-based iterative reconstruction techniques—Both model-based IRTs (Veo and IMR) followed a nearly flat trend in image noise reduction; almost the same image noise was present along the dose spectrum as shown in Figure 4. On average, the noise reduction was higher with model-based IRTs than with the highest-level hybrid IRTs (p < 0.05): Noise reduction ranged from 49.3% to 68.3% for algorithm D and from 59.1% to 81.1% for algorithm E (Fig. 5). For the latter algorithm, the strength selections (L1–L3) showed an inverse relationship with image noise, as we previously discussed when applying strength levels of hybrid IRTs. Image noise reduction was higher for algorithm E when its highest selection (L3) was compared with algorithm D (Fig. 4) for all datasets (p < 0.05) except the dataset obtained with settings of 100 kVp and 25 mAs (p > 0.05). Extracolonic Fat Attenuation Uniformity of CT numbers (measured in Hounsfield units) has been reported as a clinical concern involving reconstruction techniques [28]. In our study, attenuation values measured on images generated using all
strengths of influence of the IRT algorithms showed no difference when compared with FBP on each scanner at half our standard dose (p > 0.05). There was a slight difference noted in the attenuation values (± 3 HU) among the different vendors (p < 0.05), as reported by other authors previously [29]. Discussion The results of our study show that when constant scanning parameters are applied on scanners made by different manufacturers, similar radiation dose and image noise values are obtained on FBP images. Additionally, on dose-modified examinations, hybrid IRTs show gradual reduction in the image noise with increasing strengths of IRTs, achieving a maximum image noise reduction of approximately 50% (up to 53.9%). The maximum noise reduction on model-based IRTs is substantially higher (up to 81.1%) than that on hybrid IRTs irrespective of the dose. Radiation dose optimization is a desirable clinical goal, and the use of IRTs facilitates decreased radiation exposure with preservation of image quality on low-dose CT examinations [8, 9]. Various manufacturers of CT systems initially introduced hybrid IRTs, which featured fast image processing and relatively low costs. Recently, two vendors have introduced advanced techniques including model-based IRTs, which use complex mathematic algorithms and are computationally demanding. Each IRT uses a specific approach for image noise reduction and features different strengths of influence over image noise and texture (Tables 3 and 4). There is ample evidence of the clinical benefits of using hybrid IRTs in dose-modified CT exami-
AJR:204, February 2015 W179
Patino et al. TABLE 3: Hybrid Iterative Reconstruction Techniques (IRTs)a Hybrid IRTs SAFIREb
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
Characteristic
iDose4c
ASiRd
Year of FDA approval
2011
2012
2011
No. of available strengths
5 Levels
7 Levels
10 Levels (0% = 100% FBP, 50% = 50% FBP, 100% = 0% FBP)
Time for image generation
Few minutes (almost real time)
Few minutes (almost real time)
Few minutes (almost real time)
Reported dose reduction
≈ 50% [16, 30, 32]
≈ 50% [22, 32]
≈ 50% [8, 31]
Note—FDA = U.S. Food and Drug Administration, FBP = filtered back projection. aAll three hybrid IRTs are statistical-based IRTs that operate in raw data and image space; FBP is used as the start point for image reconstruction. bSiemens Healthcare. cPhilips Healthcare. dGE Healthcare.
nations of the abdomen [12–14, 16, 22, 30– 32]. A few reports have also emerged using one of the model-based IRTs in ultralow-dose examinations [15, 33]. The FDA has recently approved a second model-based IRT for clinical use. There are no reports to date comparing hybrid IRTs and model-based IRTs regarding the extent of image noise reduction with various dose-modified protocols. This study shows the impact of various IRTs on image noise and the influence of their strengths on image noise. Because implementation of these techniques in clinical practice can be complex, this experience can assist in optimizing the scanning parameters of standard- and modified-dose abdominal CT protocols for imaging small patients. The differences found in image noise on 120-kVp datasets with 25- and 100-mAs acquisitions on scanner C compared with scanners A and B were because of the slightly higher radiation output from scanner C for the same scanning parameters and variations in the image reconstruction kernels among the scanners for generating the FBP images. Hybrid IRTs iterate in both the image and the raw data space. Each algorithm takes a slightly different approach for noise reduction and subsequent image quality improvement [9]. These algorithms offer various strengths (in some techniques, combinations of IRT and FBP) that influence image noise and ultimately image quality. Similar image noise reduction across scanners from different manufacturers was achieved using specific strengths of hybrid IRTs when half the standard radiation dose was applied. For instance, noise reduction for algorithm A at 60% was equivalent to noise reduction for algorithm B at level 5 and algorithm C at level 4. Additionally, noise reduction was similar for algorithm A at 50%, algorithm B at level 4, and algorithm C at level 3. These similarities have served as a reference for our
W180
TABLE 4: Model-Based Iterative Reconstruction Techniques (IRTs) Model-Based IRTs Veoa
Characteristic
IMRb
Year of FDA approval
2011
2013
Operation level of the algorithm
Uses statistical and system (x-ray beam) models and uses computationally intensive constrained optimization algorithms that iteratively arrive at the solution
Uses statistical and system (x-ray beam) models and uses computationally intensive constrained optimization algorithms that iteratively arrive at the solution; offers user control that governs the spatial resolution and noise reduction
No. of available strengths
1
3
Time for image generation
> 20 min
5 min
Reported dose reduction
> 50% [15]
> 50% [34]
Note—FDA = U.S. Food and Drug Administration, FBP = filtered back projection. aGE Healthcare. bPhilips Healthcare.
clinical practice, which cares for a diverse patient population demanding flexible and cutting-edge CT technology. On the basis of our study, we have adapted strengths on lowdose CT examinations that require high contrast, such as CT colonoscopy, kidney stone protocols, and CT angiographies, to achieve diagnostic quality examinations: We use algorithm A at 60%, algorithm B at level 5, and algorithm C at level 4. Similarly, the use of strength level 4 for algorithm B and level 3 for algorithm C has been successfully implemented for routine abdominopelvic examinations [8, 22, 30–32]. To achieve image noise comparable to that seen using algorithms B and C, a strength of 50% for algorithm A may be more appropriate than the currently used 30% for algorithm A at our institution. Knowledge about the various CT technologies is now essential because it is not rare for patients to undergo follow-up examinations at different facilities. Thus, being able to obtain images of similar image quality is highly desirable to reduce any ambiguity
among readers. In our experience in a highvolume abdominal imaging department at a large academic institution, attaining CT examinations with similar imaging quality assists in minimizing interpreter variation and preserving workflow, particularly when CT systems from different vendors are used. On the basis of the results of this investigation, we think that 50% dose reduction from the standard dose is possible when lownoise images are obtained using hybrid IRTs. However, it is advisable to apply the highest hybrid IRT strengths to obtain the benefits of maximum image noise reduction. Noise reduction was substantially higher with model-based IRTs than with hybrid IRTs: Model-based IRTs achieved noise reduction up to 81.1% compared with 53.9% reduction with hybrid IRTs. This noise reduction remained robust for all the datasets irrespective of radiation dose, yielding a nearly flat noise measurement trend. The modelbased IRTs preferentially operate in the raw data domain and use various models and
AJR:204, February 2015
IRT Performance for Noise Reduction FBP
ASiR 50%
ASiR 10%
ASiR 60%
ASiR 30%
ASiR 80%
25
Noise (SD of CT Number) (HU)
Noise (SD of CT Number) (HU)
20
15
10
5
0
25
20
Noise (SD of CT Number) (HU)
25
20
FBP
SAFIRE level 3
SAFIRE level 1
SAFIRE level 4
SAFIRE level 2
SAFIRE level 5
iDOSE level 4
iDOSE level 1
iDOSE level 5
iDOSE level 3
iDOSE level 6
10
5
0
50 75 100 Tube Current–Exposure Time Product (mAs)
FBP
15
25
50 75 100 Tube Current–Exposure Time Product (mAs)
B
A
Fig. 3—Graphs show noise measurements for examinations performed using three hybrid iterative reconstruction techniques (IRTs) compared with filtered back projection (FBP) as reference standard. Hybrid IRT images were reconstructed on scanner console using five predetermined levels per IRT. All examinations were performed using tube voltage of 120 kVp. Results are shown for four tube current– exposure time product settings. A–C, Noise measurements at five strength levels for three hybrid IRTs compared with FBP: algorithm A (ASiR, GE Healthcare) (A), algorithm B (iDose4, Philips Healthcare) (B), and algorithm C (SAFIRE, Siemens Healthcare) (C).
15
10
5
0
25
50 75 100 Tube Current–Exposure Time Product (mAs)
C complex and computationally demanding algorithms. Unlike hybrid IRTs, they model CT system optics in addition to system statistics, taking into account the x-ray beam physics. These approaches lead to considerable noise 25 Average Noise (SD of CT Number) (HU)
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
25
and artifact reductions along with improvements in spatial resolution and low-contrast detectability [33]. Theoretically model-based IRTs share few operational basic principles; however, their exact implementations are pro-
prietary to each vendor and little information about the details of each of the specific methods is available. Iteration methods vary between the two algorithms (i.e., Veo and IMR); the iteration methods with IMR are parallel to
FBP scanner A FBP scanner B
20
Algorithm D Algorithm E level L3
15
10
5
0
0
2
4 CTDIvol (mGy)
6
8
Fig. 4—Graph shows image noise and volume CT dose index (CTDIvol) measurements for soft-tissue regions obtained on two different scanners using filtered back projection (FBP) and two model-based iterative reconstruction techniques (IRTs): algorithm D (Veo, GE Healthcare) and algorithm E (IMR, Philips Healthcare). Scanner A was 64-MDCT unit (Discovery CT750 HD, GE Healthcare) and scanner B was 256-MDCT unit (Brilliance iCT, Philips Healthcare). All examinations were performed using tube voltage of 120 kVp. For algorithm E, L3 refers to strength level.
AJR:204, February 2015 W181
Patino et al. 25
25
FBP Noise (SD of CT Number) (HU)
Noise (SD of CT Number) (HU)
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
Veo 20
15
10
5
0
IMR level L2
IMR level L1
IMR level L3
20
15
10
5
0
25 50 75 100 Average Tube Current–Exposure Time Product (mAs)
FBP
25 50 75 100 Average Tube Current–Exposure Time Product (mAs)
A
B
Fig. 5—Graphs show noise measurements for examinations performed using two model-based iterative reconstruction techniques (IRTs) compared with filtered back projection (FBP) as reference standard. All examinations were performed using tube voltage of 120 kVp. Results are shown for four tube current–exposure time product settings. A, Noise measurements for model-based IRT algorithm D (Veo, GE Healthcare) compared with FBP. B, Noise measurements at three strength levels (L1, L2, L3) for model-based IRT algorithm E (IMR, Philips Healthcare) compared with FBP.
expedite the image generation time. IMR also offers choices for controlling noise and spatial resolution with three strength selections (L1, L2, or L3); image noise, however, is minimally different among the three selections. When image noise reduction was compared between the two model-based IRTs, higher noise reduction was obtained with the highest level (L3) of algorithm E along the dose spectrum. However, noise on images obtained using both model-based IRTs was markedly reduced compared with FBP and hybrid IRTs. Because images processed with modelbased IRTs have the lowest image noise irrespective of the scanning protocol, these algorithms provide an opportunity to perform ultralow-dose CT examinations [15]. The implementation of model-based IRTs can be less complex than the implementation of hybrid IRTs because users do not have to select an IRT strength with particular influence on image noise as is necessary for the hybrid IRTs. Also, fast image reconstruction times obtained with IMR result in more efficient and desirable workflow, providing an opportunity for future implementation of ultralow-dose protocols for routine examinations. However, clinical studies are necessary to confirm IMR’s observed potential. The high computational demands and higher costs of model-based IRTs limit availability to a few institutions. This study has some limitations. First, the phantom was limited to a representation of a small adult, excluding various body sizes. Second, because of the many datasets derived from this study and the potential of a re-
W182
call bias of phantom anatomy, no subjective assessments of the images were performed for noise, texture, artifacts, detectability, and diagnostic effectiveness for hybrid IRTs and model-based IRTs. Additional objective measurements such as contrast-to-noise ratio and spatial resolution were not assessed. It is important to be aware that the magnitude of noise is one of the determinants of image quality but is not the only one; therefore, achieving the greatest amount of noise reduction does not necessarily mean attaining better image quality and better diagnostic performance. Additional clinical studies are required to evaluate the impact of different IRT strengths on low-contrast detectability and diagnostic effectiveness. Third, lower tube voltages (< 100 kVp) were not evaluated because our practice does not include those parameters for routine CT examinations of the abdomen. However, the study objectives of quantification and comparison of image noise reduction were achieved. Also, our described approach for protocol optimization with the use of IRTs can be potentially undertaken for other acquisition parameters and used for imaging body regions other than the abdomen. The results of this study lead to five conclusions. First, when constant scanning parameters were used, the radiation dose and image noise for FBP were similar for three CT scanners made by different manufacturers. Second, on low-dose examinations, hybrid IRTs achieve significantly lower noise than the conventional FBP. Third, specific strengths of
hybrid IRTs influence image noise in a comparable fashion across scanners from different manufacturers to allow optimization of abdominal CT protocols in small patients. Fourth, recently introduced model-based IRTs show greater noise reduction than hybrid IRTs and lay a foundation for further improvements in radiation dose reduction on abdominal CT examinations. Fifth, attenuation values are preserved using IRTs. References 1. OECD website. Medical technologies. In: Health at a glance 2011: OECD indicators. dx.doi.org/. Published 2011. Accessed October 9, 2014 2. Rao VM, Levin DC, Parker L, Frangos AJ, Sunshine JH. Trends in utilization rates of the various imaging modalities in emergency departments: nationwide Medicare data from 2000 to 2008. J Am Coll Radiol 2011; 8:706–709 3. Mettler FA Jr, Thomadsen BR, Bhargavan M, et al. Medical radiation exposure in the U.S. in 2006: preliminary results. Health Phys 2008; 95:502–507 4. Berrington de González A, Mahesh M, Kim K, et al. Projected cancer risks from computed tomographic scans performed in the United States in 2007. Arch Intern Med 2009; 169:2071–2077 5. Pearce MS, Salotti JA, Little MP, et al. Radiation exposure from CT scans in childhood and subsequent risk of leukaemia and brain tumours: a retrospective cohort study. Lancet 2012; 380:499–505 6. Desai GS, Thabet A, Elias AYA, Sahani DV. Comparative assessment of three image reconstruction techniques for image quality and radiation dose in patients undergoing abdominopelvic
AJR:204, February 2015
Downloaded from www.ajronline.org by Library Of Medicine on 01/25/15 from IP address 128.103.149.52. Copyright ARRS. For personal use only; all rights reserved
IRT Performance for Noise Reduction multidetector CT examinations. Br J Radiol 2013; 86:20120161 7. McCollough CH, Primak AN, Braun N, Kofler J, Yu L, Christner J. Strategies for reducing radiation dose in CT. Radiol Clin North Am 2009; 47:27–40 8. Hara AK, Wellnitz CV, Paden RG, Pavlicek W, Sahani DV. Reducing body CT radiation dose: beyond just changing the numbers. AJR 2013; 201:33–40 9. Willemink MJ, de Jong PA, Leiner T, et al. Iterative reconstruction techniques for computed tomography. Part 1. Technical principles. Eur Radiol 2013; 23:1623–1631 10. Zhao J, Jin Y, Lu Y, Wang G. A filtered backprojection algorithm for triple-source helical conebeam CT. IEEE Trans Med Imaging 2009; 28:384–393 11. Thrall JH. Radiation exposure in CT scanning and risk: where are we? Radiology 2012; 264:325–328 12. Lee SJ, Park SH, Kim AY, et al. A prospective comparison of standard-dose CT enterography and 50% reduced-dose CT enterography with and without noise reduction for evaluating Crohn disease. AJR 2011; 197:50–57 13. Schindera ST, Diedrichsen L, Müller HC, et al. Iterative reconstruction algorithm for abdominal multidetector CT at different tube voltages: assessment of diagnostic accuracy, image quality, and radiation dose in a phantom study. Radiology 2011; 260:454–462 14. Prakash P, Kalra MK, Kambadakone AK, et al. Reducing abdominal CT radiation dose with adaptive statistical iterative reconstruction technique. Invest Radiol 2010; 45:202–210 15. Pickhardt PJ, Lubner MH, Kim DH, et al. Abdominal CT with model-based iterative reconstruction (MBIR): initial results of a prospective trial comparing ultralow-dose with standard-dose imaging. AJR 2012; 199:1266–1274 16. Kalra MK, Woisetschlager M, Dahlstrom N, et al. Radiation dose reduction with sinogram affirmed iterative reconstruction technique for abdominal computed tomography. J Comput Assist Tomogr
2012; 36:339–346 17. Hara AK, Paden RG, Silva AC, et al. Iterative reconstruction technique for reducing body radiation dose at CT: feasibility study. AJR 2009; 193:764–771 18. Gervaise A, Osemont B, Lecocq S, et al. CT image quality improvement using adaptive iterative dose reduction with wide-volume acquisition on 320-detector CT. Eur Radiol 2012; 22:295–301 19. Yoon MA, Kim SH, Lee JM, et al. Adaptive statistical iterative reconstruction and Veo: assessment of image quality and diagnostic performance in CT colonography at various radiation doses. J Comput Assist Tomogr 2012; 36:596–601 20. Rampado O, Bossi L, Garabello D, Davini O, Ropolo R. Characterization of a computed tomography iterative reconstruction algorithm by image quality evaluations with an anthropomorphic phantom. Eur J Radiol 2012; 81:3172–3177 21. Zalis ME, Perumpillichira JJ, Kim JY, Del Frate C, Magee C, Hahn PF. Polyp size at CT colonography after electronic subtraction cleansing in an anthropomorphic colon phantom. Radiology 2005; 236:118–124 22. Fuentes-Orrego JM, Kambadakone AR, Hayano K, Hahn PF, Sahani DV. Dose-modified 256MDCT of the abdomen using low tube current and hybrid iterative reconstruction. Acad Radiol 2013; 20:1405–1412 23. Mueck FG, Korner M, Scherr MK, et al. Upgrade to iterative image reconstruction (IR) in abdominal MDCT imaging: a clinical study for detailed parameter optimization beyond vendor recommendations using the adaptive statistical iterative reconstruction environment (ASIR). Rofo 2012; 184:229–238 24. Singh S, Kalra MK, Hsieh J, et al. Abdominal CT: comparison of adaptive statistical iterative and filtered back projection reconstruction techniques. Radiology 2010; 257:373–383 25. Kambadakone AR, Chaudhary NA, Desai GD, Nguyen DD, Kulkarni NM, Sahani DV. Low-dose
MDCT and CT enterography of patients with Crohn disease: feasibility of adaptive statistical iterative reconstruction. AJR 2011; 196:(web) W743–W752 26. McNitt-Gray MF. AAPM/RSNA physics tutorial for residents: topics in CT—radiation dose in CT. RadioGraphics 2002; 22:1541–1553 27. Solomon JB, Christianson O, Samei E. Quantitative comparison of noise texture across CT scanners from different manufacturers. Med Phys 2012; 39:6048–6055 28. Noël PB, Fingerle AA, Renger B, Münzel D, Rummeny EJ, Dobritz M. Initial performance characterization of a clinical noise-suppressing reconstruction algorithm for MDCT. AJR 2011; 197:1404–1409 29. Hahn PF, Blake MA, Boland GW. Adrenal lesions: attenuation measurement differences between CT scanners. Radiology 2006; 240:458–463 30. Baker ME, Dong F, Primak A, et al. Contrast-tonoise ratio and low-contrast object resolution on full- and low-dose MDCT: SAFIRE versus filtered back projection in a low-contrast object phantom and in the liver. AJR 2012; 199:8–18 31. Fuentes-Orrego JM, Sahani DV. Low-dose CT in clinical diagnostics. Expert Opin Med Diagn 2013; 7:501–510 32. Desai G, Fuentes-Orrego JM, Kambadakone AR, Sahani DV. Performance of iterative reconstruction and automated tube voltage selection on the image quality and radiation dose in abdominal CT scans. J Comput Assist Tomogr 2013; 37:897–903 33. Deák Z, Grimm JM, Treitl M, et al. Filtered back projection, adaptive statistical iterative reconstruction, and a model-based iterative reconstruction in abdominal CT: an experimental clinical study. Radiology 2013; 266:197–206 34. Suzuki S, Haruyama T, Morita H, Takahashi Y, Matsumoto R. Initial performance evaluation of iterative model reconstruction in abdominal computed tomography. J Comput Assist Tomogr 2014; 38:408–414
AJR:204, February 2015 W183
