Image quality comparison of a nonlinear image-based noise reduction technique with a hybrid-type iterative reconstruction for pediatric computed tomography

(1)

Image quality comparison of a nonlinear

image‑based noise reduction technique with a hybrid‑type iterative reconstruction for

pediatric computed tomography

著者渡邊翔太

著者別表示 WATANABE Shota journal or

publication title

博士論文本文Full 学位授与番号 13301甲第5289号

学位名博士（保健学）

学位授与年月日 2021‑03‑22

URL http://hdl.handle.net/2297/00065176

doi: https://doi.org/10.1016/j.ejmp.2020.06.015

Creative Commons : 表示 ‑ 非営利 ‑ 改変禁止 http://creativecommons.org/licenses/by‑nc‑nd/3.0/deed.ja

(2)

Contents lists available atScienceDirect

Physica Medica

journal homepage:www.elsevier.com/locate/ejmp

Original paper

Image quality comparison of a nonlinear image-based noise reduction technique with a hybrid-type iterative reconstruction for pediatric computed tomography

Shota Watanabe

^a,b

, Katsuhiro Ichikawa

^c,⁎

, Hiroki Kawashima

^c

, Yuki Kono

^b,d

, Hiroyuki Kosaka

^d

, Koji Yamada

^d

, Kazunari Ishii

^e

aDivision of Positron Emission Tomography, Institute of Advanced Clinical Medicine, Kindai University Hospital, 377-2 Ohno-Higashi, Osakasayama, Osaka 589-8511, Japan

bGraduate School of Medical Science, Kanazawa University, 5-11-80 Kodatsuno, Kanazawa, Ishikawa 920-0942, Japan

cInstitute of Medical, Pharmaceutical and Health Sciences, Kanazawa University, 5-11-80 Kodatsuno, Kanazawa, Ishikawa 920-0942, Japan

dRadiology Center, Kindai University Hospital, 377-2 Ohno-Higashi, Osakasayama, Osaka 589-8511, Japan

eDepartment of Radiology, Kindai University Faculty of Medicine, 377-2 Ohno-Higashi, Osakasayama, Osaka 589-8511, Japan

A R T I C L E I N F O

Keywords:

Computed tomography Iterative reconstruction Image quality assessment Pediatric abdominal imaging

A B S T R A C T

Purpose:To compare computed tomography (CT) image properties between a vendor-independent image-based noise reduction technique, Image-space Noise Reduction (iNoir) and a hybrid-type iterative reconstruction technique, Adaptive Statistical Iterative Reconstruction (ASIR).

Methods:A cylindrical water phantom, corresponding to pediatric body size, containing soft-tissue-equivalent rod and 12-mg iodine/ml rod was scanned at size-specific dose estimates of 8.4 and 16.7 mGy. For assessments of image quality and noise texture change, task-based system performance function (SPF) and peak frequency difference (PFD) were compared, respectively, among filtered back projection (FBP), IR image with 50%- blending rate (50%ASIR), 100%ASIR, 50%iNoir, and 100%iNoir. Human observer test for pediatric CT images was performed by radiologists.

Results:For the soft-tissue contrast, SPF²of 100%iNoir was the highest. The average SPF²between 0.1 and 0.5 cycles/mm for 100%iNoir increased by approximately 70% compared with FBP, while ASIR indicted slight increases in the frequency region of > 0.2 cycles/mm. For the iodine contrast, 100%iNoir indicated highest values at the spatial frequencies corresponding pediatric artery diameters. The PFDs of iNoir were negligible and lower than that of ASIR. The results of human observer test supported results of SPF²and PFD.

Conclusions:Compared with ASIR, iNoir provided better image quality for pediatric abdominal CT without compromising noise texture change.

1. Introduction

Compared to adults, pediatric patients are more sensitive to ionizing radiation and have greater risk of developing cancer[1]. To mitigate this issue, various techniques, such as automated tube current modulation and the bismuth breast shield, have been developed for dose reduction in pediatric computed tomography (CT) examinations[2,3].

In addition, vendors have recently developed iterative reconstruction (IR) techniques to suppress increased noise in low-dose CT images and

can potentially reduce the radiation dose without compromising image quality[4–6]. Because IR techniques commonly require projection data (raw data), the algorithms are diverse and vendor speciﬁc[7]. There- fore, IR’s noise reduction eﬀect cannot be obtained in CT systems not equipped with an IR algorithm. In contrast, an image-based noise reduction technique, Image-space Noise Reduction (iNoir), has been developed as a vendor-neutral solution for CT noise reduction and is currently commercially available. iNoir can be used with any CT images because it is provided as a function in a three-dimensional (3D) CT

https://doi.org/10.1016/j.ejmp.2020.06.015

Received 26 July 2019; Received in revised form 13 June 2020; Accepted 13 June 2020

⁎Corresponding author at: Institute of Medical, Pharmaceutical and Health Sciences, Kanazawa University, 5-11-80 Kodatsuno, Kanazawa, Ishikawa 920-0942, Japan.

E-mail addresses:shouta-w@med.kindai.ac.jp(S. Watanabe),ichikawa@mhs.mp.kanazawa-u.ac.jp(K. Ichikawa),

kawa3@med.kanazawa-u.ac.jp(H. Kawashima),kounoyuuki1229@gmail.com(Y. Kono),h-kosaka@med.kindai.ac.jp(H. Kosaka), k-yamada@med.kindai.ac.jp(K. Yamada),ishii@med.kindai.ac.jp(K. Ishii).

T

(3)

workstation series, VirtualPlace (AZE Ltd., Kanagawa, Japan), which can receive CT images and send (back) the iNoir-processed images over a network. iNoir reduces image noise through iterative processes that detect image noise in a CT image and then subtract it from the image.

The CT images with slice thicknesses 2 mm or less can be applied to the iNoir process.

The iNoir’s image properties are potentially nonlinear, different from filtered back projection (FBP) images with linear properties, similar to IR images with spatial resolution properties that vary depending on image noise levels and object contrasts[8,9]. In addition, IR images affect noise texture, which can be evaluated with the peak spatial frequency shift presented in the noise power spectrum (NPS) [10,11]. Thus, iNoir may cause noise texture changes. As mentioned above, iNoir is the image-based technique that can be used for any CT systems. If iNoir can provides a noise reduction performance with less spatial resolution loss and texture change compared to existing IR techniques, iNoir may become a useful vendor-neutral tool for image quality improvement or dose reduction.

The present study comparatively evaluated FBP, iNoir, and a hybrid-type IR, Adaptive Statistical Iterative Reconstruction (ASIR), by means of the task transfer function (TTF) as a task-based spatial resolution measurement, NPS by using a phantom corresponding to a pediatric body size, and a human observer test for pediatric clinical CT images.

2. Materials and methods 2.1. Phantom

Fig. 1shows a schematic of the cylindrical phantom we used. Two kinds of rods for measuring TTF, each with a diameter of 30 mm and a height of 30 mm, were placed as shown in thefigure. The cylindrical phantom was filled with water. The phantom’s outer diameter was 170 mm, which corresponds to a representative abdomen size of a two- year-old child[12]. The two rods were made of a soft-tissue-equivalent material and an iodine-enhanced-vessel-equivalent material with a concentration of 12-mg iodine/ml, which present CT numbers of approximately 50 Hounsfield units (HU) and 300 HU at 120 kV, respectively. The water-only part was used for the NPS measurement.

2.2. Data acquisition and reconstruction method

All images were acquired with the Discovery CT750 HD (GE

Healthcare UK Ltd., Buckinghamshire, England), a multi-slice CT scanner. The scan parameters were 120 kV, 100 and 200 mA, 0.5 s/

rotation, a pitch factor of 0.984, and a detector conﬁguration of 64 × 0.625 mm. The volume CT dose index (CTDIvol) values for 100 and 200 mA were 4.2 mGy and 8.4 mGy. An International Atomic Energy Agency (IAEA) survey in 2012 reported median CTDIvolvalues of 5.0, 6.3, 7.6 mGy for abdominal CT in 1–5, 5–10, and 10–15 year-old pediatric patients, respectively, based on data from 146 CT facilities in 40 countries[13]. We selected the dose levels used in our study in accordance with the median CTDIvolof the IAEA survey. Corresponding size-speciﬁc dose estimates (SSDE) [12], calculated from the CTDIvol

and the phantom diameter of 17 cm, were 8.4 mGy and 16.7 mGy, respectively.

FBP and ASIR images were reconstructed using a slice thickness of 0.625 mm, a display ﬁeld-of-view of 256 mm, and a reconstruction kernel for Standard. The FBP images were transported to the commercial workstation, VirtualPlace, and iNoir images were obtained by processing the FBP images on VirtualPlace. The processing time for iNoir was ~2 min per one patient with ~400 images whereas reconstruction time for ASIR was equivalent to that for FBP. iNoir needs additional times for data transfer to the workstation for iNoir and sending (back) from the workstation. The times were ~20 s for 400 images, respectively.

The ASIR algorithm consists of a single backward projection step and subsequent iterative noise reduction processes in image space. For the ASIR algorithm, the FBP image blending rate can be set from 0% to 100% to control the noise reduction level (100%: strongest)[7]. Similar to ASIR, iNoir has blending settings from 0% to 100%. In this study, 50% and 100% blending rates for both ASIR and iNoir were used (50%ASIR and 100%ASIR; 50%iNoir and 100%iNoir, respectively). As the image qualities of FBP have been accepted in clinicalﬁelds to date, we used FBP images with no additional processing as the control data of this study.

2.3. TTF and NPS measurement

The phantom was carefully placed such that the rod objects’central axes were parallel to the rotation axes of the CT system. To improve TTF measurement accuracy, we used an image-averaging technique to signiﬁcantly reduce the image noise by averaging many rod images with precisely the same rod positioning[9,14–16]. Therefore, the acquisition for the rod part was repeated 20 times and obtained 300 CT images used for the averaging. For the disk image (axial image of the

Fig. 1.Schematic of cylindrical phantom containing two rods made of a soft-tissue-equivalent material and an iodine-enhanced-vessel-equivalent material with a concentration of 12-mg iodine/ml.

S. Watanabe, et al. Physica Medica 76 (2020) 100–108

101

(4)

rod object) in the averaged image, a one-dimensional (1D) edge spread function was obtained with the circular edge technique previously reported[8,15,16]. The bin width in the binning process, used to create equidistant edge spread function data and simultaneously reduce noise, was set to one ﬁfth of the pixel pitch (0.5 mm) corresponding to a displayﬁeld-of-view of 256 mm. NPS was measured from each uniform image for the water-only part, and NPS results of 400 images, obtained from four replicate acquisitions, were averaged. For the NPS calcula- tion, we used an established method using 2D Fourier transform [17–19]. The region of interest size was set to 128 × 128 pixels at the center of the image.

2.4. System performance function

Samei and Richard used the following detectability index, d', to assess the IR techniques’imaging performance:

∫

=

d TTF (u)

NPS(u)S (u)du

'2 2

2

(1) whereudenotes the spatial frequency andS(u) is the spectrum of the signal to be detected. The d′² value is aﬁgure of merit that in- corporates square of system performance function (SPF) TTF²(u)/NPS (u) and imaging task S²(u) [20]. This index is similar to the pre- whitening signal-to-noise ratio that is based on an ideal observer model [20–22].

In this study, we focused on this SPF expressed as

= SPF u TTF u

NPS u

( ) ( )

( ).

2 2

(2) Because the TTFs we measured were speciﬁc for rod objects with the soft-tissue and iodine contrasts presenting circular edges, we used this SPF to evaluate the edge-preserving noise reduction performance for the limited but typical two conditions.

Previous studies have reported that the diameters of abdominal aortas were 6.67 ± 0.69 to 12.33 ± 1.01 mm for patients aged between 1 and 11 years and of the hepatic artery was 2.7 ± 0.5 mm for patients aged between 5 months and 15 years[23,24]. Based on these reported diameters, we calculated the vessel signal spectrumS(u) by using the following equation:

= ∅

S u J π u∅ ( ) (π u)

2

1

(3) whereJ1() denotes aﬁrst-order Bessel function andϕis the diameter (mm)[25,26].ϕvalues were set to 2.5, 5.0, and 10.0 mm. Moreover, the highest contributing spatial frequency−u corresponding toϕ was calculated using the mean-square-root bandwidth according to the following equation[27]:

∫

− =

∫

∞ u ∞

u S u du S u du

| ( )|

2 0

2 2

0

2 (4)

and then SPF²values at−ufor the iodine contrast were compared among FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR.

2.5. Peak frequency diﬀerence measurement

To compare the noise texture change caused by iNoir and ASIR processes, the peak frequency diﬀerence (PFD) between a target image and an FBP image was estimated using normalized NPS considering a human visual response [28]. The normalized NPS was calculated by normalizing NPS to its integral across all frequencies. The displayﬁeld- of-viewF, the viewing distanceR, and the size of displayed imageD, for calculating the human visual response, were F = 256 mm, R= 400 mm, andD= 205 mm, according to the human observer test situation described later.

2.6. Human observer test

We performed a human observer test to assess the ASIR and iNoir clinical images of 10 pediatric abdominal CT cases (four boys and six girls; mean age ± standard deviation, 8.5 ± 3.4 years) using relative visual grading analysis[5]. This study was approved by the ethics re- view board of our institution. Prior informed patient consent was waived because the study used existing CT images including raw data.

Patient records and information were anonymized and deidentiﬁed prior to analysis. The authorization number was 29–106.

Ten cases were retrospectively selected. All images were acquired using Discovery CT750 HD and reconstructed with a slice thickness of 0.625 mm, a displayﬁeld of view of 256 mm, and a reconstruction kernel for Standard. The window level (WL) and window width (WW) were adjusted to 50 HU and 300 HU, respectively. The observers were three radiologists with > 15 years clinical experience for pediatric CT images reconstructed with FBP and hybrid-type IRs and they scored the relative visual grade in reference to a corresponding FBP image. In each observation, an FBP image as reference and a corresponding image each of 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR were displayed side by side on the workstation with blinding image reconstruction algorithm. The observer was not allowed to change WL and WW; reading time was not limited. The observers assessed image quality against FBP image in terms of overall image quality and noise texture, reviewing the entire sequence of CT slices of each patient. The overall image quality was graded using the following 7-point scale:−3, deﬁnitely worse;−2, worse; −1, slightly worse; 0, equal; +1, slightly better; +2, better;

+3, deﬁnitely better. The noise texture was graded by using a 4-point scale:−3, deﬁnite change;−2, change;−1, slight change; 0, equal.

The median values of overall image quality and noise texture were compared among the 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR images by using the Wilcoxon signed rank test with Bonferroni cor- rection (n = 4). Differences of p < 0.0125 were considered as in- dicative of statistical significance. All statistical analyzes were performed by using SPSS Statistics 17.0 (SPSS, IBM, Tokyo, Japan). The size of the displayed image and distance between observers and the display were 205 and 400 mm, respectively. The CTDIvolvalues of the 10 pediatric patients were obtained from dose reports provided by the CT scanner as a digital imaging and communications in medicine (DICOM)-compliant image. Moreover, the SSDE values were calculated from the CTDIvol values and the conversion factor, which was de- termined in accordance with the effective diameters for each patient.

The eﬀective diameters were calculated from the anterior-posterior and lateral dimensions measured using an image at a slice level of the upper abdomen including liver.

3. Results

3.1. Image quality assessment

Fig. 2shows the TTFs of FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR for the soft-tissue contrast. Compared to FBP, the TTFs of iNoir and ASIR decreased as the blending rate increased. The TTFs of 100%iNoir and 50%ASIR were equivalent. For the iodine contrast results shown inFig. 3, the TTFs of FBP, 50%iNoir, and 100%iNoir were equivalent. The TTF of ASIR was superior to that of FBP in frequencies > 0.4 cycles/mm for 8.4 mGy and > 0.3 cycles/mm for 16.7 mGy, respectively.

Fig. 4shows measured NPSs. The NPS results of iNoir and ASIR were signiﬁcantly lower than those of FBP. In comparison with the same blending rate (between 50%iNoir and 50%ASIR and between 100%iNoir and 100%ASIR), the noise level in the low frequencies of iNoir was lower than that of ASIR for both 8.4 and 16.7 mGy. However, the noise level of iNoir was higher than that of ASIR in the middle-to- high frequencies, which resulted in diﬀerences in the noise peak frequency described later.

(5)

The SPF²of 100%iNoir was the highest for the soft-tissue contrast, as shown inFig. 5. The average SPF²between 0.1 and 0.5 cycles/mm for 100%iNoir increased by 68.6% for 8.4 mGy and 73.5% for 16.7 mGy compared with FBP. 100%ASIR indicated a diﬀerent trend in which the slope gradually decreased with increasing frequency.

For the iodine contrast, the SPF²of iNoir was improved relative to that of the soft-tissue contrast (Fig. 6). The average SPF²between 0.1 and 0.5 cycles/mm for 100%iNoir increased by 157.4% for 8.4 mGy and 170.4% for 16.7 mGy compared to FBP. On the other hand, ASIR showed characteristic (unnatural) properties in which the SPF²value turned from a decrease to an increase at a certain frequency which varied depending on the blending rate. For 100%ASIR, the SPF²values at higher frequencies exceeded the level at the lowest frequency, which was not realistic compared to typical SPF of FBP with monotonically decreasing.

The spatial frequencies of−ucorresponding to the vessel diameters of 2.5, 5.0, and 10.0 mm were 0.18, 0.10, and 0.05 cycles/mm, respectively. The SPF²values of 100%iNoir for the iodine contrast showed the

highest value at the spatial frequencies. SPF²at 0.18 cycles/mm (the 2.5-mm diameter) with 8.4/16.7 mGy increased by 124.7/134.8% and 48.6/58.5% compared with FBP for 100%iNoir and 100%ASIR, respectively.

3.2. PFD measurement

The PFD results are summarized inTable 1. The PFD values of 50%

iNoir, 100%iNoir, 50%ASIR, and 100%ASIR were 0.01, 0.02, 0.03, and 0.09 cycles/mm at 8.4 mGy and 0.02, 0.03, 0.05, and 0.10 cycles/mm at 16.7 mGy, respectively. The PFD of iNoir was lower than that of ASIR for each comparison and was negligibly small.

3.3. Human observer test

The overall image quality score of 100%iNoir was signiﬁcantly higher than those of the other images, and the 100%ASIR score was signiﬁcantly lower than those of the other images. There was no Fig. 2.TTF measurement results for the soft-tissue-equivalent rod for FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR at (a) 8.4 and (b) 16.7 mGy.

Fig. 3.TTF measurement results for the 12-mgI iodine/ml rod for FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR at (a) 8.4 and (b) 16.7 mGy.

103

(6)

significant difference between 50%iNoir and 50%ASIR (Fig. 7a). For the noise texture change from FBP, 100%ASIR scored significantly lower than did the other images. There were no significant differences between 50%iNoir and 50%ASIR, 100%iNoir and 50%ASIR, respectively (Fig. 7b). The average CTDIvoland SSDE of the patients enrolled were 4.2 ± 1.5 mGy and 8.2 ± 2.4 mGy, respectively.

3.4. Phantom and clinical image

Fig. 8shows the representative transverse soft-tissue-equivalent rod images of FBP, 50%ASIR, 100%ASIR, 50%iNoir, and 100%iNoir. Al- though the image noise was remarkably decreased by 100%ASIR, sig- niﬁcant image blurring and noise texture change was observed in the image. iNoir naturally decreased the image noise and well preserved the object edge. Fig. 9 shows iodine rod images. Although the 100%ASIR gave images with a sharper object edge, the image appearance became unnatural. iNoir naturally reduced the image noise in a manner similar to that of the soft-tissue rod images. The abdominal, non-enhanced, CT images of FBP, 100%ASIR, and 100%iNoir of a six-

year-old girl are shown inFig. 10. The CTDIvolvalue indicated on the operator console and SSDE value for this case were 5.1 mGy and 10.0 mGy, respectively. Corresponding to the phantom image evaluation for soft-tissue contrast and PFD analysis, 100%ASIR gave remarkable blurring and noise texture change, and 100%iNoir gave effective noise reduction while preserving organ edges and detailed structures of tissues.

4. Discussion

We compared CT image qualities of iNoir with FBP and ASIR using phantom tests corresponding to pediatric abdominal CT and a human observer test using clinical images. The edge-preserving noise reduction performance of iNoir evaluated by SPF was higher than that of ASIR for the soft-tissue contrast (50 HU). For the iodine contrast (300 HU), iNoir was superior to ASIR at the low spatial frequencies corresponding to the pediatric abdominal vessel diameters. The peak frequencies observed in the NPS results were diﬀerent, and the shifts from FBP in iNoir were negligible, in contrast with clearly determinable shifts in ASIR. The Fig. 4.NPS measurement results for FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR at (a) 8.4 and (b) 16.7 mGy.

Fig. 5.SPF²results for the soft-tissue-equivalent rod for FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR at (a) 8.4 and (b) 16.7 mGy.

(7)

results of the human observer test indicated that iNoir provided better image quality and smaller noise texture change than ASIR, supporting the results of the phantom study.

The object-edge preservation capability of ASIR depended on the

object contrast. This result agreed with the previous reports using hybrid-type IRs[8,9,15] For the soft-tissue contrast, TTF of ASIR was significantly lower than that of FBP, while the NPS was also lower than that of FBP. Thus, the SPF²improvement of ASIR was limited in the middle-to-high frequencies and, moreover, not remarkable. Thisfinding was because the TTF maintenance and NPS reduction were almost in a trade-offrelationship, which is similar to low-pass filtering without edge preservation. The result corresponded to the phantom and abdominal images inFigs. 8b, c, and10b in which object and organ edges were blurred with the decreased noise. Although TTF of iNoir was also lower than that of FBP, the degree was less than ASIR. Therefore, iNoir provided improved SPF²compared with FBP over the almost entire frequency range. In the phantom and abdominal images, edge blurring on object and organs was well suppressed, corresponding to the TTF results.

For the iodine contrast, TTF enhances at middle-to-high spatial frequencies were presented in the results of ASIR, while TTF of iNoir was equivalent to that of FBP. Consequently, the SPF²of ASIR exhibited Fig. 6.SPF²results for the 12-mgI iodine/ml rod for FBP, 50%iNoir, 100%iNoir, 50%ASIR, and 100%ASIR at (a) 8.4 and (b) 16.7 mGy.

Table 1

PFD in the normalized NPS considering a human visual response. The PFD of iNoir was smaller than that of ASIR.

PFD

8.4 mGy 16.7 mGy

50%iNoir 0.01 0.02

100%iNoir 0.02 0.03

50%ASIR 0.03 0.05

100%ASIR 0.09 0.10

PFD: peak frequency diﬀerence, iNoir: Image-space Noise Reduction, ASIR:

Adaptive Statistical Iterative Reconstruction.

Fig. 7.Boxplots show the human observer test results for (a) overall image quality and (b) noise texture change from FBP.

105

(8)

impractical curves, whereas, in general, the SPF²curves must show a monotonic decrease. This impractical feature of ASIR evaluated by SPF² may relate to the unrealistic images with unnaturally sharpened object edges and very smoothed backgrounds, as indicated inFig. 9. Similar impractical SPF²curve shapes were also previously reported for another hybrid-type IR, ADMIRE provided by Siemens[15], with an impression of unrealistic image. iNoir improved SPF more signiﬁcantly with the iodine contrast than with the soft-tissue contrast. This result corresponded to the phantom images presenting well-preserved object edge and naturally reduced noise (Fig. 9d and e).

In addition, the average SPF²of iNoir between 0.1 and 0.5 cycles/

mm increased in comparison with FBP. These trends were similar to the properties achieved by dose increase. Thus, iNoir would have potential to reduce dose or achieve edge-preserving noise reduction without dose

increase for pediatric abdominal CT which need special concern of radiation dose because of greater risk for cancer development.

As shown inFig. 4, the use of ASIR shifted the peak frequency of NPS to lower frequencies as the blending rates increased. The peak frequency shift was found in other hybrid-type IRs including new IR techniques such as ASIR-V [10,11,29,30]. By contrast, the peak frequency shifts of iNoir were negligible. The PFD of 50%ASIR became small (0.03) and was comparable to iNoir. This result corresponded to the observer test results of noise texture change that were not significantly different between 50%ASIR and 50%iNoir or between 50%ASIR and 100%iNoir. Although the influence of PFD on image di- agnosis has not been clinically demonstrated, we presume that iNoir would obtain great acceptance for observers because the spatial frequency distribution of noise is similar to that of FBP.

Fig. 8.Transverse soft-tissue-equivalent rod images of (a) FBP, (b) 50%ASIR, (c) 100%ASIR, (d) 50%iNoir, and (e) 100%iNoir at 8.4 mGy. The image noise is remarkably decreased by 100%iNoir without conspicuous blurring.

Fig. 9.Transverse 12-mg iodine/ml rod images of (a) FBP, (b) 50%ASIR, (c) 100%ASIR, (d) 50%iNoir, and (e) 100%iNoir at 8.4 mGy. Although the circular edges of 100%ASIR are sharper than those of other images, the natural image appearance is impaired.

(9)

As a result, for both the soft-tissue and iodine contrasts, iNoir’s showed notable noise reductions with SPF²curves that monotonically decreases with the increase of spatial frequency similarly to FBP's SPF curves. In addition, the noise texture change was mostly negligible.

Thus, our results indicated that iNoir could reduce CT image noise, exhibiting more natural image appearances than with ASIR.

TTF and NPS used in this study can offer quantitative measures with signal responses and noise amounts as a function of spatial frequency, respectively. Thus, the results tend to be understandable for comparing different systems or judging the image quality relating to clinical ben- efits. However, these methods need specific phantoms; thus, the measurement results are possible to be phantom-specific, even with the task-based schemes as applied in this study. On the other hand, the reference-free image quality assessment methods have been proposed for non-clinical images such as screen pictures, particulate matter (PM2.5) images, sonar images, virtual reality images, and camera images [31–35]. These methods reportedly can offer good quality predictions for assessing respective categorized images without reference images and specific phantoms. Thus, these methods may become alternations of TTF and NPS currently proposed for CT images.

Though the good accuracies of TTF and NPS for the quantitative measurements and the phantom accessibility in the clinical settings are presently favorable, the reference-free methods need to be investigated for the CT image quality measurements in future studies.

Our study had several limitations. ASIR was the only IR technique compared to iNoir. Since there are various commercially available IR techniques with diﬀerent properties[7–11,29,30,36], the performance advantage of iNoir in this study would vary depending on other IR techniques’performances. We only examined two SSDE levels (8.4 and 16.7 mGy) with one phantom size (17 cm). The average SSDE for clinical cases was 8.2 ± 2.4 mGy, which corresponded to the lower dose of 8.4 mGy in our phantom tests. Therefore, smaller phantom sizes and corresponding lower doses should be assessed for pediatric patients with smaller body sizes. We used only two contrast levels of 50 and 300 HU; thus, the spatial resolutions for other contrasts, such as 1000 HU for bones and−100 for fats, were not examined in this study. There- fore, we should investigate other contrasts in future work. Finally, we did not assess iNoir’s artifact reduction capability because the user's manual stated that iNoir was developed only to achieve noise

reduction. However, it was reported that the combined use of IR techniques and a metal artifact reduction technique improved image quality and strongly reduces metal artifacts[37]. Hence, the additional investigation for the combined use of iNoir and metal artifact reduction techniques would be required.

5. Conclusions

The noise reduction, without noise texture change and with eﬀec- tively preserved object edges, was achieved at contrasts of 50 and 300 HU by the image-based noise reduction algorithm, iNoir. The results suggested that iNoir would provide better image qualities for pediatric abdominal CT compared to ASIR, as a useful vendor-neutral tool.

Funding

This research did not receive any speciﬁc grant from funding agencies in the public, commercial, or not-for-proﬁt sectors.

Declaration of Competing Interest

The authors declare that they have no known competingﬁnancial interests or personal relationships that could have appeared to inﬂu- ence the work reported in this paper.

References

[1] Brenner DJ, Hall EJ. Computed tomography–an increasing source of radiation ex- posure. N Engl J Med 2007;357:2277–84.https://doi.org/10.1056/

NEJMra072149.

[2] Greess H, Lutze J, Nömayr A, Wolf H, Hothorn T, Kalender WA, et al. Dose reduction in subsecond multislice spiral CT examination of children by online tube current modulation. Eur Radiol 2004;14:995–9.https://doi.org/10.1007/s00330-004- 2301-9.

[3] Coursey C, Frush DP, Yoshizumi T, Toncheva G, Nguyen G, Greenberg SB. Pediatric chest MDCT using tube current modulation: eﬀect on radiation dose with breast shielding. Am J Roentgenol 2008;190:W54–61.https://doi.org/10.2214/AJR.07.

2017.

[4] Sohi B, Kim M-J, Yoon C-S, Kim DW, Hong JH, Lee M-J. Eﬀects of adaptive statistical iterative reconstruction on radiation dose reduction and diagnostic accuracy of pediatric abdominal CT. Pediatr Radiol 2014;44:1541–7.https://doi.org/10.1007/

s00247-014-3058-1.

Fig. 10.Abdominal images of (a) FBP, (b) 100%ASIR, and (c) 100%iNoir of a six-year-old girl. The image noise is decreased by 100%iNoir without conspicuous image blurring, whereas remarkable image blurring and noise texture changes are observed for 100%ASIR.

107

(10)

[5] Miéville FA, Gudinchet F, Rizzo E, Ou P, Brunelle F, Bochud FO, et al. Paediatric cardiac CT examinations : impact of the iterative reconstruction method ASIR on image quality–preliminaryﬁndings. Pediatr Radiol 2011;41:1154–64.https://doi.

org/10.1007/s00247-011-2146-8.

[6] Singh S, Kalra MK, Shenoy-Bhangle AS, Saini A, Gervais DA, Westra SJ, et al.

Radiation dose reduction with hybrid iterative reconstruction for pediatric CT.

Radiology 2012;263:537–46.https://doi.org/10.1148/radiol.12110268/-/DC1.

[7] Willemink MJ, de Jong PA, Leiner T, de Heer LM, Nievelstein RAJ, Budde RPJ, et al.

Iterative reconstruction techniques for computed tomography Part1: technical principles. Eur Radiol 2013;23:1623–31.https://doi.org/10.1007/s00330-012- 2765-y.

[8] Richard S, Husarik DB, Yadava G, Murphy SN, Samei E. Towards task-based assessment of CT performance: system and object MTF across diﬀerent reconstruction algorithms. Med Phys 2012;39:4115–22.https://doi.org/10.1118/1.4725171.

[9] Takata T, Ichikawa K, Mitsui W, Hayashi H, Minehiro K, Sakuta K, et al. Object shape dependency of in-plane resolution for iterative reconstruction of computed tomography. Phys Medica 2017;33:146–51.https://doi.org/10.1016/j.ejmp.2017.

01.001.

[10] Christianson O, Chen JJS, Yang Z, Saiprasad G, Dima A, Filliben JJ, et al. An improved index of image quality for task-based performance of CT iterative reconstruction across three commercial implementations. Radiology 2015;275:725–34.https://doi.org/10.1148/radiol.15132091.

[11] Chen B, Christianson O, Wilson JM, Samei E. Assessment of volumetric noise and resolution performance for linear and nonlinear CT reconstruction methods. Med Phys 2014;41:071909https://doi.org/10.1118/1.4881519.

[12] Boone JM, Strauss KJ, Cody DD, McCollough CH, McNitt-Gray MF, Toth TL. Size- speciﬁc dose estimates (SSDE) in pediatric and adult body CT examinations Report of AAPM Task Group 204. Am Assoc Phys Med 2011.

[13] Vassileva J, Rehani MM, Applegate K, Ahmed NA, Al-dhuhli H, Al-naemi HM. IAEA survey of paediatric computed tomography practice in 40 countries in Asia, Europe, Latin America and Africa : procedures and protocols. Eur Radiol 2013;23:623–31.

https://doi.org/10.1007/s00330-012-2639-3.

[14] Urikura A, Ichikawa K, Hara T, Nishimaru E, Nakaya Y. Spatial resolution measurement for iterative reconstruction by use of image-averaging techniques in computed tomography. Radiol Phys Technol 2014;7:358–66.https://doi.org/10.

1007/s12194-014-0273-2.

[15] Kawashima H, Ichikawa K, Matsubara K, Nagata H, Takata T, Kobayashi S. Quality evaluation of image-based iterative reconstruction for CT: Comparison with hybrid iterative reconstruction. J Appl Clin Med Phys 2019;20:199–205.https://doi.org/

10.1002/acm2.12597.

[16] Ichikawa K, Kawashima H, Shimada M, Adachi T, Takata T. A three-dimensional cross-directional bilateralﬁlter for edge-preserving noise reduction of low-dose computed tomography images. Comput Biol Med 2019;111:103353https://doi.org/

10.1016/j.compbiomed.2019.103353.

[17] Hanson KM. Detectability in computed tomographic images. Med Phys 1999;6:441–51.https://doi.org/10.1118/1.594534.

[18] Kijewski MF, Judy PF. The noise power spectrum of CT images. Phys Med Biol 1987;32:565–75.https://doi.org/10.1088/0031-9155/32/5/003.

[19] Boedeker KL, Cooper VN. Mcnitt-Gray MF. Application of the noise power spectrum in modern diagnostic MDCT: partⅠ. Measurement of noise power spectra and noise equivalent quanta. Phys Med Biol 2007;52:4027–46.https://doi.org/10.1088/

0031-9155/52/14/002.

[20] Samei E, Richard S. Assessment of the dose reduction potential of a model-based iterative reconstruction algorithm using a task-based performance metrology. Med Phys 2015;42:314–23.https://doi.org/10.1118/1.4903899.

[21] Boedeker KL, Cooper VN. Mcnitt-Gray MF. Application of the noise power spectrum

in modern diagnostic MDCT: partⅡ. Measurement of noise power spectra and noise equivalent quanta. Phys Med Biol 2007;52:4047–61.https://doi.org/10.1088/

0031-9155/52/14/003.

[22] Miura Y, Ichikawa K, Fujimura I, Hara T, Hoshino T. Comparative evaluation of image quality among diﬀerent detector conﬁgurations using area detector computed tomography. Radiol Phys Technol 2018;11:54–60.https://doi.org/10.1007/

s12194-017-0437-y.

[23] Akturk Y, Ozbal Gunes S. Normal abdominal aorta diameter in infants, children and adolescents. Pediatr Int 2018;60:455–60.https://doi.org/10.1111/ped.13542.

[24] Hashikura Y, Kawasaki S, Okumura N, Ishikawa S, Matsunami H, Ikegami T, et al.

Prevention of hepatic artery thrombosis in pediatric liver transplantation.

Transplantation 1995;60:1109–12.

[25] NickoloﬀEL. Measurement of the PSF for a CT scanner: appropriate wire diameter and pixel size. Phys Med Biol 1988;33:149–55.

[26] Urikura A, Hara T, Ichikawa K, Nishimaru E, Hoshino T, Yoshida T, et al. Objective assessment of low-contrast computed tomography images with iterative reconstruction. Phys Medica 2016;32:992–8.https://doi.org/10.1016/j.ejmp.2016.

07.003.

[27] Wang G, Vannier MW. Longitudinal resolution in volumetric x-ray computerized tomography—Analytical comparison between conventional and helical computerized tomography. Med Phys 1994;21:429–33.https://doi.org/10.1118/1.597306.

[28] Solomon JB, Christianson O, Samei E. Quantitative comparison of noise texture across CT scanners from diﬀerent manufacturers. Med Phys 2012;39:6048–55.

https://doi.org/10.1118/1.4752209.

[29] Euler A, Solomon J, Marin D, Nelson RC, Samei E. A third-generation adaptive statistical iterative reconstruction technique: phantom study of image noise, spatial resolution, lesion detectability, and dose reduction potential. Am J Roentgenol 2018;210:1301–8.

[30] Miéville FA, Gudinchet F, Brunelle F, Bochud FO, Verdun FR. Iterative reconstruction methods in two diﬀerent MDCT scanners: Physical metrics and 4-al- ternative forced-choice detectability experiments - A phantom approach. Phys Medica 2013;29:99–110.https://doi.org/10.1016/j.ejmp.2011.12.004.

[31] Gu K, Zhou J, Qiao J, Zhai G, Lin W, Bovik AC. No-reference quality assessment of screen content pictures. IEEE Trans Image Process 2017;26:4005–18.

[32] Gu K, Qiao J, Li X. Highly eﬃcient picture-based prediction of PM2.5 concentration.

IEEE Trans Ind Electron 2019;66:3176–84.https://doi.org/10.1109/TIE.2018.

2840515.

[33] Chen W, Gu K, Lin W, Xia Z, Le Callet P, Cheng E. Reference-free quality assessment of sonar images via contour degradation measurement. IEEE Trans Image Process 2019;28:5336–51.https://doi.org/10.1109/TIP.2019.2910666.

[34] Gu K, Jakhetiya V, Qiao JF, Li X, Lin W, Thalmann D. Model-based referenceless quality metric of 3D synthesized images using local image description. IEEE Trans Image Process 2018;27:394–405.https://doi.org/10.1109/TIP.2017.2733164.

[35] Liu Y, Gu K, Wang S, Zhao D, Gao W. Blind quality assessment of camera images based on low-level and high-level statistical features. IEEE Trans Multimed 2018;21:135–46.https://doi.org/10.1109/TMM.2018.2849602.

[36] Funama Y, Taguchi K, Utsunomiya D, Oda S, Katahira K, Tokuyasu S, et al. Image quality assessment of an iterative reconstruction algorithm applied to abdominal CT imaging. Phys Medica 2014;30:527–34.https://doi.org/10.1016/j.ejmp.2014.02.

005.

[37] Wellenberg RHH, Boomsma MF, Van Osch JAC, Vlassenbroek A, Miles J, Edens MA, et al. Computed tomography imaging of a hip prosthesis using iterative model- based reconstruction and orthopaedic metal artefact reduction : a quantitative analysis. J Comput Assist Tomogr 2016;40:971–8.https://doi.org/10.1097/RCT.

0000000000000449.