Dose reduction potential in dual-energy subtraction chest radiography based on the relationship between spatial-resolution property and segmentation accuracy of the tumor area


ACTA IMEKO 
ISSN: 2221-870X 
June 2022, Volume 11, Number 2, 1 - 8 

 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 1 

Dose reduction potential in dual-energy subtraction chest 
radiography based on the relationship between spatial-
resolution property and segmentation accuracy of the tumor 
area 

Shu Onodera1, Yongbum Lee2, Tomoyoshi Kawabata1 

1 Department of Radiology Division of Medical Technology, Tohoku University Hospital, Sendai, Miyagi, Japan 
2 Graduate School of Health Sciences, Niigata University, Niigata, Japan 

 

 

Section: RESEARCH PAPER  

Keywords: Chest radiography; X-ray; U-net; deep learning; flat panel detector 

Citation: Shu Onodera, Yongbum Lee, Tomoyoshi Kawabata, Dose reduction potential in dual-energy subtraction chest radiography based on the 
relationship between spatial-resolution property and segmentation accuracy of the tumor area, Acta IMEKO, vol. 11, no. 2, article 28, June 2022, identifier: 
IMEKO-ACTA-11 (2022)-02-28 

Section Editor: Francesco Lamonaca, University of Calabria, Italy 

Received September 30, 2021; In final form February 23, 2022; Published June 2022 

Copyright: This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 License, which permits unrestricted use, 
distribution, and reproduction in any medium, provided the original author and source are credited. 

Corresponding author: Shu Onodera, e-mail: onodera@rad.hosp.tohoku.ac.jp  

 

1. INTRODUCTION 

Chest radiography is the most basic diagnostic imaging 
procedure for lung diseases. However, the amount of X-rays that 
the patient is exposed to is enormous, including in the case of 
standing position imaging which is usually performed during 
medical examinations and bedside imaging for critically ill 
patients [1]. Compared with computed tomography (CT) 
examination, which provides three-dimensional information, the 
amount of exposure in radiography is very low (CT: 10 mSv, 
chest radiography: 0.1 mSv) [2]-[5] and its importance in terms 
of convenience of examination cannot be underestimated. 

Usually, a high voltage of approximately 120 kV is applied 
during chest radiography to emphasize the contrast of the lung 
field rather than that of the ribs [6]. Nevertheless, the shadow of 
the rib remains on the image, making it difficult to detect the 
shadow of the soft tissue that overlaps that of the ribs. To 
address this problem, it has been developed an energy 
subtraction process [7] in which two types of image data with 
different radiographic energy characteristics are obtained with a 
single exposure, the bone shadows are removed through weighed 
differentiation of the respective images, and the image of the soft 
tissue alone is segmented (hereinafter, referred to as soft tissue 
imaging). Once bone shadows are removed, it becomes easier to 
detect tumours in soft tissue images. However, because the 
image quality in chest radiography considerably varies for 

ABSTRACT 
We investigated the relationship between the spatial-resolution property of soft tissue images and the lesion detection ability using U-
net. We aimed to explore the possibility of dose reduction during energy subtraction chest radiography. The correlation between the 
spatial-resolution property of each dose image and the segmentation accuracy of the tumor area in the four regions where the tumor 
was placed was evaluated using linear regression analysis. The spatial-resolution property was determined by task-based evaluation, 
and the task-based modulation transfer function (TTF) was computed as its index. TTFs of the reference dose image and the 75 % dose 
image showed almost the same frequency characteristics regardless of the location of the tumor, and the Dice coefficient also high. 
When the tumor was located in the right supraclavicular region and under 50 % dose, the frequency characteristics were significantly 
reduced, and the Dice coefficient was also low. Our results showed a close relationship between the spatial-resolution property and the 
segmentation accuracy of tumor area using deep learning in dual-energy subtraction chest radiography. In conclusion, a dose reduction 
of approximately 25 % compared to the conventional method can be achieved. The limitations are the shape of the simulated mass and 
the use of chest phantom. 

mailto:onodera@rad.hosp.tohoku.ac.jp


 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 2 

different parts due to factors such as the amount of radiation 
reaching the detector and the amount of scatter radiation, 
detectability may differ depending on the location of the tumour. 
In particular, the scatter radiation generated from the clavicle and 
scapula is believed to significantly affect the upper lobe of the 
lung, which is a common site of adenocarcinoma [8]. The spatial-
resolution property of the image is also a very important factor 
in tumour detection [9]. The spatial-resolution property is a 
measure of the sharpness of an image and is an important 
characteristic that determines the detectability of lesions in X-ray 
images. Chest radiography images are generally subjected to 
several image-processing techniques to improve image quality, 
and these processing tools lead to a nonlinear behaviour that 
depends on image quality, which is different for different parts 
[10].  Thus, the quality of the soft tissue image also shows 
nonlinear behaviour, and task-based evaluation in a 
measurement environment that reflects clinical conditions is 
necessary to determine the spatial-resolution property. 

The use of computer-aided diagnosis (CAD) in diagnostic 
imaging has increased [11], [12]. Earlier, image interpretation was 
performed by radiologists, based on their cultivated experience. 
However, with the recent increase in the use of radiography and 
the consequent increase in the number of images, CAD was 
introduced to reduce the burden on radiologists. CAD based on 
deep learning has been attracting attention recently, and it may 
soon be possible to detect a tumour even in a low-dose image 
with high noise. 

In previous reports on energy subtraction-treated chest 
radiographs, visual evaluations of images acquired by the 
computed radiography (CR) systems have been reported [13]-
[17]. The purpose of this study was to investigate the relationship 
between the spatial-resolution property of soft tissue images 
obtained by the flat panel detector (FPD) system and the lesion 
detection ability based on deep learning and explore the 
possibility of dose reduction during energy subtraction chest 
radiography. 

2. MATERIALS AND METHODS 

2.1. Image acquisition  

An acrylic cylindrical simulated tumour with a diameter of 20 
mm and thickness of 3 mm was placed in four regions on the 
chest phantom (right supraclavicular, left middle lung, right 
lower lung, and mediastinum), Figure 1.  

Bone structure such as the clavicle, shoulder blade, and ribs 
as well as soft tissues as the mediastinum and pulmonary 
vascularity are located in the chest phantom. The single-exposure 
dual-energy subtraction system, [18], [19] CALNEO Dual 
(FUJIFILM Medical Co. Ltd. Tokyo. Japan, with a pixel size of 
0.15 mm), was used in this study. The FPD implemented in this 

system consisted of two stacked scintillators. Normal energy 
images were collected in the first layer (cesium iodide scintillator), 
and the second layer (gadolinium sulfide scintillator) collected 
high-energy images transmitted through the first layer. Table 1 
lists the imaging conditions used. The source-image distance was 
fixed at 180 cm, the field size was 43.2 cm, the image depth was 
12 bits, and the tube voltage was 115 kV. Three types of imaging 
doses were used: a standard dose of 1.6 mA s, which was then 
reduced by 25 % to 1.25 mA s and then reduced by 50 % to 
0.8 mA s, and 100 images of each type were acquired. 

2.2. Calculation of the spatial-resolution property 

Chest radiography images are generally subjected to several 
image-processing techniques to improve image quality. 
Frequency-processing and dynamic-range compression 
processing are typical examples. However, these processing tools 
lead to a nonlinear behaviour that depends on image quality, 
which is different for different parts. Therefore, in this study, the 
spatial-resolution property was determined by task-based 
evaluation, and the task-based modulation transfer function 
(TTF) was computed as its index [20]. The TTF calculation 
process is shown in Figure 2. The edge spread function (ESF) 
for the cylinder was obtained by averaging the profiles that cross 
the edge of the cylinder, measured from the centre in the 
direction of radiation. Next, TTF was calculated using the 
Fourier transform of the line spread function obtained by 
differentiating the ESF. One of the factors to be considered 
when determining the TTF of a soft tissue image is the signal-to-
noise ratio (SNR) of the image. Because images with a low SNR 
create large errors in the calculation results, in this study, the 
image without acrylic was subtracted from the image with acrylic, 
and an image with a high SNR was created through the additive 
average of 100 such images, which was then used to calculate the 
TTF (Figure 3). 

2.3. Building the deep learning environment 

CAD using deep learning has been an area of active research 
in recent years and has a wide range of applications in medical 

 

Figure 1. Phantom image and placement of acrylic cylinder. 

 

Figure 2. TTF calculation process. 

Table 1. Imaging conditions. 

Source 
image 

distance  
in cm 

Field  
size  

in cm 

Tube  
voltage  

in kV 

Image  
depth  
in bit 

Dose  
in mA s 

180 43.2 115 12 

1.6 (reference) 

1.2 (25 % down) 

0.8 (50 % down) 



 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 3 

imaging, such as lesion detection, area extraction, and image 
noise reduction. Image segmentation refers to the process of 
dividing an image into regions corresponding to each object. 
Because the target areas in medical imaging are organs or lesions, 
the positional information must be specified in the original input 

image at the time of segmentation. U-net [21] is a typical example 
of a deep convolutional neural network for image segmentation. 
The present study deals with the detection of lung tumours using 
U-net in soft tissue images. Figure 4 shows the structure of the 
U-net used in this study. The usage environment of U-net in this 
research is as follows: OS: Windows10, Framework: Python3.7, 
TensorFlow, Keras, CPU: Core i7-10750H, Memory: 16G. The 
ReLu function and sigmoid function were used as activation 
functions, cross entropy as the loss function, and Adam as the 
learning optimization algorithm. 

2.4. Data set for deep learning 

The acquired soft tissue image (window width: 8500, window 
level: 8100, 14 bits) was segmented into 128 × 128 pixels centred 
around the tumour and converted to png format (window width: 
255, window level: 128, 8bits). Fifty standard-dose images were 
input in U-net as training data and 50 reduced-dose images as 
evaluation data, and learning was conducted by setting the 
number of epochs to 30. The teaching data for the soft tissue 
images containing the tumour were created by binarizing the 
image into the tumour area and other areas (Figure 5). 
 

 

Figure 3. Creating the TTF calculation image. 

 

Figure 4. The structure of the U-net in this study. 



 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 4 

2.5. Evaluation of the segmented tumour area 

The Dice coefficient [22] was calculated as the degree of 
similarity between the output image and the teacher image to 
evaluate the extraction accuracy of the tumour region using U-
net. The Dice coefficient is defined by the following formula: 

𝐷𝑖𝑐𝑒(𝐴, 𝐵) =
2|𝐴 ∩ 𝐵|

|𝐴| + |𝐵|
 . (1) 

Here, A denotes the tumour region in the teaching image 
(region with a digital value of 255), and B denotes the tumour 
region in the output image (region with a digital value of 255). 

2.6. Evaluation of the correlation between spatial-resolution 
property and segmentation accuracy of tumor area 

In this study, the correlation between the spatial-resolution 
property of each dose image and the segmentation accuracy of 
the tumour area in the four regions where the tumour was placed 
was evaluated using linear regression analysis. A scatter plot was 
created by treating the TTF and Dice coefficient of each dose 
image from 0.2 to 1.2 cycle/mm (intervals of 0.2 cycle/mm) as 
variables. The Dice coefficient for the reference dose image was 
set to 1. 

3. RESULTS 

Figure 6 shows the TTF results for each condition. No 
difference was observed between the TTFs of the reference dose 
image and the 75 % dose image in the supraclavicular region, 
where the contrast was low due to the influence of scattered 

radiation from the thoracic spine, clavicle, and scapula; however, 
the similarity decreased significantly in the TTF of 50 % dose 
images. In contrast, in the middle and lower lung regions where 
the effect of scattered radiation was small and the contrast was 
high compared to the supraclavicular region, the TTFs were 
generally high and the difference in values between doses was 
small. In the mediastinum, TTFs were low as in the 
supraclavicular region because of the low contrast due to the 
scattered radiation from the heart and sternum, but the decrease 
was not as high as that in the supraclavicular region; in 
comparison, the TTF in the supraclavicular region was the lowest 
among all the other regions. 

Table 2 shows the average values of the Dice coefficients of 
the 50 datasets for each condition. The Dice coefficient between 
the segmented tumour area and the teaching data in the 75 % 
dose image showed a generally high value of approximately 0.96, 
regardless of the location of the tumour. Furthermore, the TTF 
of the 75 % dose image showed a value similar to that of the 
reference dose image regardless of the location of the tumour. In 
contrast, the Dice coefficient in the 50 % dose image was as low 
as 0.937 when the tumour was located in the supraclavicular 
region. Likewise, the TTF of the 50 % dose image in which the 
tumour was located in the supraclavicular region showed a lower 
value compared to the reference dose image. 

Figure 7 shows the actual tumour area segmented by U-net. 
In the 75 % dose condition, the segmented images were highly 
similar to the teaching data, regardless of the location of the 
tumour. However, in the 50 % dose condition and when the 
tumour was located in the right supraclavicular region, the 
segmented region was slightly larger compared with the teaching 
data. Figure 8 to Figure 11 show the correlation between TTF 
and Dice coefficient in soft tissue images. A positive correlation 
was observed between the TTF and Dice coefficient of every 
dose image at all spatial frequencies in the right supraclavicular 
and the right lower lung regions and between the frequencies of 
0.2 to 0.8 cycle / mm in the mediastinum section. In contrast, no 
correlation was observed between the TTF and Dice coefficients 
in any of the spatial frequencies in the left middle lung region. 

4. DISCUSSION 

In the case of the simulated tumour located in the right 
supraclavicular region and under 50 % dose, both the TTF and 
Dice coefficients showed significantly low values. One reason for 
this could be that the contrast of the tumour was reduced by 
scattered radiation mainly from the clavicle and scapula due to 
the complicated bone structure of the supraclavicular region. A 
second reason could be the fact that the tumour area could not 
be segmented accurately because of the increased image noise 
because the amount of radiation reaching the detector was 
smaller than that reaching other parts. 

In the case in which the simulated tumour was located in the 
mediastinum region, the value of TTF was not very different 
from that when the tumour was in the middle and lower lung 
regions, and the Dice coefficient also showed a similar value. In 
the mediastinum region, the amount of radiation reaching the 
detector was less than that of the middle and lower lung regions, 

 

Figure 5. Creation of teaching data. 

 

Figure 6. TTFs for each condition. 

Table 2. Dice coefficients for each of the conditions. 

mA s 
Right 

supraclavicular 
Left middle 

lung 
Right lower 

lung 
Mediastinum 

1.25 0.960 0.960 0.959 0.971 

0.8 0.937 0.969 0.963 0.967 



 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 5 

and the amount of scattered radiation from the sternum and 
heart was also large. Therefore, under the 50 % dose condition, 
the TTF and Dice coefficients were perceived to be as low as the 
right supraclavicular region. However, as the tumour in this area 
had fewer pulmonary blood vessels around it than other sites, the 

structure was relatively simple and the tumour area could be 
segmented accurately (Figure 12). 

The results of this study show that there is a high degree of 
correlation between the spatial resolution of the soft tissue image 
and the segmentation accuracy of the tumour area using deep 

 

Figure 7. Mass region segmentation using for each of the conditions. 

 

Figure 8. Between TTF and dice coefficient (right supraclavicular). 

 

Figure 9. Between TTF and dice coefficient (left middle lung). 

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference



 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 6 

learning in the supraclavicular, lower lung, and mediastinum 
regions. In the 75 % dose images, the TTF was high regardless 
of the tumour location, and the Dice coefficient was also high. 
In contrast, in the 50 % dose images, when the tumour was 
present in the supraclavicular region, the TTF was significantly 
reduced, and the Dice coefficient was also low. In other words, 
if the radiation dose is reduced to 50 % of the conventional 
radiation condition, tumour that develop in the supraclavicular 
region may not be segmented accurately due to a decrease in 
TTF. No correlation was confirmed between TTF and the Dice 
coefficient in the middle lung area. This could be because there 
was no difference in TTFs of the dose images between 0.2 and 
0.6 cycle / mm, and between the 0.8 and 1.2 cycle / mm, there 

was no difference in the TTFs of the 75 % dose image and the 
reference dose image. Among the four sites examined in this 
study, the highest amount of radiation reached the detector from 
the middle lung area, and the amount of scattered radiation from 
the surroundings was also small. Therefore, no correlation could 
be confirmed between the TTF and the Dice coefficient in the 
middle lung region, and the Dice coefficients of all dose images 
showed a high value of approximately 0.96. The discussion above 
suggests that, with single-exposure dual-energy subtraction chest 
radiography by the FPD system, it may be possible to reduce the 
dose by approximately 25 % compared to the conventional 
method. 

Figure 13 shows the detection quantum efficiency (DQE) in 
the radiation qualities of the RQA9 of the CR system, which was 
manufactured by the same company as the CALNEO Dual 
system used in this study [23], [24]. Because lung tumors are the 
targets of this study, we focused on the value of the spatial 
frequency of 1 cycle / mm [25]. The DQE value at 1 cycle / mm 
is approximately 0.5 for the CALNEO Dual and about 0.2 for 
the CR system, respectively, and the detection quantum 
efficiency of the CALNEO Dual is approximately 2.5 times 
higher. A system with an excellent DQE has a high degree of 
freedom in adjusting the balance between sharpness and 
graininess through image processing [26]. Therefore, selecting 

 

Figure 10. Between TTF and dice coefficient (right lower lung). 

 

Figure 11. Between TTF and dice coefficient (mediastinum). 

 

Figure 12. Pulmonary vessels around the mass in the mediastinum. 

 

Figure 13. DQE for CALNEO dual and CR system with RQA9 spectra. 

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference

● 50% dose

▲ 75% dose

■ reference



 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 7 

parameters with good spatial-resolution properties for multi-
frequency processing in single-exposure dual-energy subtraction 
chest radiography using FPD could lead to further dose 
reduction. 

As a limitation of this study, we would first mention the 
structure of the simulated tumors. In this study, an acrylic 
material with a simple cylindrical structure was used as the 
simulated tumor. Actual lesions with increased malignancy, such 
as spiculated lesions, often have more complex structures, and in 
such cases, the results may differ. Moreover, in this study, all 
measurements from beginning to end were performed using a 
phantom, and the effects of heartbeat, which is a major problem 
in actual clinical practice, [27] have not been considered. 
However, to reduce the effects of heartbeat, the imaging time 
was shortened to the largest extent possible, and measurements 
were performed in a very short interval of approximately 10 ms, 
as is done in clinical practice; hence, we hope that the results will 
not be greatly affected. 

5. CONCLUSIONS 

In this study, we clarified the relationship between the spatial 
resolution of single-exposure dual-energy subtraction chest 
radiography using the FPD system and the segmentation 
accuracy of the tumour area using deep learning. The TTFs of 
the reference dose image and the 75 % dose image showed 
almost the same frequency characteristics regardless of the 
location of the tumour, and the Dice coefficient also showed a 
high value. When the tumour was located in the right 
supraclavicular region and under 50 % dose, the frequency 
characteristics were significantly reduced, and the Dice 
coefficient was also low. Therefore, a close relationship between 
the spatial-resolution property and the segmentation accuracy of 
the tumour area was confirmed using deep learning in single-
exposure dual-energy subtraction chest radiography using the 
FPD system, and it may be possible to achieve dose reduction of 
approximately 25 % compared to the conventional method. 

REFERENCES 

[1] UNSCEAR. Medical Radical Exposures. Sources and Effects of 
Ionizing Radiation, UNSCER 2008 Report. New York: United 
Nations; 2010. Annex A.  

[2] L. J. M. Kroft, L. van der Velden, I. H. Girón, J. J. H. Roelofs, A. 
de Roos, J. Geleijns, Added value of ultra-low-dose computed 
tomography, dose equivalent to chest X-ray radiography, for 
diagnosing chest pathology, J. Thorac Imaging, 34 (2019), pp. 179-
186. 
DOI: 10.1097/RTI.0000000000000404  

[3] R. Ward, W. D Carrol, P. Cunningham, S. A Ho, M. Jones, W. 
Lenney, D. Thonpson, F. J Gilchrist, Radiation dose from 
common radiological investigations and cumulative exposure in 
children with cystic fibrosis: An observational study from a single 
UK centre, Observational Study, 7 (2017), pp. 1-5.  
DOI: 10.1136/bmjopen-2017-017548  

[4] S. Singh, M. K. Kalra, R. D. Ali Khawaja, A. Padle, S. Pourjabbar, 
D. Lira, J. A. O. Shepard, S. R. Digumarthy, Radiation dose 
optimization and thoracic computed tomography, Radiol. Clin. 
North Am., 52(2014), pp. 1-15. 
DOI: 10.1016/j.rcl.2013.08.004  

[5] International Commission on Radiological Protection, The 2007 
Recommendations of the International Commission on 
Radiological Protection, ICRP Publication 103, Ann. ICRP 37 (2-
4). 

[6] O. W. Hamer, C. B. Sirlin, M. Strotzer, I. Borisch, N. Zorger, S. 
Feuerbach, M. Volk, Chest radiography with a flat-panel detector: 

Image quality with dose reduction after copper filtration. 
Radiology, 237 (2005), pp- 691-700. 
DOI: 10.1148/radiol.2372041738  

[7] M. Fukao, K. Kawamoto, H. Matsuzawa, O. Honda, T. Iwaki, T. 
Doi, Optimization of dual-energy subtraction chest radiography 
by use of a direct-conversion flat-panel detector system, Radiol 
Phys Technol, 8 (2015), pp. 46-52. 
DOI: 10.1007/s12194-014-0285-y  

[8] K. Honda, Y. Matsui, H. Imai, Regional distribution of lung 
cancer, Haigan, 23 (1983), pp. 11-21. 

[9] Y. Fujimura, H. Nishiyama, T. Masumoto, S. Kono, Y. Kitagawa, 
T. Ikeda, T. Furukawa, T. Ishida, Investigation of reduction of 
exposure dose in digital mammography: Relationship between 
exposure dose and image processing, JpnSoc, Nihon Hoshasen 
Gijutsu Gakkai Zasshi, 64(2008), pp. 259-267. 
DOI: 10.6009/jjrt.64.259  

[10] K. Kishimoto, E. Ariga, R. Ishigaki, M. Imai, K. Kawamoto, K. 
Kobayashi, M. Sawada, K. Noto, M. Nakamae, R. Higashide, 
Study of appropriate dosing in consideration of image quality and 
patient dose on the digital radiography, JpnSoc, Nihon Hoshasen 
Gijutsu Gakkai Zasshi, 67(2011), 1381-1397. 
DOI: 10.6009/jjrt.67.1381  

[11] K. Doi, Current status and future potential of computer-aided 
diagnosis in medical imaging, Br J Radiol, 78 (2005), Spec No 1, 
S3-S19. 
DOI: 10.1259/bjr/82933343  

[12] H. Fujita, Present status of mammography CAD system, Med 
Imaging Technol, 1 (2003), 27-33. 

[13] S. Kido, J. Ikezoe, H. Naito, J. Arisawa, S. Tamura, T. Kozuka, W. 
Ito, K. Shimura, H. Kato, Clinical evaluation of pulmonary 
nodules with single-exposure dual-energy subtraction chest 
radiography with an iterative noise-reduction algorithm, 
Radiology, 194 (1995), pp. 407-412. 
DOI: 10.1148/radiology.194.2.7824718  

[14] S. Kido, K. Kuriyama, N. Hosomi, E. Inoue, C. Kuroda, T. Horai, 
Low-cost soft-copy display accuracy in the detection of pulmonary 
nodules by single-exposure dual-energy subtraction: comparison 
with hard-copy viewing, J Digit Imaging, 2000, pp. 33-37. 
DOI: 10.1007/BF03168338  

[15] J. R. Wilkie, M. L. Giger, M. R. Chinander, T. J. Vokes, R. M. 
Nishikawa, M. D. Carlin, Investigation of physical image quality 
indices of a bone densitometry system, Med Phys, 31 (2004), pp. 
873-881. 
DOI: 10.1118/1.1650528  

[16] S. Kido, H. Nakamura, W. Ito, K. Shimura, H. Kato, 
Computerized detection of pulmonary nodules by single-exposure 
dual-energy computed radiography of the chest (Part 1), Eur J 
Radiol, 44(2002), 198-204. 
DOI: 10.1016/s0720-048x(02)00268-1  

[17] S. Kido, K. Kuriyama, C. Kuroda, H. Nakamura, W. Ito, K. 
Shimura, H. Kato, Detection of simulated pulmonary nodules by 
single- exposure dual-energy computed radiography of the chest: 
Effect of a computer-aided diagnosis system (Part 2), Eur J Radiol, 
44 (2002), pp. 205-209.  
DOI: 10.1016/s0720-048x(02)00269-3  

[18] L. Shi, M. Lu, N. R. Bennett, E. Shapiro, J. Zhang, R. Colbeth, J. 
S. Lack, A. S. Wang, Characterization and potential applications of 
a dual-layer flat-panel detector, Med Phys, 47(2020), Epub 2020 
May 18, pp. 3332-3343. 
DOI: 10.1002/mp.14211  

[19] M. Lu, A. Wang, E. Shapiro, A. Shiroma, J. Zhang, J. Steiger, J. S. 
Lack, Dual energy imaging with a dual-layer flat panel detector, 
SPIE Med Imaging;10948 Physics of Medical Imaging, Sandiego, 
United States, 2019  
DOI: 10.1117/12.2513499  

[20] S. Richard, D. B. Husarik, G. Yadava, S. N. Murphy, E. Samei, 
Towards task-based assessment of CT performance: System and 
object MTF across different reconstruction algorithms, Med Phys, 
39(2012), 4115-4122. 
DOI: 10.1118/1.4725171  

https://doi.org/10.1097/RTI.0000000000000404
https://doi.org/10.1136/bmjopen-2017-017548
https://doi.org/10.1016/j.rcl.2013.08.004
https://doi.org/10.1148/radiol.2372041738
https://doi.org/10.1007/s12194-014-0285-y
https://doi.org/10.6009/jjrt.64.259
https://doi.org/10.6009/jjrt.67.1381
https://doi.org/10.1259/bjr/82933343
https://doi.org/10.1148/radiology.194.2.7824718
https://doi.org/10.1007/BF03168338
https://doi.org/10.1118/1.1650528
https://doi.org/10.1016/s0720-048x(02)00268-1
https://doi.org/10.1016/s0720-048x(02)00269-3
https://doi.org/10.1002/mp.14211
https://doi.org/10.1117/12.2513499
https://doi.org/10.1118/1.4725171


 

ACTA IMEKO | www.imeko.org June 2022 | Volume 11 | Number 2 | 8 

[21] O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional 
networks for biomedical image segmentation Lecture Notes in 
Computer Sciences (including SubserLect Notes ArtifIntellLect 
Notes Bioinformatics). 2015; 9351: pp. 234-241. 

[22] B. Sahiner, A. Pezeshk, L. M. Hadjiiski, X. Wang, K. Drukker, K. 
H. Cha, R. M. Summers, M. L. Giger, Deep learning in medical 
imaging and radiation therapy. Med Phys, 46(2019), Epub 2018 
Nov 20, e1-e36. 
DOI: 10.1002/mp.13264  

[23] IEC. 62220-1. Medical Electrical Equipment-Characteristics of 
Digital X-Ray Imaging Devices part 1: Determination of Detective 
Quantum Efficiency. International Electrotechnical Commission; 
2003. 

[24] IEC. 62220-1-2. Medical Electrical Equipment-Characteristics of 
Digital X-Ray Imaging Devices part 1-2: Determination of 
Detective Quantum Efficiency-Detectors Used in Mammography. 
International Electrotechnical Commission; 2007. 

[25] T. Yokoi, T. Takata, K. Ichikawa, Investigation of image quality 
identification utilizing physical image quality measurement in 
direct- and indirect-type of flat panel detectors and computed 
radiography, Nihon Hoshasen Gijutsu Gakkai Zasshi, 67(2011), 
pp. 1415-1425. 
DOI: 10.6009/jjrt.67.1415  

[26] A. R. Cowen, A. G. Davies, M. U. Sivananthan, The design and 
imaging characteristics of dynamic, solid-state, flat-panel x-ray 
image detectors for digital fluoroscopy and fluorography, Clin 
Radiol, 63(2008), 1073-1085. 
DOI: 10.1016/j.crad.2008.06.002  

[27] European Commission, CHEST. (LUNGS AND HEART) PA 
and Lateral Projections. European Guidelines on Quality Criteria 
for Diagnostic Radiographic Image. EUR. Luxembourg: CEC; 
1996:12:16260 EN. 

 

 

https://doi.org/10.1002/mp.13264
https://doi.org/10.6009/jjrt.67.1415
https://doi.org/10.1016/j.crad.2008.06.002