Open Access
Add to BibSonomy
Abstract
High-dynamic-range (HDR) X-ray CT imaging is effective in detecting some complex structures. For previous low-dynamic-range (LDR) imaging detectors, multi-energy LDR image sequence fusion can extend the dynamic range, but the efficiency is decreased. However, with the application of HDR imaging devices, traditional fixed-energy X-ray imaging can cause incongruity within energy, dynamic range, and the equivalent thickness of the workpiece at different projection angles. Then, the projection has a blurred edge, and the CT image quality is poor because of scattering and the inadequate dose. In this paper, a new HDR X-ray CT imaging method with energy self-adaptation between scanning angles for HDR imaging devices is studied. Low-energy prescanning is used to determine the initial scanning energy and obtain the edge contour information with an attenuating effect on scattering. By establishing a mathematical model between the gray level of the projection and the transmission voltage, the transmission energy at each angle is adjusted adaptively. Then, the prescanning and energy self-adaption scanning projections are fused to obtain the complete projection of the complex workpiece. Finally, a conventional reconstruction algorithm is used to reconstruct the HDR CT image. The experimental results show that the proposed imaging method can achieve HDR CT imaging of complex structures with high reconstruction quality, clear edge details, and high completeness.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
In X-ray CT imaging, the selection of X-ray tube voltage is very important, as it determines the ray energy. For a certain thickness of workpiece, if the ray energy is too high, there will be overexposure because of the lower dynamic range and ray scattering. Moreover, if the ray energy is too low, there will be low contrast because of the inadequate dose. Therefore, for a complicated workpiece, the construct is complex, and the equivalent thickness of X-ray attenuation will be different at every projection angle. However, common fixed-energy CT imaging has the problem of mismatch among the optimal exposure energy, effective thickness, and dynamic range of the imaging system. This problem readily causes artifacts, which make the image quality poor. Then, the mismatch will affect the defect detection and construct analysis [1]. Only if the ray energy at every angle is the most appropriate can a projection image with a clear internal structure and full edge information be obtained [2,3]. Thus, choosing the optimal energy based on only the operator’s personal experience is unreliable, especially in CT scanning.
Impeded by the effectiveness of low-dynamic-range (LDR) imaging detectors, whose bit depth is 8-bit or 12bit, a considerable amount of effective projection information is usually lost because of overexposure and underexposure. Based on LDR projection sequences, HDR fusion can capture more complete reconstruction information at every projection angle [4,5]. For the variable energy image sequences at every CT rotation angle, Liu et al. succeeded in obtaining the complete internal structural information about complicated structural components by using multi-energy image sequence fusion [6], Li et al. calculated the weighted fusion coefficient to capture image sequences based on a linear constraint with variable energy, confirming that HDR X-ray imaging with energy self-adaption is possible for complex workpieces [7]. This fusion method mainly solves the problem of HDR imaging for LDR detectors. However, this solution necessitates multiple exposures at every angle. As such, for a complete CT scan, the number of exposures is very large. The imaging efficiency is also very low, and this factor influences the lifetime of the X-ray source.
With the improvement of the imaging devices, HDR detectors become universal, whose bit depth is 14-bit or 16-bit. For an HDR detector, in general, all construct information about the workpiece can be obtained at once. Therefore, HDR fusion of LDR image sequences has no advantage in terms of imaging efficiency. However, for a complicated workpiece, the HDR detector still cannot overcome the problem caused by the different equivalent thicknesses at every projection angle. The construct of the workpiece is more complex, and the effects of scattering and the inadequate dose are more obvious. Thus, complete edge information usually cannot be obtained.
Complete edge information is also essential for improving the effect of reconstruction, but most of the information is lost due to scattering and the inadequate dose. Along with the study of the multi-spectrum imaging, distinguishing high- and low-energy areas in multi-material CT is also a good way [8,9]. However, this method focuses on the distinction between different materials. For the complicated object, multi-spectrum method is not effective because the edge information is lost during the projection acquisition stage [10]. Therefore, researchers have tried to use the multimodal information to compensate, providing a priori information for CT reconstruction by the optical scanning and ultrasonic scanning [11]. Despite the advantage of a priori reconstruction, there is still the problem of matching the prior images and the CT images. Moreover, the validity of this method cannot be verified when detecting more complicated object. Additionally, the use of low-energy rays instead of visible light is a good way to achieve a dose that can determine the outline of the workpiece. In the work of Xiaogang Yang and Mirone, a low-energy projection image with full edge information was used in network training, which obtained a perfect projection with complete information [12,13].
This paper proposes a new imaging method with variable energy between the scanning angles. The method includes self-adaptive variable-energy (SAVE) projection acquisition and image fusion stages. During the acquisition stage, the CT projection is collected by low-energy imaging with the aim of minimizing the loss of edge information and confirms the initial ray energy. Then, SAVE acquisition is used to adapt to the different equivalent thicknesses at every projection angle. Finally, the projection from SAVE acquisition and the low-energy projection are fused to obtain a higher quality projection and an HDR CT image.
The rest of the paper is organized as follows: the first section addresses the experimental setup in which the dynamic range problem arises. The following sections describe the concepts used in the proposed method in the context of the proposed experiment. Next, an experiment is used to verify the proposed method. The final section is a discussion and conclusion.
2. Experimental setup
2.1 Conditions and difficulty
This experiment will verify the disadvantage of fixed-energy CT imaging. Fixed-energy X-ray imaging leads to the appearance of blurry edges and affects the reconstructed CT image quality. This approach also produces noise and artifacts, which reduce the contrast ratio of the reconstruction [14]. Figure 1 is the actual titanium metal workpiece used in the experiment, and the maximum thickness ratio is 25 (mm)/1 (mm) = 25.
In the traditional fixed-energy CT scan imaging process, the X-ray voltage and current are not changed, but for complex structural parts, the effective thickness is constantly changing during scanning. Because of the interaction between the X-rays and complex structural parts, in this case, high-quality imaging is often impossible [15]. The low energy and large effective thickness for some projection angles result in an invalid area in the projection because of scattering and the inadequate dose.
Experimental Verification: Projection acquisition is carried out at every projection angle using an experimental CT system consisting of an X-ray source (COMET MXR-451HP/11 450KV) and a detector (PerkinElmer XRD 1621 AN14 ES). The bit depth of the detector is 16-bit. The distance between the source and detector is 1400 mm. The distance between the source and rotation center is 1030 mm. The tube voltage and current are 200 kV and 1.5 mA, respectively, and the scanning mode is circular trajectory scanning. The schematic diagram of the CT imaging system is shown in Fig. 2. From Fig. 2, the placements of the X-ray source, detector, and workpiece are described clearly. Also the CT rotation is performed by the rotating mechanism with the same direction. Get the projection at the fixed rotation interval, by adjusting the X-ray source’s tube voltage and current. Partial projections are shown in Fig. 3. Additionally, for each projection angle, statistical analysis is performed to obtain the trend of the minimum gray value about the projection, which is shown in Fig. 4.
As shown in Fig. 3, a clear edge cannot be obtained at most scan angles, and the projection information about the thin part is lost. Additionally, as shown in Fig. 4, the gray level at each angle varies greatly, with a minimum of 13509 and a maximum of 21035, and the proportion of the gap in the total range rises to 11.48%. This is because of the different effective thicknesses of the complicated workpiece at every projection angle. If the imaging ray energy does not match the effective thickness, the projection quality will be lower. Then, a higher quality CT image of a complicated object cannot be obtained [16]. Therefore, it is necessary to use variable-energy acquisition.
2.2 Adaptive acquisition
For adaptive acquisition, the two problems of improving the projection quality and compensating for the missing projection edge should be solved. Here, low-energy prescanning is used to compensate for the loss of edge information, and adaptive-energy prediction is used to improve the quality.
2.2.1 Low-energy prescanning
Based on the characteristics of X-rays, when the X-ray energy is less than 1 MeV, the X-ray interactions with matter are mainly Compton scattering. Additionally, when the energy continues to decrease, scattering is not obvious [17,18]. Therefore, low-energy prescanning is performed to ensure the integrity of the edge information. In the prescanning, the edge location needs to be defined. At the same time, the part projection of the workpiece’s thinner regions should exist, which can be used to restore the edge of high-energy imaging. Therefore, it is necessary to combine with the object to determine the prescanning energy by the priori of X-ray attenuation and experiment. In order to make the prescanning energy suitable for all the scanning angles, the one scanning angle with the minimum effective thickness of workpiece only be considered.
Because the gray level can reflect the effective thickness of the workpiece, prescanning can also determine the projection angle with the maximum effective thickness for the full projection angle. Then, the estimation of maximum energy can be performed at all CT rotation angles. For adaptive-energy prediction, the initial ray energy should be determined. Therefore, based on the estimated maximum energy, the upper limit of the energy is set in the process of the adaptive-energy scanning process. This procedure can improve the efficiency of the adaptive adjustment. For full-angle projections, the minimum gray level of the imaging area of the workpiece is analyzed in all CT projections.
2.2.2 Estimation of the maximum voltage
For the starting angle, the maximum ray energy needs to be set. In Eq. (1), the minimum gray level Hmin corresponds to the maximum effective thickness of the workpiece and the maximum imaging energy. At the starting angle, variable-energy imaging is necessary to determine the maximum imaging energy. Because adjusting the tube voltage is more efficient than adjusting the tube current in industrial X-ray imaging, the voltage is gradually increased to obtain the projection sequences, the current is kept constant.
Take an example of one scanning angle, with increasing ray energy, the edge gradually vanishes, which are shown in Fig. 5. According to the marked projection edge, compared to the edge at 160 kV, the edge at 240 kV is reduced by 107 pixels. Thus, according to Fig. 5, the edge projection information can be obtained with the lower ray energy, and the internal structure gradually becomes clearer with increasing voltage. Therefore, in the projection sequences with different voltages, the highest quality projection, corresponding to the maximum voltage, can be determined. The highest projection quality of the internal structure is evaluated and selected by a gray level distribution histogram, and the imaging energy condition is set to the maximum voltage of the adaptive scan.
The gray level histogram results show that when the voltage is low, the distributions in Figs. 6(a) and 6(b) show many pixels lower than 10,000, which indicates that the current voltage is not enough to image the internal construct of the workpiece. With increasing voltage, the distributions in Figs. 6(c) and 6(d) mostly gather in the range of 20,000-40,000, which is empirically the best gray level range for a 16-bit detector [6]. In fact, the internal structure of the image is complete for the projections at 220kV and 240kV, as shown in Fig. 5. Because of the extensive edge scattering and the loss of the edge structure, it is not appropriate to continue to increase the voltage. Accordingly, 220 kV is chosen as the maximum voltage and can be set as the initial voltage of the adaptive-energy acquisition process. Additionally, this gray level range of 20000-40000 is defined as the reference gray level range.
2.2.3 Self-adaptive variable-energy (SAVE) scanning
After determining the starting angle, maximum voltage and reference gray level range, auto adjustment of the voltage is important to improve the imaging efficiency at the next projection angle. The tube voltage V (kV) is determined based on the empirical Eq. for X-ray intensity I0. Specifically, I0 is proportional to the square of the voltage V [19].
where i is the tube current, which does not change; Z is the atomic number of the target material; and $K \approx (1.1\sim 1.4) \times {10^{ - 6}}$ is a ratio coefficient. Based on the X-ray attenuation character, the intensity I passing through the workpiece is expressed as Eq. (3). In Eq. (3), S(E) is the coefficient of the ray energy spectrum, d is the effective thickness of the workpiece, and μ(E) represents the attenuation coefficient. Let Based on Eqs. (2)–(4), the imaging intensity I can be revised as follows: Changing to the next angle, at the same imaging voltage and current, the imaging intensity I is different because the thickness d is different. Therefore, the gray level of the next angle will deviate from the reference gray level range. Because the difference in the effective thickness is less between two adjacent angles, the deviation of α is less, and there is approximately no change. Therefore, the tube voltage needs to be adjusted to reach the reference gray level range. For the numerical calculation, this reference gray level range is replaced by the reference average gray level I1, which is the average gray level of projection at the maximum voltage V1. During CT scanning, the current angle is relative to the previous angle. Thus,3. Projection fusion of SAVE scanning and prescanning
As shown in Fig. 5, the internal structure in the projection gradually becomes clear with the increasing of voltage, but the edge is lost. The projections acquired with SAVE scanning and prescanning could be fused at every scanning angle to obtain the complete information.
By the equivalent energy method, Eq. (3) can be changed to the ideal X-ray attenuation model [20]. Two imaging energies, E1 and E2 (E1<E2), are set; the initial ray intensities are I01 and I02, and the attenuated ray intensities are I1 and I2, respectively. Namely,
- (1) Effective area fusion: For prescanning, the effective areas in the optimal imaging gray level range are all fused to SAVE scanning.
- (2) Edge area fusion: Because the ray energy of prescanning is lower than that of SAVE scanning, for the workpiece’s thicker area, the projection quality is poor because of the inadequate dose. To improve fusion quality, the fusion should consist of only the lost edge information from SAVE scanning, which has higher quality than that of prescanning, and the other area is not changed. Namely, the fusion area is limited to the lost edge area.
4. Experimental results and discussion
4.1 Experimental parameters
The experimental system is a CT system consisting of an X-ray source (COMET MXR-451HP/11 450KV) and a detector (PerkinElmer XRD 1621 AN14 ES). The distance between the source and detector is 1400 mm. The distance between the source and rotation center is 1030 mm. Projection collection of the actual complex workpiece (Fig. 1) is carried out in two steps. According to the structure, size and material of the workpiece (Fig. 1), by X-ray imaging experiment, the prescanning energy conditions of 160 kV and 1.5 mA are determined respectively. The other imaging conditions are consistent with those used in fixed-energy scanning. Then, the SAVE scanning is performed. The current is also set to 1.5 mA. The voltage is controlled by the SAVE method (Fig. 7). For these CT scans, the angular sampling interval is 1°, and the total angular distance is 360°. The detector element size is 0.2mm, and the projection dimension is 760 × 190.
4.2 Projection collection and fusion
For the complex object in Fig. 1, if the existing fixed-energy CT scanning method is used, the distribution of the minimum gray level of all rotation angles is shown in Fig. 4. Because the effective thickness of the complex object has a notable difference at all angles, the gray level distribution also has a significant difference. Thus, the partial projection angles have an inadequate dose, which influences CT reconstruction quality. Then, the SAVE method is used to collect the projection. The contrasting minimum gray level distribution is shown in Fig. 8. And, the adaptive X-ray tube voltages at all the scanning angles are shown in Fig. 9.
In Fig. 8, the red curve demonstrates the change in the gray level quality in fixed-energy imaging mode at all angles, and the blue curve represents the corresponding change for adaptive-energy imaging. With the SAVE method, the gray level distribution at all the angles is relatively stable, which can ensure effective imaging. Also from Fig. 9, the distribution of the voltage at all the angles can be corresponding to the gray distribution of the single high energy scanning (Fig. 8). For example, in Fig. 8, the gray at 0°, 180°, and 360° are the highest, which are corresponding to the minimum effective thickness. So that will use the minimum tube voltages, which are shown in Fig. 9. So the distribution of the voltages at all the angles can illustrate the effectiveness of SAVE method. However, the edge information is lost (Fig. 10). Then, the prescanning projection and SAVE scanning projection are fused to obtain the complete projection information.
Fig. 10. Projection and fusion projection at one angle: (a) SAVE scanning projection, (b) prescanning projection, (c) projection of the effective fusion area, (d) projection of the edge fusion area.
In Fig. 10, for the SAVE scanning projection, the edge is lost, but the internal structure is clear. However, in the prescanning projection, the edge is complete, whereas the internal structure is blurred and the contrast is lower. With fusion, the fused projection has a complete edge and a clear internal structure. Of course, if only the edge area of the SAVE projection is fused the, the contrast with the internal structure is higher.
4.3 CT reconstruction
Here, the FDK reconstruction algorithm is used to obtain the CT image [21]. The CT image dimension is 512 × 512 × 190, the voxel size is 0.2mm. In order to analyze the CT result, the two-dimension slice is used. For the complex object in Fig. 1, the voltage is set to 220 kV, and the current is set to 1.5 mA. Then, CT projections are acquired, and the CT image is reconstructed. The CT reconstruction image is shown in Fig. 11.
From Fig. 11, because of the loss of projection information, the reconstruction information is not complete, and the edge of the object is lost. Through SAVE scanning, excellent results can be obtained. A comparison with many other scanning methods was performed.
The results shown in Fig. 12 clearly illustrate the improvement in the reconstruction slice by the method proposed in this paper, especially for the edge of the object. Compared to the other images, the fusion projection reconstruction image has a higher contrast ratio and more clearly presents the complex structure. Partial comparisons of the details labeled in Fig. 12 are shown in Fig. 13.
Fig. 12. CT reconstruction by different processes: (a) prescanning reconstruction, (b) SAVE scanning reconstruction, (c) projection reconstruction of the effective fusion area, and (d) projection reconstruction of the edge fusion area.
Fig. 13. Detailed comparisons of three areas labeled in Fig. 12. (a)-(c): prescanning, (d)-(f): SAVE scanning, (g)-(i): effective fusion area, (k)-(l): edge fusion area.
The areas labeled in Fig. 13 are representative edges of the complex object. They cannot be reconstructed clearly by the existing fixed-energy CT scanning method (Fig. 11). However, in prescanning, the edge information is improved (Figs. 13(a) and 13(b)), and in SAVE scanning, a better reconstruction result can be obtained for the thicker area (Fig. 13(f)) because of the increased ray intensity. However for the thinner edge areas, the full edge information cannot be obtained by SAVE scanning (Figs. 13(d) and 13(e)). By the effective fusion area, all the structure information can get (Figs. 13(g), 13(h) and 13(i)). Then, by the fusion of the edge area, as shown in Figs. 13(j), 13(k) and 13(l), superior edge reconstruction can be obtained. Therefore, the fusion image shows the shape characteristics of the workpiece very effectively. The artifacts are smaller, the edge and background are clearer, and the image quality is higher. In addition, three-dimensional visualization is used to verify that the method proposed in this paper can completely reconstruct the complex structure, which is shown in Fig. 14.
Fig. 14. Three-dimensional visualization result. Left: the whole structure, Middle: the sharp corner structure, Right: the hole structure.
4.4 Experimental sample
In order to verify the universality of the proposed method, another experiment of the engine blade is performed. In Fig. 15, the blade is also a complicated object, with the different thickness at the different CT rotation angle, similar to Fig. 1.
Based on the above the process: pre-scanning, SAVE scanning, projection fusion, and CT reconstruction, the HDR imaging is performed. According to the size and shape, the tube voltage and tube current of the pre-scanning are set to 280 kV and 1.6 mA. In SAVE scanning, the range of the tube voltage is 350 kV to 430 kV, and the tube current is set to 1.6 mA. CT result of one section is shown in Fig. 16(b). Also the CT result of the fixed energy scanting with 430 kV and 1.6 mA is shown in Fig. 16(a).
Fig. 16. CT reconstruction by different processes: (a) reconstruction with the fixed energy scanning, (b) reconstruction with SAVE scanning and projection fusion.
From Fig. 16, for the blade, because of the complicated structure, the traditional CT scanning with the fixed-energy can’t get the clear edge information again. By SAVE scanning and projection fusion, the full edge can be obtained. Also three-dimensional visualization is used to show the complete structure (Fig. 17). So it is further proved that the proposed method is effective for complex structure workpiece.
5. Conclusion
For a complex workpiece analyzed by an HDR imaging device, it is difficult for the existing fixed-energy imaging method to capture full projection information, such as a clear internal structure and complete edge information. In this paper, an adaptive-energy CT imaging method that operates between scan angles is proposed to solve this problem. The new method contains three parts: prescanning, SAVE scanning, and fusion of prescanning and SAVE scanning. Prescanning obtains clear edge information and sets the initial scanning angle and voltage. SAVE scanning improves the matching between the energy and the effective thickness of the scanned workpiece and obtains clear internal construct information about the workpiece. These projections are then preprocessed by fusion and reconstruction. In the fusion procedure, effective area fusion and edge area fusion are separately used to fuse the prescanning projection and SAVE scanning projection. Then, the FDK algorithm was used to reconstruct the HDR-CT image. In the experiment, an actual titanium metal workpiece with a complicated structure is used. With the energy self-adaptation acquisition and edge area fusion, the gray level consistency of the projections is enhanced. Additionally, the CT results are obviously improved with more complete information and higher image quality. At last, the engine blade is used to verify the effectiveness of the proposed method further.
Funding
National Natural Science Foundation of China (61571404, 61871351, 61801437, 61971381); Science Foundation of Shanxi Province (201801D221207, 201801D221206).
Acknowledgments
We are grateful to the team led by professor Sun Yi from Dalian University of Technology for their design of the test workpiece.
Disclosures
The authors declare no conflicts of interest.
References
1. H. Jiang, “Computed Tomography: Principles, Design, Artifacts, and Recent Advances,” SPIE-the International Society for Optical Engineering. 2003, 30–38 (2009).
2. J. T. Wei, Y. Han, and P. Chen, “Improved contrast of materials based on multi-voltage images decomposition in X-ray CT,” Meas. Sci. Technol. 27(2), 025402 (2016). [CrossRef]
3. D. L. Chiffre, S. Carmignato, J. P. Kruth, R. Schmitt, and A. Weckenmann, “Industrial applications of computed tomography,” CIRP Ann. 63(2), 655–677 (2014). [CrossRef]
4. P. Chen, Y. Han, and J. X. Pan, “High-Dynamic-Range CT Reconstruction Based on Varying Tube-Voltage Imaging,” PLoS One 10(11), e0141789 (2015). [CrossRef]
5. M. A. Haidekker, D. K. Morrison, A. Sharma, and E. Burke, “Enhanced dynamic range x-ray imaging,” Comput. Biol. Med. 82(C), 40–48 (2017). [CrossRef]
6. B. Liu, Y. Han, J. X. Pan, and P. Chen, “Multi-energy image sequence fusion based on variable energy X-ray imaging,” J. X-Ray Sci. Technol. 22(2), 241–251 (2014). [CrossRef]
7. Y. H. Li, Y. Han, and P. Chen, “X-ray energy self-adaption high dynamic range (HDR) imaging based on linear constraints with variable energy,” IEEE Photonics J. 10(1), 1 (2017). [CrossRef]
8. A. Macovski, R. E. Alvarez, J. L. H. Chan, J. P. Stonestrom, and L. M. Zatz, “Energy dependent reconstruction in X-ray computerized tomography,” Comput. Biol. Med. 6(4), 325–336 (1976). [CrossRef]
9. A. Alpers and P. Gritzmann, “On double-resolution imaging in discrete tomography,” SIAM J. Discrete Math. 32(2), 1369–1399 (2018). [CrossRef]
10. Y. M. Guo, L. Zeng, C. X. Wang, and L. L. Zhang, “Image reconstruction model for the exterior problem of computed tomography based on weighted directional total variation,” Appl Math Model. 52, 358–377 (2017). [CrossRef]
11. M. J. Schrapp and G. T. Herman, “Data fusion in X-ray computed tomography using a superiorization approach,” Rev. Sci. Instrum. 85(5), 053701 (2014). [CrossRef]
12. X. G. Yang, V. D. Andrade, W. Scullin, E. L. Dyer, N. Kasthuri, F. D. Carlo, and D. Gürsoy, “Low-dose x-ray tomography through a deep convolutional neural network,” Sci. Rep. 8(1), 1–13 (2018). [CrossRef]
13. A. Mirone, E. Brun, and P. Coan, “A dictionary learning approach with overlap for the low dose computed tomography reconstruction and its vectorial application to differential phase tomography,” PLoS One 9(12), e114325 (2014). [CrossRef]
14. P. Chen and Y. Han, “Varying-energy CT imaging method based on EM-TV,” Meas. Sci. Technol. 27(11), 114004 (2016). [CrossRef]
15. P. C. Johns and M. Yaffe, “Scattered radiation in fan beam imaging systems,” Med. Phys. 9(2), 231–239 (1982). [CrossRef]
16. H. Kanamori, N. Nakamori, K. Inoue, and E. Takenaka, “Effects of scattered X-rays on CT images,” Phys. Med. Biol. 30(3), 239–249 (1985). [CrossRef]
17. T. Y. Niu and L. Zhu, “Overview of X-ray Scatter in Cone-beam Computed Tomography and Its Correction Methods,” Curr. Med. Imaging Rev. 6(2), 82–89 (2010). [CrossRef]
18. J. Rinkel, L. Gerfault, F. Estève, and J. M. Dinten, “A new method for x-ray scatter correction: first assessment on a cone-beam CT experimental setup,” Phys. Med. Biol. 52(15), 4633–4652 (2007). [CrossRef]
19. M. C. Chen, H. M. Li, Z. Y. Chen, and J. Shen, “An examination of mass thickness measurements with X-ray sources,” Appl. Radiat. Isot. 66(10), 1387–1391 (2008). [CrossRef]
20. J. T. Wei, P. Chen, and J. X. Pan, “Multi-voltage Digital Radiography images fusion based on gray consistency,” Nondestructive Evaluation testing: New Technology & Application IEEE. 209–211 (2013).
21. L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beam algorithm,” J. Opt. Soc. Am. 1(6), 612–619 (1984). [CrossRef]
