Development of a multi-axis X-ray CT for highly accurate inspection of electronic devices Toru Kano 1, Michihiko Koseki 2 More info about this article: http://www.ndt.net/?id=20843 1 Tokyo University of Technology, 1404-1 Katakura-cho, Hachioji, Tokyo, Japan, e-mail: kanohtr@stf.teu.ac.jp 2 Shinshu University, 3-15-1 Tokida, Ueda, Nagano, Japan, e-mail: koseki@shinshu-u.ac.jp Abstract X-ray CT is a non-destructive inspection device that can generate cross-sectional images of an object. This technology has been primarily used as a diagnostic tool and an inspection technique. However, significant problems with metal artifacts remain. A metal artifact is a strong radial noise, which makes it difficult to diagnose patients and inspect products that contains metal implants. Although many studies for metal artifact reduction were conducted, it has never reached a drastic solution. In particular, it's very tough to inspect electronic devices containing a lot of metals. In this paper, we propose new concepts of X-ray CT for inspecting electronic devices that is based on multi-axis rotation mechanism. The feasibility of the multi-axis X-ray CT was verified by a preliminary experiment and simulation. Keywords: X-ray CT, metal artifact, non-destructive inspection, reconstruction algorithm 1 Background Computed tomography (CT) is a method for generating cross-sectional images of an object. This technology is commonly used as a diagnostic tool in medicine such as in the early detection of cancer, as well as for non-destructive inspection in industrial engineering applications. Despite its widespread use, technical challenges remain. A pervasive problem is an imaging artifact caused by radial noise resultant of objects containing high-z materials; typically metals (Figure 1). Metal artifacts make it difficult to diagnose patients and inspect products containing metal implants, and this continues to be problematic in CT imaging. E. Meyer et al. proposed normalize metal artifact reduction (NMAR) [1], which is composed of a segmentation and a normalizing operation. This method partially removed metal artifacts, but other streaking artifacts were generated. Also, it is necessary to extract accurate metal regions by thresholding for applying this algorithm. Zhao et al. proposed a method using a wavelet-based weighting function in the field of view [2], Zhang et al. modified metal regions on cross-sectional images by using projection data from two directions [3], and Abdoli et al. corrected projection data using weighted virtual sinograms [4]. There are also several methods based on reconstructed images. M. Bal et al. conducted adaptive filtering algorithm [5], and segmented reconstructed images into different material classes. Y. Chen et al. enhanced metal artifacts with a large-scale non-local mean filter [6], and segmented images into metal parts and artifact parts. T. Kano et al. focused on X-ray energy and proposed an iterative reconstruction algorithm from a deteriorated image [7]. Although many other methods of metal artifact reduction have proposed, most of them include the process of interpolation of projection data, and it s really difficult to apply them to electronic devices that contains a lot of metals, because they have many regions that have to be interpolated, and make strong metal artifacts due to saturation of X-ray transmission intensities. In this paper, we propose a new X-ray CT system based on multi-axis rotation mechanism, in order to establish highly accurate inspection technology of electronic devices. Figure 1: CT images showing metal artifacts on an electronic circuit and a cell phone. 1
2 Metal artifacts Causes of metal artifacts can be divided into two main factors shown in the following: The X-ray energy spectrum is not considered properly during image reconstruction. The X-ray transmission intensities are saturated -- X-rays don't reach the detector. The above two causes are often treated as the same, and the differences are rarely mentioned even in articles related to metal artifact reduction. However, the amount of available information is quite different between them. 2.1 Cases where X-rays reach the detector In cases where enough X-rays reach the detector, metal artifacts are due to improper consideration of the X-ray energy spectrum. We have examined about the fact, and obtained some positive results [7]. Modern CT imaging systems define the projection data for reconstruction calculation as the following equation. =, where denotes the material-specific X-ray attenuation coefficient, and is the path length. This equation indicates that is linear with respect to an object thickness when the object is homogeneous, but this is only true for monochromatic X-ray fields. Because attenuation depends on X-ray energies, the actual projection data becomes the integration of energy as follows: = ln in exp. in In order to solve the discrepancy due to the nonlinearity of projection data, we assumed the distribution of X-ray attenuation coefficients, and embedded the information in iterative calculation. The equation for the calculation is: + = + [ + ln { in exp( ) } ], in where is a coefficient that decides the feedback ratio. By using this calculation, we confirmed that metal artifacts were theoretically reduced on simulation. That is to say, when X-rays transmitted through metals reach the detector, metal artifacts can be technically removed. Figure 2: Metal artifact reduction using iterative reconstruction based on X-ray energy. Figure 3: The simulation result of metal artifact reduction. is the stomach phantom containing metals, is the application result. 2.2 Cases where X-rays don t reach the detector On the other hand, in cases that the X-ray transmission intensities are saturated by metals, the saturation regions on the projection data aren't available in any correction processing. Although some metal artifacts would be reduced by interpolating them, it's impossible to restore accurate information of non-metallic regions that were lost by metals, and the non-metallic regions adjoining metals can't be reconstructed. The new X-ray imaging mechanism, which is proposed in this paper, targets at such cases. 2
3 Multi-axis X-ray CT Consider a projection to an electronic device from a certain direction. Even if X-ray transmission intensities are saturated at the direction by superimposed metals, there will be a lot of three-dimensional directions that X-ray intensities aren't saturated (Figure 4). Therefore, if projection data of the object in various postures can be acquired by using a multi-axis mechanism, lost nonmetallic regions will be interpolated, and accurate reconstruction can be conducted. In this study, we propose two concepts of multi-axis X-ray CT below. Both mechanisms are assumed to place in a single-axis X-ray CT for industrial fields that the X-ray source is fixed and the stage rotates. Single-axis projection Multi-axis projection Figure 4: A basic idea of multi-axis X-ray CT. 3.1 Dome-type mechanism The dome-type mechanism is composed of: the rotation of the stage, the rotation to the tilt direction of the dome, and the rotation of the upper side of the dome (Figure 5 ). By placing an object on the top of the dome, X-ray attenuation due to the dome itself can be suppressed. However, since the rotation range of the tilt direction of the dome is limited, it cannot give the object arbitrary posture angle. 3.2 Gimbal-type mechanism The gimbal-type mechanism is composed of: the rotation of the stage, and the rotations of two frames (Figure 5 ). An object is fixed using the central holder, and this mechanism can give the object arbitrary posture angle. However, it's necessary to control the posture and perform reconstruction considering attenuation due to the stepping motors and the frames. Dome-type Gimbal-type Figure 5: Concepts of multi-axis mechanisms. 3.3 Reconstruction algorithm As long as projection data are recorded with angles of each axis of the mechanism, we can perform back-projection calculations to three-dimensional space using the Feldkamp method. In this calculation, if regions that transmission distance of metals is short are preferentially back-projected, metal artifacts will be reduced. Also, the more projection data you collect, the more quality the reconstructed image has. However, a large amount of projection directions means a waste of a huge memory space and a significant increase in calculation time. In order to develop a practical inspection technology for electronic devices, we should avoid unnecessary projections as much as possible for saving memory space and calculation time. As a new projection flow, we propose following steps (Figure 6): 1. Perform a conventional X-ray projection with a single-axis rotation. 2. Extract saturation regions and directions based on the projection data. 3. Collect additional projection data around the saturation directions by rotating other two-axes. 4. Perform the Feldkamp algorithm using data that aren t saturated. If enough data for performing back-projection are collected, metal artifacts will be reduced. Furthermore, by fusing this method with iterative reconstruction algorithm which is considering X-ray energy, the effectiveness will be more improved. Figure 6: The process flow of the multi-axis X-ray CT. 3
4 Preliminary experiment As a fundamental experiment for validation of our proposed method, we conducted CT imaging of an experimental sample (Figure 7), which is composed of a cross-shaped metal column and a resin. One projection was performed while placing the sample horizontally (Figure 8 ), and the other projection was performed while tilting the stage with 30 degrees (Figure 8 ). You can see that more information about the resin was collected and saturation regions were reduced in the latter projection result. Also, by replacing the non-metal regions of the former (horizontal) projection with the corresponding regions of the latter (tilting 30 degrees) projection for these reconstruction results, the metal artifacts in the former projection were partially reduced (Figure 9). This result indicates that image quality could be improved and metal artifacts could be reduced by using projection data collected from various directions three-dimensionally. Figure 7: An experimental sample. Figure 8: The preliminary experiment. is the projection result with horizontal placing and is the projection result with 30 degrees tilting. ` (c) Figure 9: The preliminary reconstruction results. and are the FBP results with horizontal placing and 30 degrees tilting, respectively. (c) is the synthesis result. 4
5 Simulation experiments In this section, the simulation experiments performed to confirm the application potency of multi-axis back projection algorithm are described. 5.1 Simulation conditions A numerical three-dimensional phantom that imitates the experimental sample shown in Figure 10 was prepared. This phantom was composed of two materials -- a resin and a cross-shaped iron. The image size was, the source-to-image distance was 1946 mm, and the number of projection was 256. The X-ray source was a cone beam, and the tube voltage was 100 KeV. Also, a discrete X-ray energy distribution and X-ray attenuation coefficients to each energy were prepared for generating metal artifacts on simulation (Table 1). The X-ray attenuation coefficients for each element were calculated from mass attenuation coefficients found in the National Institutes of Standards and Technology (NIST) database [8]. Energy X-ray Air, dry Resin Iron [KeV] Intensity [/mm] [/mm] [/mm] 10 0.000 6.170E-04 0.39948 134.3 20 1.604 9.374E-05 0.06800 20.22 30 26.93 4.263E-05 0.03608 6.438 40 49.12 2.994E-05 0.02797 2.857 50 46.77 2.506E-05 0.02468 1.542 60 42.78 2.259E-05 0.02290 0.9488 Figure 10: A numerical three-dimensional phantom composed of a resin embedded with a cross-shaped iron. is a crosssectional image at the height of the center, and is the volume rendering result. 80 46.31 2.003E-05 0.01953 0.4687 100 14.00 1.857E-05 0.01733 0.2927 Table 1: X-ray intensity and attenuation coefficients for materials used in forward projection. 5.2 Multi-axis projection In this simulation, we added a rotation angle that tilts three-dimensional space on the stage in addition to a rotation angle of the stage (Figure 11). By changing continuously with the change of from to, a simple multi-axis projection is performed. The multi-axis back projection can also be performed by Feldkamp algorithm based on recorded values of and. The projection result at the initial state ( =, = ) is shown in Figure 12. Figure 11: Projection image for illustration purposes and definition of two angles. Figure 12: The projection result of the phantom shown in Figure 10 at the initial state ( =, = ). As experiments of multi-axis projection, three kinds of rotation patterns were performed -- =, =, =. The relationship between two rotate angles are shown in Figure 11. The condition = indicates an ordinary projection, and other two conditions indicate multi-axis projection. 5
5.3 Multi-axis back projection Back projection results for each condition are shown in Figure 13. As cross-sectional images for presentation, three different heights ( =, 8, ) were selected. At the height of =, there is only the resin region, and at the heights of = 8,, there is the cross-shaped iron embedded in the resin. In the result at =, while metal artifacts didn't generate at the height of =, distinctive metal artifacts were generated from the iron at the heights of = 8 and. On the other hand, in the results of =, 8 4 at the height of = 8, we can see that metal artifacts were reduced and the cross-shape would be observed. However, we also see that the results at other heights were changed. Although there were no metal artifacts at =, =, some artifacts were generated for multi-axis projection results at the height of =. This is because metal artifacts that were generated only two-dimensionally came to spread to three-dimensional space by multi-axis projection. Here, remember that the metal artifacts were reduced at the height of = 8. These results indicate that multiaxis projection could distribute and weaken metal artifacts three-dimensionally. Furthermore, if metal artifacts were weakened and saturation regions of projection data reduced, you can apply our previous iterative algorithm that is considering X-ray energy. 6 Conclusion We proposed new systems of multi-axis X-ray CT, and conducted preliminary experiments both on actual CT scan and simulation. As the result, it was indicated that metal artifact would be reduced and electronic devices could be inspected with a high degree of accuracy. By using the two proposed mechanisms selectively depending on the situation, and combining them with our previous algorithm based on X-ray energy, effective metal artifact reduction will be realized. As a next step, we will first implement a function selecting rotation angles of multi-axis projection automatically. Acknowledgements We wish to acknowledge valuable discussions with Mr. Kazuo Kikuchi and Mr. Takashi Tomitsuka of Comscantecno Co., Ltd. References [1] E. Meyer, R. Raupach, M. Lell, B. Schmidt and M. Kachelrieß, Normalized metal artifact reduction (NMAR) in computed tomography, Medical Physics, 37-10 (2010), 5482-5493. [2] S. Zhao, K.T. Bae, B. Whiting and G. Wang, A wavelet method for metal artifact reduction with multiple metallic objects in the field of view, Journal of X-ray Science and Technology, 10-1-2 (2002), 67-76. [3] Y. Zhang, L. Zhang, X.R. Zhu, A.K. Lee, M. Chambers and L. Dong, Reducing metal artifacts in cone-beam CT images by preprocessing projection data, International Journal of Radiation Oncology Biology Physics, 67-3 (2007), 924-932. [4] M. Abdoli, M. R. Ay, A. Ahmadian, R. A. Dierckx and H. Zaidi, Reduction of dental filling metallic artifacts in CTbased attenuation correction of PET data using weighted virtual sinograms optimized by a genetic algorithm, Medical Physics 37-12 (2010), 6166-6177. [5] M. Bal and L. Spies, Metal artifact reduction in CT using tissue-class modeling and adaptive prefiltering, Medical Physics 33-8 (2006), 2852-2859. [6] Y. Chen, Y. Li, H. Guo, Y. Hu, L. Luo, X. Yin, J. Gu and C. Toumoulin, CT Metal Artifact Reduction Method Based on Improved Image Segmentation and Sinogram In-Painting, Mathematical Problems in Engineering 2012 (2012), 786281. [7] T. Kano and M. Koseki, A new metal artifact reduction algorithm based on a deteriorated CT image, Journal of X- Ray Science and Technology 24-6 (2016), 901-912. [8] J. H. Hubbell and S. M. Seltzer, "Tables of X-Ray Mass Attenuation Coefficients and Mass Energy-Absorption Coefficients from 1 kev to 20 MeV for Elements Z = 1 to 92 and 48 Additional Substances of Dosimetric Interest, NIST, http://www.nist.gov/pml/data/ (2015) 6
= = 8 = Line charts of relationship 7th Conference on Industrial Computed Tomography, Leuven, Belgium (ict 2017) = = = Figure 13: The simulation results of multi-axis projection and back projection. 7