- Email: [email protected]
Segmentation of Synchrotron Radiation micro-Computed Tomography Images using Energy Minimization via Graph Cuts
Applied Radiation and Isotopes 70 (2012) 1284–1287
Contents lists available at SciVerse ScienceDirect
Applied Radiation and Isotopes journal homepage: www.elsevier.com/locate/apradiso
Segmentation of Synchrotron Radiation micro-Computed Tomography Images using Energy Minimization via Graph Cuts Anderson A.M. Meneses a,b, Alessandro Giusti c, Andre´ P. de Almeida b,d,n, Liebert Nogueira d, Delson Braz d, Carlos E. de Almeida e, Regina C. Barroso b a
´, Brazil Federal University of Western Para Physics Institute, Rio de Janeiro State University, Brazil c IDSIA (Dalle Molle Institute for Artificial Intelligence), University of Lugano, Switzerland d Nuclear Engineering Program, Federal University of Rio de Janeiro, Brazil e Radiological Sciences Laboratory, Rio de Janeiro State University, Brazil b
a r t i c l e i n f o
a b s t r a c t
Available online 12 November 2011
The research on applications of segmentation algorithms to Synchrotron Radiation X-Ray microComputed Tomography (SR-mCT) is an open problem, due to the interesting and well-known characteristics of SR images, such as the phase contrast effect. The Energy Minimization via Graph Cuts (EMvGC) algorithm represents state-of-art segmentation algorithm, presenting an enormous potential of application in SR-mCT imaging. We describe the application of the algorithm EMvGC with swap move for the segmentation of bone images acquired at the ELETTRA Laboratory (Trieste, Italy). & 2011 Elsevier Ltd. All rights reserved.
Keywords: Synchrotron Radiation Phase Contrast Imaging Micro-Computed Tomography Image Segmentation Energy Minimization via Graph Cuts
1. Introduction Synchrotron Radiation (SR) (Elder et al., 1947) facilities provide high brilliance X-rays with very high flux at small source size compared to tube X-rays, enabling investigations of samples in the micro- and even the sub-micrometer levels. Therefore microComputed Tomography (mCT) (Hounsfield, 1973) obtained with SR X-ray conjugates several qualities for the investigation of biomedical structures such as high brilliance and high space resolution. Besides the characteristics resulting from the high coherence and monochromaticity of the beam, it is also possible to achieve the enhancing of contrast during imaging due to wavefield phase information. This interesting specificity of the SR, the phase contrast effect, is useful and important in many biomedical applications (Lewis, 2004). Thus, given the special characteristics of SR-mCT and the possibilities of the current third generation SR facilities, especially regarding their application in medicine and biology, image processing algorithms applied to SR-mCT segmentation are currently being investigated. As examples of the application of novel approaches to CT images, Krebs et al. (2009) describe the segmentation of high resolution CT images using fuzzy approaches for the assessment of trabecular distances. Meneses et al. (2008, 2009,
n Corresponding author at: Federal University of Rio de Janeiro, Nuclear Engineering Program, 21941-972 Rio de Janeiro, Brazil. E-mail address: [email protected] (A.P. de Almeida).
0969-8043/$ - see front matter & 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.apradiso.2011.11.011
2010) reported the application of Artificial Neural Networks (ANNs) to the segmentation of SR-mCT biomedical images, as well as the proposal and assessment of ANNs training strategies for segmentation with respect to SR-mCT bone images for histomorphometry applications. Since the mCT imaging modality is only at its beginning (Chappard et al., 2008), algorithms that yield high-quality results shall be also investigated and assessed in order to be applied to this imaging modality. One of those state-of-art algorithms that yield remarkable results is the min-cut/max-flow algorithm (Greig et al., 1989; Boykov et al., 2001; Kolmogorov and Zabih, 2004; Boykov and Kolmogorov, 2004) used for segmentation based on energy minimization. The energy minimization problem is relevant in many computer vision areas. The contributions of Boykov et al. (2001), especially regarding the swap move and expansion move algorithms, represent a breakthrough in image segmentation research, yielding fast and accurate results in many image processing fields. For simplicity, we refer to the application of the min-cut/max-flow swap move algorithm for the minimization of energy as Energy Minimization via Graph Cuts (EMvGC). The specimens used in this work were imaged at the SYRMEP (SYnchrotron Radiation for MEdical Physics) beam line at Elettra Synchrotron Laboratory in Trieste, Italy during the project 20090192 (Dose Effects of Radiotherapic Procedures on Bone Structure Using SR-mCT—Part II). In this project we intend to evaluate possible alterations in microstructure of ribs of Wistar rats due to radiotherapy simulation of breast cancer since radiation induced rib fracture has been recognized as a normal tissue
A.A.M. Meneses et al. / Applied Radiation and Isotopes 70 (2012) 1284–1287
1285
Laboratory, Trieste, Italy. SYRMEP beam line provides a monochromatic laminar-section X-ray with a maximum area of about 160 5 mm2 at 20 keV, at a distance of about 23 m from the source. The system consists of a Si (111) crystal working at Bragg configuration. The useful energy range is 8–35 keV. The intrinsic energy resolution of the monochromator is about 10 3. The typical flux measured at the sample position at 17 keV is about 1.6 108 photons/mm2 s with a stored electron beam of 300 mA as ELETTRA operates at 2 GeV (Abrami et al., 2005). The samples, mounted on a rotation stage, were illuminated with monochromatic radiation (E¼15 keV) with a sample-to-detector distance d¼10 cm. The detector system is a water-cooled CCD camera (Photonics Science XDIVHR), a 12/16 bit, 4008 2672 full frame, 4.5 mm pixel size CCD camera with a field of view of 18 12 mm2. It is coupled to a Gadox scintillator placed on a straight fiber optics coupler.
Fig. 1. Three-dimensional visualization of dorsal portion of Wistar rat rib bone (sample E8D).
(a) (b)
(c)
Fig. 2. Slice 305 of Wistar rat rib (bone from the dorsal portion; sample E8D), with the indication of: (a) marrow; (b) bone; (c) cartilage.
complication after conventional radiotherapy when the radiation filed is the thoracic region, such as radiotherapy for breast or lung cancer (Pettersson et al., 2009). The quantification of bone microstructures includes the extraction of morphological parameters directly from mCT scans, which have to be segmented. Thus, the aim of this work is to assess the application of the algorithm EMvGC for SR-mCT to tomographic scans, such as the one depicted in Fig. 1, in order to eliminate non-bone portions such as marrow and cartilage (Fig. 2) and to obtain images with better quality, considering the scope of the aforementioned project, that is, regarding the evaluation and quantification of bone microstructures. The remainder of this article is organized as follows. The materials and methods are described in Section 2. The results yielded by EMvGC are presented in Section 3. The conclusion is presented in Section 4.
2. Materials and methods 2.1. Synchrotron radiation micro-computed tomography The images of interest were obtained at the SYnchrotron Radiation for MEdical Physics (SYRMEP) beamline at the ELETTRA
2.2. Samples characteristics Female Wistar Rats (Rattus norvegicus), were obtained from the Radiological Sciences Laboratory of the Rio de Janeiro State University, in Brazil. Animals (n ¼5), aging about three months and weighting about 200 g, were maintained on a 12 h light/dark cycle with food and water provided ad libitum. The animals were maintained under standard animal facility condition until the sacrifice. Their ribs were removed and immediately fixed with 10% formaldehyde neutral buffered solution. Afterwards, the specimens were carefully cleaned and allowed to air dry. The dorsal region of each rib was then imaged. 2.3. Energy minimization via graph cuts In computer vision, many tasks that involve the assignment of a label to pixels are naturally stated in terms of energy minimization. According to the notation used by Boykov et al. (2001), a piecewise smooth labeling f consistent with the observed data is sought, in order to minimize the energy function. According to f, a label fpAL, where L is a finite set of labels, is assigned to each pixel pAP. The energy functions E(f) considered for minimization via Graph Cuts are of the form X X Eðf Þ ¼ V p,q ðf p ,f q Þ þ Dp ðf p Þ: ð1Þ fp,qg A N
pAP
The function Dp(fp) typically corresponds to (fp Ip)2 and measures the disagreement between fp and the observed intensity Ip, for each pixel p. The function Vp,q(fp,fq) computes the cost of assigning fp and fq to the pair of pixels {p, q}AN, where the set N often comprises adjacent pixels, although not necessarily. During the evaluation of E(f), relatively high values of Vp,q indicate that f does not correspond to a piecewise smooth labeling. Min-cut/max-flow optimization problems may be formulated as follows. Given a weighted graph G/V,ES, where V is a collection of nodes (or vertices) connected by a collection of edges E, with two distinguishable nodes called terminals (or source and sink), the minimum cut problem is to find the cut C E that separates the terminal nodes, with minimum cost. The cost of the cut 9C9 corresponds to the sum of its edge weights. Boykov et al. (2001) proposed two algorithms for multidimensional energy minimization, achieving approximate solutions with guaranteed optimality bounds: the swap move algorithm and the expansion move algorithm. Unlike other algorithms such as simulated annealing, those algorithms provide solutions in a nearly linear time in practice. The library used in our implementations (gco-v3.0) is available for research purposes at http://vision.csd.uwo.ca/code/.
1286
A.A.M. Meneses et al. / Applied Radiation and Isotopes 70 (2012) 1284–1287
3. Results For each tomographic scan, 900 projections of the sample were acquired for equally spaced rotation angles and measurement time of about 1 s over a total rotation of 1801. The Tomo Project program, developed at SYRMEP (Montanari, 2003), was used for tomographic reconstruction. From each sample, a centered region of interest (ROI) was selected using the software ImageJ. Two-dimensional slices and three-dimensional reconstructed images obtained for dorsal portions of a given specimen (named E8D) are presented in Figs. 1 and 2, respectively. In the preprocessing phase, a median filter was used in order to reduce noise. It is well known that information carried by the phase of X-Ray wavefield, or phase effects, may be converted into image contrast, enhancing detail visibility, especially if compared to the poor contrast given by media with small X-ray absorption differences. Small amounts of marrow and cartilage, for quantification purposes and as depicted in Fig. 2, may be regarded as artifacts in the image and should be removed with the application of the EMvGC algorithm. Fig. 3 is the result of the application of the EMvGC algorithm to the image depicted in Fig. 2. Fig. 4 corresponds to the threedimensional visualization after the application of EMvGC to all
slices of the tomographic scan depicted in Fig. 1. In order to achieve those results in the tomographic scan segmentation, the EMvGC algorithm considered 3D information in our implementations, that is, besides the bi-dimensional information contained at each slice, the information of adjacent slices was also considered during the execution of the algorithm. The results of the EMvGC algorithm were also compared to the manual segmentation using the Dice Similarity Coefficient (Dice, 1945; Zijdenbos et al., 1994), for which the algorithm obtained the value of 99.94%, indicating that the automated segmentation using EMvGC was accurately performed, coping with the tasks of removing non-bone portions of the slices such as cartilage and marrow depicted in Fig. 2.
4. Conclusion The present article reports the application of EMvGC algorithm to SR-mCT biomedical images. EMvGC is a state-of-art algorithm, which is fast and accurate, with many applications in computer vision. We presented the results of the automated segmentation of a Wistar Rat rib scan, in which the algorithm achieved a remarkable performance, with a Dice Similarity Coefficient corresponding to 99.94%, in comparison to manual segmentation.
Acknowledgment We warmly thank Giuliana Tromba and Nicola Sodini from the SYRMEP beamline for their invaluable help during the experiment. The authors thank to the Brazilian agencies CNPq, CAPES, FAPESPA and FAPERJ for partial financial support. The images were obtained during a visit (R.C.B.) within the Associate Scheme of the Abdus Salam International Centre for Theoretical Physics (ICTP). The authors are thankful to Cherley Borba (LCR/UERJ) who made the samples available. References
Fig. 3. Result of the application of the EMvGC algorithm to the slice depicted in Fig. 2.
Fig. 4. Three-dimensinoal visualization of dorsal portion of Wistar rat rib bone (sample E8D) after the segmentation with EMvGC.
Abrami, A., Arfelli, F., Barroso, R.C., et al., 2005. Medical applications of synchrotron radiation at the SYRMEP beamline of ELETTRA. Nucl. Instrum. Methods A 548, 221–227. Boykov, Y., Veksler, O., Zabih, R., 2001. Fast approximate energy minimization via graph cuts. IEEE Trans. Pattern Anal. Mach. Intell. 23, 1222–1239. Boykov, Y., Kolmogorov, V., 2004. An experimental comparison of min-cut/maxflow algorithms for energy minimization in vision. IEEE Trans. Pattern Anal. Mach. Intell. 26, 1124–1137. Chappard, D., Basle´, M.-F., Legrand, E., Audran, M., 2008. Trabecular bone microarchitecture: a review. Morphologie 92, 162–170. Dice, L.R., 1945. Measures of the amount of ecologic association between species. Ecology 26, 297–302. Elder, F.R., Gurewitsch, A.M., Langmuir, R.V., Pollock, H.C., 1947. Radiation from electrons in a synchrotron. Phys. Rev. 71, 829–830. Greig, D.M., Porteous, B.T., Seheult, A.H., 1989. Exact maximum a posteriori estimation for binary images. J. Royal Stat. Soc. B 51, 271–279. Hounsfield, G.N., 1973. Computerized transverse axial scanning (tomography): Part 1. Description of system. Br. J. Radiol. 46, 1016–1022. Kolmogorov, V., Zabih, R., 2004. What energy functions can be minimized via graph cuts? IEEE Trans. Pattern Anal. Mach. Intell. 26, 147–159. ¨ Krebs, A., Graeff, C., Frieling, I., Kurz, B., Timm, W., Engelke, K., Gluer, C.-C., 2009. High resolution computed tomography of the vertebrae yields accurate information on trabecular distances if processed by 3D fuzzy segmentation techniques. Bone 44, 145–152. Lewis, R.A., 2004. Medical phase contrast x-ray imaging: current status and future prospects. Phys. Med. Biol. 49, 3573–3583. Meneses, A.A.M., Pinheiro, C.J.G., Schirru, R., Barroso, R.C., Braz, D., Oliveira, L.F., 2008. Artificial neural networks applied to bone recognition in X-ray computer microtomography imaging for histomorphometric analysis. In: Proceedings of the IEEE Nuclear Science Symposium and Medical Imaging Conference Record, pp. 5309–5313. Meneses, A.A.M., Pinheiro, C.J.G., Gambardella, L.M., Schirru, R., Barroso, R.C., Braz, D., Oliveira, L.F., 2009. Neural computing for quantitative analysis of human bone trabecular structures in synchrotron radiation X-ray mCT images.
A.A.M. Meneses et al. / Applied Radiation and Isotopes 70 (2012) 1284–1287
In: Proceedings of the IEEE Nuclear Science Symposium and Medical Imaging Conference Record, pp. 3437–3441. Meneses, A.A.M., Pinheiro, C.J.G., Rancoita, P., Schaul, T., Gambardella, L.M., Schirru, R., Barroso, R.C., Oliveira, L.F., 2010. Assessment of neural networks training strategies for histomorphometric analysis of synchrotron radiation medical images. Nucl. Instr. Methods A 621, 662–669. Montanari, F., 2003. SYRMEP TOMO Project, tutorial.
1287
Pettersson, N., Nyman, J., Johansson, K., 2009. Radiation-induced rib fractures after hypofractionated stereotactic body radiation therapy of non-small cell lung cancer: a dose- and volume-response analysis. Radiother. Oncol. 91, 360–368. Zijdenbos, A.P., Dawant, B.M., Margolin, R.A., Palmer, A.C., 1994. Morphometric analysis of white matter lesions in MR images: method and validation. IEEE Trans. Med. Imaging 13, 716–724.
Contents lists available at SciVerse ScienceDirect
Applied Radiation and Isotopes journal homepage: www.elsevier.com/locate/apradiso
Segmentation of Synchrotron Radiation micro-Computed Tomography Images using Energy Minimization via Graph Cuts Anderson A.M. Meneses a,b, Alessandro Giusti c, Andre´ P. de Almeida b,d,n, Liebert Nogueira d, Delson Braz d, Carlos E. de Almeida e, Regina C. Barroso b a
´, Brazil Federal University of Western Para Physics Institute, Rio de Janeiro State University, Brazil c IDSIA (Dalle Molle Institute for Artificial Intelligence), University of Lugano, Switzerland d Nuclear Engineering Program, Federal University of Rio de Janeiro, Brazil e Radiological Sciences Laboratory, Rio de Janeiro State University, Brazil b
a r t i c l e i n f o
a b s t r a c t
Available online 12 November 2011
The research on applications of segmentation algorithms to Synchrotron Radiation X-Ray microComputed Tomography (SR-mCT) is an open problem, due to the interesting and well-known characteristics of SR images, such as the phase contrast effect. The Energy Minimization via Graph Cuts (EMvGC) algorithm represents state-of-art segmentation algorithm, presenting an enormous potential of application in SR-mCT imaging. We describe the application of the algorithm EMvGC with swap move for the segmentation of bone images acquired at the ELETTRA Laboratory (Trieste, Italy). & 2011 Elsevier Ltd. All rights reserved.
Keywords: Synchrotron Radiation Phase Contrast Imaging Micro-Computed Tomography Image Segmentation Energy Minimization via Graph Cuts
1. Introduction Synchrotron Radiation (SR) (Elder et al., 1947) facilities provide high brilliance X-rays with very high flux at small source size compared to tube X-rays, enabling investigations of samples in the micro- and even the sub-micrometer levels. Therefore microComputed Tomography (mCT) (Hounsfield, 1973) obtained with SR X-ray conjugates several qualities for the investigation of biomedical structures such as high brilliance and high space resolution. Besides the characteristics resulting from the high coherence and monochromaticity of the beam, it is also possible to achieve the enhancing of contrast during imaging due to wavefield phase information. This interesting specificity of the SR, the phase contrast effect, is useful and important in many biomedical applications (Lewis, 2004). Thus, given the special characteristics of SR-mCT and the possibilities of the current third generation SR facilities, especially regarding their application in medicine and biology, image processing algorithms applied to SR-mCT segmentation are currently being investigated. As examples of the application of novel approaches to CT images, Krebs et al. (2009) describe the segmentation of high resolution CT images using fuzzy approaches for the assessment of trabecular distances. Meneses et al. (2008, 2009,
n Corresponding author at: Federal University of Rio de Janeiro, Nuclear Engineering Program, 21941-972 Rio de Janeiro, Brazil. E-mail address: [email protected] (A.P. de Almeida).
0969-8043/$ - see front matter & 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.apradiso.2011.11.011
2010) reported the application of Artificial Neural Networks (ANNs) to the segmentation of SR-mCT biomedical images, as well as the proposal and assessment of ANNs training strategies for segmentation with respect to SR-mCT bone images for histomorphometry applications. Since the mCT imaging modality is only at its beginning (Chappard et al., 2008), algorithms that yield high-quality results shall be also investigated and assessed in order to be applied to this imaging modality. One of those state-of-art algorithms that yield remarkable results is the min-cut/max-flow algorithm (Greig et al., 1989; Boykov et al., 2001; Kolmogorov and Zabih, 2004; Boykov and Kolmogorov, 2004) used for segmentation based on energy minimization. The energy minimization problem is relevant in many computer vision areas. The contributions of Boykov et al. (2001), especially regarding the swap move and expansion move algorithms, represent a breakthrough in image segmentation research, yielding fast and accurate results in many image processing fields. For simplicity, we refer to the application of the min-cut/max-flow swap move algorithm for the minimization of energy as Energy Minimization via Graph Cuts (EMvGC). The specimens used in this work were imaged at the SYRMEP (SYnchrotron Radiation for MEdical Physics) beam line at Elettra Synchrotron Laboratory in Trieste, Italy during the project 20090192 (Dose Effects of Radiotherapic Procedures on Bone Structure Using SR-mCT—Part II). In this project we intend to evaluate possible alterations in microstructure of ribs of Wistar rats due to radiotherapy simulation of breast cancer since radiation induced rib fracture has been recognized as a normal tissue
A.A.M. Meneses et al. / Applied Radiation and Isotopes 70 (2012) 1284–1287
1285
Laboratory, Trieste, Italy. SYRMEP beam line provides a monochromatic laminar-section X-ray with a maximum area of about 160 5 mm2 at 20 keV, at a distance of about 23 m from the source. The system consists of a Si (111) crystal working at Bragg configuration. The useful energy range is 8–35 keV. The intrinsic energy resolution of the monochromator is about 10 3. The typical flux measured at the sample position at 17 keV is about 1.6 108 photons/mm2 s with a stored electron beam of 300 mA as ELETTRA operates at 2 GeV (Abrami et al., 2005). The samples, mounted on a rotation stage, were illuminated with monochromatic radiation (E¼15 keV) with a sample-to-detector distance d¼10 cm. The detector system is a water-cooled CCD camera (Photonics Science XDIVHR), a 12/16 bit, 4008 2672 full frame, 4.5 mm pixel size CCD camera with a field of view of 18 12 mm2. It is coupled to a Gadox scintillator placed on a straight fiber optics coupler.
Fig. 1. Three-dimensional visualization of dorsal portion of Wistar rat rib bone (sample E8D).
(a) (b)
(c)
Fig. 2. Slice 305 of Wistar rat rib (bone from the dorsal portion; sample E8D), with the indication of: (a) marrow; (b) bone; (c) cartilage.
complication after conventional radiotherapy when the radiation filed is the thoracic region, such as radiotherapy for breast or lung cancer (Pettersson et al., 2009). The quantification of bone microstructures includes the extraction of morphological parameters directly from mCT scans, which have to be segmented. Thus, the aim of this work is to assess the application of the algorithm EMvGC for SR-mCT to tomographic scans, such as the one depicted in Fig. 1, in order to eliminate non-bone portions such as marrow and cartilage (Fig. 2) and to obtain images with better quality, considering the scope of the aforementioned project, that is, regarding the evaluation and quantification of bone microstructures. The remainder of this article is organized as follows. The materials and methods are described in Section 2. The results yielded by EMvGC are presented in Section 3. The conclusion is presented in Section 4.
2. Materials and methods 2.1. Synchrotron radiation micro-computed tomography The images of interest were obtained at the SYnchrotron Radiation for MEdical Physics (SYRMEP) beamline at the ELETTRA
2.2. Samples characteristics Female Wistar Rats (Rattus norvegicus), were obtained from the Radiological Sciences Laboratory of the Rio de Janeiro State University, in Brazil. Animals (n ¼5), aging about three months and weighting about 200 g, were maintained on a 12 h light/dark cycle with food and water provided ad libitum. The animals were maintained under standard animal facility condition until the sacrifice. Their ribs were removed and immediately fixed with 10% formaldehyde neutral buffered solution. Afterwards, the specimens were carefully cleaned and allowed to air dry. The dorsal region of each rib was then imaged. 2.3. Energy minimization via graph cuts In computer vision, many tasks that involve the assignment of a label to pixels are naturally stated in terms of energy minimization. According to the notation used by Boykov et al. (2001), a piecewise smooth labeling f consistent with the observed data is sought, in order to minimize the energy function. According to f, a label fpAL, where L is a finite set of labels, is assigned to each pixel pAP. The energy functions E(f) considered for minimization via Graph Cuts are of the form X X Eðf Þ ¼ V p,q ðf p ,f q Þ þ Dp ðf p Þ: ð1Þ fp,qg A N
pAP
The function Dp(fp) typically corresponds to (fp Ip)2 and measures the disagreement between fp and the observed intensity Ip, for each pixel p. The function Vp,q(fp,fq) computes the cost of assigning fp and fq to the pair of pixels {p, q}AN, where the set N often comprises adjacent pixels, although not necessarily. During the evaluation of E(f), relatively high values of Vp,q indicate that f does not correspond to a piecewise smooth labeling. Min-cut/max-flow optimization problems may be formulated as follows. Given a weighted graph G/V,ES, where V is a collection of nodes (or vertices) connected by a collection of edges E, with two distinguishable nodes called terminals (or source and sink), the minimum cut problem is to find the cut C E that separates the terminal nodes, with minimum cost. The cost of the cut 9C9 corresponds to the sum of its edge weights. Boykov et al. (2001) proposed two algorithms for multidimensional energy minimization, achieving approximate solutions with guaranteed optimality bounds: the swap move algorithm and the expansion move algorithm. Unlike other algorithms such as simulated annealing, those algorithms provide solutions in a nearly linear time in practice. The library used in our implementations (gco-v3.0) is available for research purposes at http://vision.csd.uwo.ca/code/.
1286
A.A.M. Meneses et al. / Applied Radiation and Isotopes 70 (2012) 1284–1287
3. Results For each tomographic scan, 900 projections of the sample were acquired for equally spaced rotation angles and measurement time of about 1 s over a total rotation of 1801. The Tomo Project program, developed at SYRMEP (Montanari, 2003), was used for tomographic reconstruction. From each sample, a centered region of interest (ROI) was selected using the software ImageJ. Two-dimensional slices and three-dimensional reconstructed images obtained for dorsal portions of a given specimen (named E8D) are presented in Figs. 1 and 2, respectively. In the preprocessing phase, a median filter was used in order to reduce noise. It is well known that information carried by the phase of X-Ray wavefield, or phase effects, may be converted into image contrast, enhancing detail visibility, especially if compared to the poor contrast given by media with small X-ray absorption differences. Small amounts of marrow and cartilage, for quantification purposes and as depicted in Fig. 2, may be regarded as artifacts in the image and should be removed with the application of the EMvGC algorithm. Fig. 3 is the result of the application of the EMvGC algorithm to the image depicted in Fig. 2. Fig. 4 corresponds to the threedimensional visualization after the application of EMvGC to all
slices of the tomographic scan depicted in Fig. 1. In order to achieve those results in the tomographic scan segmentation, the EMvGC algorithm considered 3D information in our implementations, that is, besides the bi-dimensional information contained at each slice, the information of adjacent slices was also considered during the execution of the algorithm. The results of the EMvGC algorithm were also compared to the manual segmentation using the Dice Similarity Coefficient (Dice, 1945; Zijdenbos et al., 1994), for which the algorithm obtained the value of 99.94%, indicating that the automated segmentation using EMvGC was accurately performed, coping with the tasks of removing non-bone portions of the slices such as cartilage and marrow depicted in Fig. 2.
4. Conclusion The present article reports the application of EMvGC algorithm to SR-mCT biomedical images. EMvGC is a state-of-art algorithm, which is fast and accurate, with many applications in computer vision. We presented the results of the automated segmentation of a Wistar Rat rib scan, in which the algorithm achieved a remarkable performance, with a Dice Similarity Coefficient corresponding to 99.94%, in comparison to manual segmentation.
Acknowledgment We warmly thank Giuliana Tromba and Nicola Sodini from the SYRMEP beamline for their invaluable help during the experiment. The authors thank to the Brazilian agencies CNPq, CAPES, FAPESPA and FAPERJ for partial financial support. The images were obtained during a visit (R.C.B.) within the Associate Scheme of the Abdus Salam International Centre for Theoretical Physics (ICTP). The authors are thankful to Cherley Borba (LCR/UERJ) who made the samples available. References
Fig. 3. Result of the application of the EMvGC algorithm to the slice depicted in Fig. 2.
Fig. 4. Three-dimensinoal visualization of dorsal portion of Wistar rat rib bone (sample E8D) after the segmentation with EMvGC.
Abrami, A., Arfelli, F., Barroso, R.C., et al., 2005. Medical applications of synchrotron radiation at the SYRMEP beamline of ELETTRA. Nucl. Instrum. Methods A 548, 221–227. Boykov, Y., Veksler, O., Zabih, R., 2001. Fast approximate energy minimization via graph cuts. IEEE Trans. Pattern Anal. Mach. Intell. 23, 1222–1239. Boykov, Y., Kolmogorov, V., 2004. An experimental comparison of min-cut/maxflow algorithms for energy minimization in vision. IEEE Trans. Pattern Anal. Mach. Intell. 26, 1124–1137. Chappard, D., Basle´, M.-F., Legrand, E., Audran, M., 2008. Trabecular bone microarchitecture: a review. Morphologie 92, 162–170. Dice, L.R., 1945. Measures of the amount of ecologic association between species. Ecology 26, 297–302. Elder, F.R., Gurewitsch, A.M., Langmuir, R.V., Pollock, H.C., 1947. Radiation from electrons in a synchrotron. Phys. Rev. 71, 829–830. Greig, D.M., Porteous, B.T., Seheult, A.H., 1989. Exact maximum a posteriori estimation for binary images. J. Royal Stat. Soc. B 51, 271–279. Hounsfield, G.N., 1973. Computerized transverse axial scanning (tomography): Part 1. Description of system. Br. J. Radiol. 46, 1016–1022. Kolmogorov, V., Zabih, R., 2004. What energy functions can be minimized via graph cuts? IEEE Trans. Pattern Anal. Mach. Intell. 26, 147–159. ¨ Krebs, A., Graeff, C., Frieling, I., Kurz, B., Timm, W., Engelke, K., Gluer, C.-C., 2009. High resolution computed tomography of the vertebrae yields accurate information on trabecular distances if processed by 3D fuzzy segmentation techniques. Bone 44, 145–152. Lewis, R.A., 2004. Medical phase contrast x-ray imaging: current status and future prospects. Phys. Med. Biol. 49, 3573–3583. Meneses, A.A.M., Pinheiro, C.J.G., Schirru, R., Barroso, R.C., Braz, D., Oliveira, L.F., 2008. Artificial neural networks applied to bone recognition in X-ray computer microtomography imaging for histomorphometric analysis. In: Proceedings of the IEEE Nuclear Science Symposium and Medical Imaging Conference Record, pp. 5309–5313. Meneses, A.A.M., Pinheiro, C.J.G., Gambardella, L.M., Schirru, R., Barroso, R.C., Braz, D., Oliveira, L.F., 2009. Neural computing for quantitative analysis of human bone trabecular structures in synchrotron radiation X-ray mCT images.
A.A.M. Meneses et al. / Applied Radiation and Isotopes 70 (2012) 1284–1287
In: Proceedings of the IEEE Nuclear Science Symposium and Medical Imaging Conference Record, pp. 3437–3441. Meneses, A.A.M., Pinheiro, C.J.G., Rancoita, P., Schaul, T., Gambardella, L.M., Schirru, R., Barroso, R.C., Oliveira, L.F., 2010. Assessment of neural networks training strategies for histomorphometric analysis of synchrotron radiation medical images. Nucl. Instr. Methods A 621, 662–669. Montanari, F., 2003. SYRMEP TOMO Project, tutorial.
1287
Pettersson, N., Nyman, J., Johansson, K., 2009. Radiation-induced rib fractures after hypofractionated stereotactic body radiation therapy of non-small cell lung cancer: a dose- and volume-response analysis. Radiother. Oncol. 91, 360–368. Zijdenbos, A.P., Dawant, B.M., Margolin, R.A., Palmer, A.C., 1994. Morphometric analysis of white matter lesions in MR images: method and validation. IEEE Trans. Med. Imaging 13, 716–724.