- Email: [email protected]
3D identification of trabecular bone fracture zone using an automatic image registration scheme: A validation study
Journal of Biomechanics 45 (2012) 2035–2040
Contents lists available at SciVerse ScienceDirect
Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com
Short communication
3D identification of trabecular bone fracture zone using an automatic image registration scheme: A validation study Simone Tassani a,b,n, George K. Matsopoulos a, Fabio Baruffaldi b a b
Institute of Communication and Computer System, National Technical University of Athens, 9 Iroon Polytechniou Street, 157 80 Zografou, Athens, Greece Laboratorio di Tecnologia Medica, Istituto Ortopedico Rizzoli, Italy
a r t i c l e i n f o
abstract
Article history: Accepted 13 May 2012
Accurate identification of the local fracture zone is an important step towards the failure assessment of trabecular bone. In previous in-vitro studies, local fracture zones were visually identified in micro-CT images by experienced observers. This is a time-consuming and observer-dependent approach and it prevents any large-scale analysis of local trabecular fracture regions. The scope of this study is the application and validation of a new registration scheme for the automatic identification of trabecular bone fracture zones. Six human trabecular specimens were extracted from different anatomical sites. Five specimens were mechanically tested and scanned using micro-CT. For each specimen pre- and post-failure microCT datasets were obtained. The sixth specimen was scanned twice without any mechanical compression and was used to test the accuracy of the proposed scheme. The registration scheme was applied to the acquired datasets for the automatic identification of the fracture zone. The proposed scheme comprises of a three-dimensional (3D) automatic registration method to define the differences between the two datasets, and the application of a criterion for defining slices of the pre-failure dataset as ‘‘broken’’ or ‘‘unbroken’’. Identifications of the fracture zones were qualitatively validated against visual identification of observers. Furthermore, ‘‘full 3D’’ fracture zone identification, based on the presented scheme, was proposed. The proposed scheme proved to be more accurate and significantly faster than the currently used visual process. & 2012 Elsevier Ltd. All rights reserved.
Keywords: Trabecular fracture zone Image processing Three-dimensional automatic registration Fracture parameters Trabecular bone
1. Introduction Trabecular mechanical behaviour is an important issue for the assessment of bone fracture risk. Numerous studies analysing the mechanical behaviour of the trabecular bone structure have been reported in the literature (Goldstein et al., 1993; Goulet et al., 1994; Ciarelli et al., 2000; Matsuura et al., 2007; Perilli et al., 2008). Several parameters related to the identification and prediction of a fracture zone, and of the fracture event in general have been used in both clinical studies (Marshall et al., 1996; McCreadie and Goldstein, 2000; Donaldson et al., 2009; Watts et al., 2009) and invitro studies (Nazarian et al., 2006, 2008; Perilli et al., 2008; Tassani et al., 2010). In-vitro studies considered the local analysis of the trabecular structure and compared it to the global one, in order to identify the weakest point of the structure and, therefore, the fracture zone.
In the reported studies, fracture zones are identified by visual inspection of each micro-CT acquired slice of the specimens, often requiring a blind comparison among operators (Nazarian et al., 2006; Perilli et al., 2008). Consequently, the whole procedure is time-consuming and operator-dependent, and therefore not applicable to large scale analysis. Automatic techniques should be applied to identify fracture zones in different specimens subjected to mechanical testing, providing accuracy and reproducibility of the results and reduction of the execution time required. In the present study, a new methodological approach based on image processing is proposed and validated for the automatic identification of trabecular bone fracture zones in micro-CT datasets after mechanical testing.
2. Materials and methods 2.1. Data acquisition
n Corresponding author at: Institute of Communication and Computer System, National Technical University of Athens, 9 Iroon Polytechniou Street, 157 80 Zografou, Athens, Greece. Tel.: þ 30 210 7723577; fax: þ30 210 7723557. E-mail address: [email protected] (S. Tassani).
0021-9290/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jbiomech.2012.05.019
Six human trabecular bone specimens, cylindrical in shape (10 mm diameter, 26 mm height), were extracted from the epiphyses of two femurs and two tibiae (Fig. 1) and embalmed, according to procedures reported in the literature (Van Sint Jan and Rooze, 1992; Ohman et al., 2008; Tassani et al., 2011).
2036
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040 Five specimens underwent compressive testing (model Mini Bionix 858, MTS Systems Corp., Minneapolis, MN, USA) (Ohman et al., 2007; Perilli et al., 2008). All specimens were scanned using a micro-CT scanner (model SkyScan 1072, SkyScan, Kontich, Belgium) before and after (pre- and post-failure datasets) the mechanical test, using an isotropic voxel size of 19.5 mm (Perilli et al., 2007a, 2008; Tassani et al., 2011). The sixth specimen was used to perform an accuracy test. Thus, the specimen was scanned twice without mechanical compression. The two resulting datasets are hereafter called TestSet-I and TestSet-II.
2.2. The proposed registration scheme Automatic identification of the fracture zones was performed in a two-step approach: (a) Application of a 3D automatic registration method. (b) Identification of fracture zones.
2.2.1. 3D automatic registration method The proposed method is a surface-based registration technique (Matsopoulos et al., 2003; Matsopoulos, 2009). The method was applied on the pre- and postfailure datasets of every specimen in order to highlight the differences related to the compressive test. Analytically the registration method comprises the following steps: Step 1: Application of a segmentation process by means of a global fixed threshold (Perilli et al., 2006, 2007b). Step 2: Definition of a measure of match (MOM) that quantifies the spatial matching between the pre- and post-failure sets. Step 3: Maximization of the MOM. The geometrical transformation employed was the rigid transformation model (van den Elsen et al., 1993). The 3D automatic registration method was applied as follows. Two subsets of the post-failure set were initially defined: the upper and the lower subset, relative to the fracture zone, consisting of a maximum of 50 contiguous slices. The upper subset consisted of slices between the first upper slice of the set, and a randomly selected slice located above the fracture zone. A similar procedure was performed for defining the lower subset. Thus, the two subsets correspond to an ‘‘unbroken region’’. These subsets are shown in Fig. 2a (yellow areas). The proposed registration method was applied twice, for aligning both the upper and the lower post-failure subsets to the pre-failure set (Fig. 2b and c).
Fig. 1. Extraction sites of trabecular specimens for femur and tibia.
2.2.2. Criterion for the identification of the fracture zone The trabecular fracture zone was defined as the region presenting brittle fracture or plastic deformation of at least one trabecula. In order to automatically identify the fracture zone on the pre-failure set, each slice was classified as ‘‘broken’’ or ‘‘unbroken’’ according to the following methodology (Fig. 3). For each slice, ROIs of the pre-failure and registered post-failure dataset were obtained from the segmented images. Each ROI was selected by the identification of single disconnected objects. Therefore, every disconnected trabecula lying on the slice was identified as a ROI. If ROIs of the pre- and registered post-failure dataset had an overlap inferior of a threshold x%, the ROI was classified as broken. The Broken
Fig. 2. Application of the proposed 3D registration scheme. (a) The pre- and post-failure sets. Upper and lower subsets are identified on the post-failure set (yellow areas) and are registered on the corresponding areas of the pre-failure set (blue areas). (b) Application of the registration method involving the upper post-failure subset and the pre-failure set. (c) Application of the registration method involving the lower post-failure subset and the pre-failure set. (d) All the slices including the misaligned zone are classified as broken using the identification criterion and identified as ‘‘fracture zone’’. (e) The Full 3D fracture zone is identified as the VOI including only the misaligned zone (red area).
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
2037
Percentage of each slice was computed as in formula: P ROIs that classif ied as broken P % Broken Percentage ¼ all ROIs Finally a median filter with a width of about 0.5 mm, 25 slices, was applied to the Broken Percentage distribution along the z-axis. The filtering procedure was introduced to remove the slices incoherently classified due to noise within the images: i.e. a slice identified as broken without having any other neighbouring slices broken. The filter is about half of the trabecular unit (sum of trabecular thickness and separation, Ulrich et al., 1999; Nazarian et al., 2006). If the Broken Percentage was different from zero, the whole slice was classified as broken. Slices identified as broken formed a fracture zone (Fig. 2d). The result of the Broken Percentage criterion was plotted along the ‘‘z’’ axis, showing the different broken degrees for every slice of the fracture zone. The threshold x% was determined by minimizing the root mean square error (RMSE) between the values identified by the automatic identification and those of the visual identification (later described). sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P 2 i ðaiðxÞ vi Þ RMSEðxÞ ¼ n where ‘ai(x)’ are the values suggested by the automatic identification for the selected x, ‘vi’ are those visually identified and ‘n’ is the total number of analysed cases. Finally, a ‘‘full 3D’’ definition of the fracture zone was obtained starting from the 3D distribution of ROIs classified as broken by the proposed criterion. A 3D Volume of Interest (VOI) is required for the calculation of morphometrical parameters. The 3D VOI was identified applying a morphological 3D dilation procedure around the broken ROIs (Serra, 1982). Each ROI was dilated in every direction by 25 pixels (about 0.50 mm radius, Nazarian et al., 2006), thereby obtaining for every ROI an ellipsoidal VOI centred on the broken trabecula. When ROIs were close enough, the VOIs fused creating a single 3D VOI (Fig. 2e). 2.3. Validation of the proposed registration scheme 2.3.1. Application on the TestSets The proposed registration scheme was applied on TestSet-I and TestSet-II in order to evaluate its accuracy in identifying fracture zones. 2.3.2. Visual inspection for the identification of the fracture zone A visual comparison was performed between the pre- and post-failure datasets for the identification of the fracture zones. These regions were identified visually, by comparing the stack of pre-failure with post-failure micro-CT, using the pre-failure cross sections as reference. Three operators performed this procedure independently on each specimen, and the pooled regions were considered as the final fracture zones of the specimen. Operators were asked to identify as the first and last broken slices the first and the last in which even a single broken trabecula was visible (Perilli et al., 2008). Visual inspection is by definition an operator-dependent procedure, and therefore disagreement on a few slices among operators is unavoidable. The pooled value of each first and last broken slice was used for the calculation of the RMSE. Moreover, the mean difference among operators was used as threshold in the validation against the proposed automatic registration scheme.
3. Results
Fig. 3. Flow chart of the identification of the fracture zone. Abbreviations: ROI: Region Of Interest. Overlap%: percentage of overlap of the current ROI with the ROI of the registered post-failure corresponding slice. x%: the value for which the root mean square error between the values identified by the automatic identification and those of the visual identification is minimized. Broken %: the ratio of the ROIs classified as broken, over the total number of ROIs for each slice.
The accuracy of the registration scheme was verified by its application on the TestSets. After application of the proposed registration scheme, the two datasets completely matched and no fracture zones were identified. The mean registration error was within the range of 0.002% (number of misaligned pixels over total number of pixels). Based on the analysis of the pre- and post-failure datasets (Fig. 4a), the fracture zone was visually identified and compared to the broken percentage distribution obtained by the proposed scheme (Fig. 4b). The ‘‘full 3D’’ broken region was also identified by means of the dilation procedure (Fig. 4c). The three operators reported a mean variation of 18 slices (about 350 mm) identifying the start and end of the fracture zones. This value was used as threshold for the validation of the registration scheme. The minimum RMSE (RMSE¼41) was found for a value of x%¼25% (Fig. 5a). The value was greater than the classification disagreement (18). The visual and automatic procedures were in agreement in seven out of ten cases (5 first and 5 last broken slices,
2038
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
Fig. 4. The complete identification process. (a) Micro-CT pre- and post-failure datasets. (b) The visual procedure is compared to the automatic registration scheme for fracture zone identification. On the vertical axis of the plot, the number of slices of the pre-failure dataset is displayed whereas the broken percentage of all ROIs for each slice, as obtained by the proposed registration scheme, is shown on the horizontal axis. (c) Finally a ‘‘full 3D’’ fracture zone is identified.
Fig. 5. Trend of RMSE as a function of the classification threshold x%, before (a) and after (b) the correction of the visual identification.
two measurements for every specimen). For the remaining three cases, the operators repeated their visual inspection. For all three cases, the operators found their initial findings inaccurate and they corrected them, suggesting the successful performance of the proposed automatic registration scheme. Taking into consideration the corrections from the three cases, the RMSE was re-calculated. The minimum RMSE (RMSE¼8) was found for a value of x%¼30% (Fig. 5b). The mean execution time for the application of the proposed registration procedure was two minutes (Pentium 4, 3.2 GHz, 2 GB RAM was used in this study). The final result was a substantial reduction of time for the identification of the fracture zone, compared with the 20–30 min required for the visual approach by each observer. In addition, the proposed scheme allows the automatic identification of the fracture zone in 3D.
4. Discussion The aim of the study was to present and validate a novel scheme for automatic identification of the bone fracture zone in trabecular specimens. The technique was validated and compared to a visual
approach, performed by three experienced operators. Moreover, a ‘‘full 3D’’ automatic identification of the fracture zone was introduced. A novel application of a 3D image registration scheme to identify the fracture zone in in-vitro data acquired using a micro-CT scanner was presented. Many image registration methods have been applied to micro-CT datasets. (Waarsing et al., 2004; Boyd et al., 2006; Badea et al., 2008; Hulme et al., 2008; Hardisty and Whyne, 2009; Baiker et al., 2010; Lee et al., 2010; Nishiyama et al., 2010; Xiao et al., 2010; Nagaraja et al., 2011), but to the authors knowledge no methods for the identification of a discontinuity of the trabecular structure are presented in the literature. In Hulme et al. (2008), a registration method was used for recording vertebral endplate deformations of functional spine units, under axial compression. However, the two applications are significantly different, since in Hulme et al. external markers were required for registration of the data, while the scheme proposed here is automatic and permits the study of a discontinuous region, such as the zone of trabecular fracture. No fracture zones were identified in the accuracy test and a small registration error was reported (0.002% registration error), probably due to partial volume effects and segmentation problems (Hara et al., 2002). Comparable results had been shown between the visual approach and the automatic registration scheme. Results confirmed
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
by the morphometrical analysis reported in Appendix A. Automatic registration exhibits superior reliability, based on the refined disagreement measurements presented. In addition, the automatic approach reduced the execution time and allowed for three-dimensional identification and representation of the fracture region. Nonetheless, the morphometrical findings should be verified in a larger scale analysis. A limitation of the registration scheme is that unbroken upper and lower subsets are required. Therefore, the fracture zone must not pass through the edges of the specimen. However, this situation is often avoided in in-vitro mechanical tests by the use of cement endcaps. Moreover, the proposed scheme can be applied to specimens where one of the two extremities is completely broken, by registering the opposite side only. The proposed automatic scheme allows the identification of the fracture region and its structural characteristics, thereby giving the opportunity to improve the knowledge about the fracture of trabecular bone. This knowledge may eventually lead to the prediction of the failure event.
Conflict of interest statement None of the authors have any financial or personal relationship with other people or organizations that could have inappropriately influenced this study.
Acknowledgements The authors would like to thank Luigi Lena for the illustrations, ¨ hman and Nikolaos Mouravand Francesca Particelli, Caroline O liansky for the technical support during the experiments. The datasets were available on http://www.physiomespace.com, and produced by Laboratorio di Tecnologia Medica, Istituto Ortopedico Rizzoli, Bologna, Italy, with the financial support of the EU Project Living Human Digital Library (LHDL IST-2004-026932). This work was partially supported by a Marie Curie Intra European Fellowship within the 7th European Community Framework Programme (project number: PIEF-GA-2009-253924; acronym: MOSAIC).
Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jbiomech. 2012.05.019.
References Badea, C.T., Schreibmann, E., Fox, T., 2008. A registration based approach for 4D cardiac micro-CT using combined prospective and retrospective gating. Medical Physics 35, 1170–1179. Baiker, M., Milles, J., Dijkstra, J., Henning, T.D., Weber, A.W., Que, I., Kaijzel, E.L., Lowik, C.W., Reiber, J.H., Lelieveldt, B.P., 2010. Atlas-based whole-body segmentation of mice from low-contrast Micro-CT data. Medical Image Analysis 14, 723–737. Boyd, S.K., Moser, S., Kuhn, M., Klinck, R.J., Krauze, P.L., Muller, R., Gasser, J.A., 2006. Evaluation of three-dimensional image registration methodologies for in vivo micro-computed tomography. Annals of Biomedical Engineering 34, 1587–1599. Ciarelli, T.E., Fyhrie, D.P., Schaffler, M.B., Goldstein, S.A., 2000. Variations in threedimensional cancellous bone architecture of the proximal femur in female hip fractures and in controls. Journal of Bone and Mineral Research 15, 32–40. Donaldson, M.G., Palermo, L., Schousboe, J.T., Ensrud, K.E., Hochberg, M.C., Cummings, S.R., 2009. FRAX and risk of vertebral fractures: the Fracture Intervention Trial (FIT). Journal of Bone and Mineral Research.
2039
Goldstein, S.A., Goulet, R., McCubbrey, D., 1993. Measurement and significance of three-dimensional architecture to the mechanical integrity of trabecular bone. Calcified Tissue International 53 (Suppl. 1), S127–S132, discussion S132–133. Goulet, R.W., Goldstein, S.A., Ciarelli, M.J., Kuhn, J.L., Brown, M.B., Feldkamp, L.A., 1994. The relationship between the structural and orthogonal compressive properties of trabecular bone. Journal of Biomechanics 27, 375–389. Hara, T., Tanck, E., Homminga, J., Huiskes, R., 2002. The influence of microcomputed tomography threshold variations on the assessment of structural and mechanical trabecular bone properties. Bone 31, 107–109. Hardisty, M.R., Whyne, C.M., 2009. Whole bone strain quantification by image registration: a validation study. Journal of Biomechanical Engineering 131, 064502. Hulme, P.A., Ferguson, S.J., Boyd, S.K., 2008. Determination of vertebral endplate deformation under load using micro-computed tomography. Journal of Biomechanics 41, 78–85. Lee, C.F., Li, G.J., Wan, S.Y., Lee, W.J., Tzen, K.Y., Chen, C.H., Song, Y.L., Chou, Y.F., Chen, Y.S., Liu, T.C., 2010. Registration of micro-computed tomography and histological images of the guinea pig cochlea to construct an ear model using an iterative closest point algorithm. Annals of Biomedical Engineering 38, 1719–1727. Marshall, D., Johnell, O., Wedel, H., 1996. Meta-analysis of how well measures of bone mineral density predict occurrence of osteoporotic fractures. British Medical Journal 312, 1254–1259. Matsopoulos, G.K., 2009. Medical image registration and fusion techniques: a review. In: Stergiopoulos, S. (Ed.), Advanced Signal Processing Handbook—Theory and Implementation for Radar, Sonar, and Medical Imaging Real-Time Systems. CRC Press, Taylor & Francis, Inc., Florida, pp. 148–221. Matsopoulos, G.K., Delibasis, K.K., Mouravliansky, N.A., Asvestas, P.A., Nikita, K.S., Kouloulias, V.E., Uzunoglu, N.K., 2003. CT-MRI automatic surface-based registration schemes combining global and local optimization techniques. Technology and Health Care 11, 219–232. Matsuura, M., Eckstein, F., Lochmuller, E.M., Zysset, P.K., 2007. The role of fabric in the quasi-static compressive mechanical properties of human trabecular bone from various anatomical locations. Biomechanics and Modeling in Mechanobiology. McCreadie, B.R., Goldstein, S.A., 2000. Biomechanics of fracture: is bone mineral density sufficient to assess risk? Journal of Bone and Mineral Research 15, 2305–2308. Nagaraja, S., Skrinjar, O., Guldberg, R.E., 2011. Spatial correlations of trabecular bone microdamage with local stresses and strains using rigid image registration. Journal of Biomechanical Engineering 133, 064502. Nazarian, A., Stauber, M., Zurakowski, D., Snyder, B.D., Muller, R., 2006. The interaction of microstructure and volume fraction in predicting failure in cancellous bone. Bone 39, 1196–1202. Nazarian, A., von Stechow, D., Zurakowski, D., Muller, R., Snyder, B.D., 2008. Bone volume fraction explains the variation in strength and stiffness of cancellous bone affected by metastatic cancer and osteoporosis. Calcified Tissue International 83, 368–379. Nishiyama, K.K., Campbell, G.M., Klinck, R.J., Boyd, S.K., 2010. Reproducibility of bone micro-architecture measurements in rodents by in vivo micro-computed tomography is maximized with three-dimensional image registration. Bone 46, 155–161. Ohman, C., Baleani, M., Perilli, E., Dall’Ara, E., Tassani, S., Baruffaldi, F., Viceconti, M., 2007. Mechanical testing of cancellous bone from the femoral head: experimental errors due to off-axis measurements. Journal of Biomechanics 40, 2426–2433. Ohman, C., Dall’Ara, E., Baleani, M., Van Sint Jan, S., Viceconti, M., 2008. The effects of embalming using a 4% formalin solution on the compressive mechanical properties of human cortical bone. Clinical Biomechanics (Bristol, Avon) 23, 1294–1298. Perilli, E., Baleani, M., Ohman, C., Baruffaldi, F., Viceconti, M., 2007a. Structural parameters and mechanical strength of cancellous bone in the femoral head in osteoarthritis do not depend on age. Bone 41, 760–768. Perilli, E., Baleani, M., Ohman, C., Fognani, R., Baruffaldi, F., Viceconti, M., 2008. Dependence of mechanical compressive strength on local variations in microarchitecture in cancellous bone of proximal human femur. Journal of Biomechanics 41, 438–446. Perilli, E., Baruffaldi, F., Bisi, M.C., Cristofolini, L., Cappello, A., 2006. A physical phantom for the calibration of three-dimensional X-ray microtomography examination. Journal of Microscopy 222, 124–134. Perilli, E., Baruffaldi, F., Visentin, M., Bordini, B., Traina, F., Cappello, A., Viceconti, M., 2007. MicroCT examination of human bone specimens: effects of polymethylmethacrylate embedding on structural parameters. Journal of Microscopy 225, 192–200. Serra, J., 1982. Image Analysis and Mathematical Morphology. Academic Press, London. Tassani, S., Ohman, C., Baleani, M., Baruffaldi, F., Viceconti, M., 2010. Anisotropy and inhomogeneity of the trabecular structure can describe the mechanical strength of osteoarthritic cancellous bone. Journal of Biomechanics 43, 1160–1166. Tassani, S., Ohman, C., Baruffaldi, F., Baleani, M., Viceconti, M., 2011. Volume to density relation in adult human bone tissue. Journal of Biomechanics 44, 103–108. Ulrich, D., van Rietbergen, B., Laib, A., Ruegsegger, P., 1999. The ability of threedimensional structural indices to reflect mechanical aspects of trabecular bone. Bone 25, 55–60. van den Elsen, P.A., Pol, E.J.D., Viergever, M.A., 1993. Medical image matching—a review with classification. IEEE Engineering in Medicine and Biology Magazine 12, 26–39.
2040
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
Van Sint Jan, S., Rooze, M., 1992. The thenar muscles: new findings. Surgical— Radiologic Anatomy 14, 325–329. Waarsing, J.H., Day, J.S., van der Linden, J.C., Ederveen, A.G., Spanjers, C., De Clerck, N., Sasov, A., Verhaar, J.A., Weinans, H., 2004. Detecting and tracking local changes in the tibiae of individual rats: a novel method to analyse longitudinal in vivo micro-CT data. Bone 34, 163–169.
Watts, N.B., Ettinger, B., LeBoff, M.S., 2009. FRAX facts. Journal of Bone and Mineral Research 24, 975–979. Xiao, D., Zahra, D., Bourgeat, P., Berghofer, P., Tamayo, O.A., Wimberley, C., Gregoire, M.C., Salvado, O., 2010. An improved 3D shape context based nonrigid registration method and its application to small animal skeletons registration. Computerized Medical Imaging and Graphics 34, 321–332.
Contents lists available at SciVerse ScienceDirect
Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com
Short communication
3D identification of trabecular bone fracture zone using an automatic image registration scheme: A validation study Simone Tassani a,b,n, George K. Matsopoulos a, Fabio Baruffaldi b a b
Institute of Communication and Computer System, National Technical University of Athens, 9 Iroon Polytechniou Street, 157 80 Zografou, Athens, Greece Laboratorio di Tecnologia Medica, Istituto Ortopedico Rizzoli, Italy
a r t i c l e i n f o
abstract
Article history: Accepted 13 May 2012
Accurate identification of the local fracture zone is an important step towards the failure assessment of trabecular bone. In previous in-vitro studies, local fracture zones were visually identified in micro-CT images by experienced observers. This is a time-consuming and observer-dependent approach and it prevents any large-scale analysis of local trabecular fracture regions. The scope of this study is the application and validation of a new registration scheme for the automatic identification of trabecular bone fracture zones. Six human trabecular specimens were extracted from different anatomical sites. Five specimens were mechanically tested and scanned using micro-CT. For each specimen pre- and post-failure microCT datasets were obtained. The sixth specimen was scanned twice without any mechanical compression and was used to test the accuracy of the proposed scheme. The registration scheme was applied to the acquired datasets for the automatic identification of the fracture zone. The proposed scheme comprises of a three-dimensional (3D) automatic registration method to define the differences between the two datasets, and the application of a criterion for defining slices of the pre-failure dataset as ‘‘broken’’ or ‘‘unbroken’’. Identifications of the fracture zones were qualitatively validated against visual identification of observers. Furthermore, ‘‘full 3D’’ fracture zone identification, based on the presented scheme, was proposed. The proposed scheme proved to be more accurate and significantly faster than the currently used visual process. & 2012 Elsevier Ltd. All rights reserved.
Keywords: Trabecular fracture zone Image processing Three-dimensional automatic registration Fracture parameters Trabecular bone
1. Introduction Trabecular mechanical behaviour is an important issue for the assessment of bone fracture risk. Numerous studies analysing the mechanical behaviour of the trabecular bone structure have been reported in the literature (Goldstein et al., 1993; Goulet et al., 1994; Ciarelli et al., 2000; Matsuura et al., 2007; Perilli et al., 2008). Several parameters related to the identification and prediction of a fracture zone, and of the fracture event in general have been used in both clinical studies (Marshall et al., 1996; McCreadie and Goldstein, 2000; Donaldson et al., 2009; Watts et al., 2009) and invitro studies (Nazarian et al., 2006, 2008; Perilli et al., 2008; Tassani et al., 2010). In-vitro studies considered the local analysis of the trabecular structure and compared it to the global one, in order to identify the weakest point of the structure and, therefore, the fracture zone.
In the reported studies, fracture zones are identified by visual inspection of each micro-CT acquired slice of the specimens, often requiring a blind comparison among operators (Nazarian et al., 2006; Perilli et al., 2008). Consequently, the whole procedure is time-consuming and operator-dependent, and therefore not applicable to large scale analysis. Automatic techniques should be applied to identify fracture zones in different specimens subjected to mechanical testing, providing accuracy and reproducibility of the results and reduction of the execution time required. In the present study, a new methodological approach based on image processing is proposed and validated for the automatic identification of trabecular bone fracture zones in micro-CT datasets after mechanical testing.
2. Materials and methods 2.1. Data acquisition
n Corresponding author at: Institute of Communication and Computer System, National Technical University of Athens, 9 Iroon Polytechniou Street, 157 80 Zografou, Athens, Greece. Tel.: þ 30 210 7723577; fax: þ30 210 7723557. E-mail address: [email protected] (S. Tassani).
0021-9290/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jbiomech.2012.05.019
Six human trabecular bone specimens, cylindrical in shape (10 mm diameter, 26 mm height), were extracted from the epiphyses of two femurs and two tibiae (Fig. 1) and embalmed, according to procedures reported in the literature (Van Sint Jan and Rooze, 1992; Ohman et al., 2008; Tassani et al., 2011).
2036
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040 Five specimens underwent compressive testing (model Mini Bionix 858, MTS Systems Corp., Minneapolis, MN, USA) (Ohman et al., 2007; Perilli et al., 2008). All specimens were scanned using a micro-CT scanner (model SkyScan 1072, SkyScan, Kontich, Belgium) before and after (pre- and post-failure datasets) the mechanical test, using an isotropic voxel size of 19.5 mm (Perilli et al., 2007a, 2008; Tassani et al., 2011). The sixth specimen was used to perform an accuracy test. Thus, the specimen was scanned twice without mechanical compression. The two resulting datasets are hereafter called TestSet-I and TestSet-II.
2.2. The proposed registration scheme Automatic identification of the fracture zones was performed in a two-step approach: (a) Application of a 3D automatic registration method. (b) Identification of fracture zones.
2.2.1. 3D automatic registration method The proposed method is a surface-based registration technique (Matsopoulos et al., 2003; Matsopoulos, 2009). The method was applied on the pre- and postfailure datasets of every specimen in order to highlight the differences related to the compressive test. Analytically the registration method comprises the following steps: Step 1: Application of a segmentation process by means of a global fixed threshold (Perilli et al., 2006, 2007b). Step 2: Definition of a measure of match (MOM) that quantifies the spatial matching between the pre- and post-failure sets. Step 3: Maximization of the MOM. The geometrical transformation employed was the rigid transformation model (van den Elsen et al., 1993). The 3D automatic registration method was applied as follows. Two subsets of the post-failure set were initially defined: the upper and the lower subset, relative to the fracture zone, consisting of a maximum of 50 contiguous slices. The upper subset consisted of slices between the first upper slice of the set, and a randomly selected slice located above the fracture zone. A similar procedure was performed for defining the lower subset. Thus, the two subsets correspond to an ‘‘unbroken region’’. These subsets are shown in Fig. 2a (yellow areas). The proposed registration method was applied twice, for aligning both the upper and the lower post-failure subsets to the pre-failure set (Fig. 2b and c).
Fig. 1. Extraction sites of trabecular specimens for femur and tibia.
2.2.2. Criterion for the identification of the fracture zone The trabecular fracture zone was defined as the region presenting brittle fracture or plastic deformation of at least one trabecula. In order to automatically identify the fracture zone on the pre-failure set, each slice was classified as ‘‘broken’’ or ‘‘unbroken’’ according to the following methodology (Fig. 3). For each slice, ROIs of the pre-failure and registered post-failure dataset were obtained from the segmented images. Each ROI was selected by the identification of single disconnected objects. Therefore, every disconnected trabecula lying on the slice was identified as a ROI. If ROIs of the pre- and registered post-failure dataset had an overlap inferior of a threshold x%, the ROI was classified as broken. The Broken
Fig. 2. Application of the proposed 3D registration scheme. (a) The pre- and post-failure sets. Upper and lower subsets are identified on the post-failure set (yellow areas) and are registered on the corresponding areas of the pre-failure set (blue areas). (b) Application of the registration method involving the upper post-failure subset and the pre-failure set. (c) Application of the registration method involving the lower post-failure subset and the pre-failure set. (d) All the slices including the misaligned zone are classified as broken using the identification criterion and identified as ‘‘fracture zone’’. (e) The Full 3D fracture zone is identified as the VOI including only the misaligned zone (red area).
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
2037
Percentage of each slice was computed as in formula: P ROIs that classif ied as broken P % Broken Percentage ¼ all ROIs Finally a median filter with a width of about 0.5 mm, 25 slices, was applied to the Broken Percentage distribution along the z-axis. The filtering procedure was introduced to remove the slices incoherently classified due to noise within the images: i.e. a slice identified as broken without having any other neighbouring slices broken. The filter is about half of the trabecular unit (sum of trabecular thickness and separation, Ulrich et al., 1999; Nazarian et al., 2006). If the Broken Percentage was different from zero, the whole slice was classified as broken. Slices identified as broken formed a fracture zone (Fig. 2d). The result of the Broken Percentage criterion was plotted along the ‘‘z’’ axis, showing the different broken degrees for every slice of the fracture zone. The threshold x% was determined by minimizing the root mean square error (RMSE) between the values identified by the automatic identification and those of the visual identification (later described). sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P 2 i ðaiðxÞ vi Þ RMSEðxÞ ¼ n where ‘ai(x)’ are the values suggested by the automatic identification for the selected x, ‘vi’ are those visually identified and ‘n’ is the total number of analysed cases. Finally, a ‘‘full 3D’’ definition of the fracture zone was obtained starting from the 3D distribution of ROIs classified as broken by the proposed criterion. A 3D Volume of Interest (VOI) is required for the calculation of morphometrical parameters. The 3D VOI was identified applying a morphological 3D dilation procedure around the broken ROIs (Serra, 1982). Each ROI was dilated in every direction by 25 pixels (about 0.50 mm radius, Nazarian et al., 2006), thereby obtaining for every ROI an ellipsoidal VOI centred on the broken trabecula. When ROIs were close enough, the VOIs fused creating a single 3D VOI (Fig. 2e). 2.3. Validation of the proposed registration scheme 2.3.1. Application on the TestSets The proposed registration scheme was applied on TestSet-I and TestSet-II in order to evaluate its accuracy in identifying fracture zones. 2.3.2. Visual inspection for the identification of the fracture zone A visual comparison was performed between the pre- and post-failure datasets for the identification of the fracture zones. These regions were identified visually, by comparing the stack of pre-failure with post-failure micro-CT, using the pre-failure cross sections as reference. Three operators performed this procedure independently on each specimen, and the pooled regions were considered as the final fracture zones of the specimen. Operators were asked to identify as the first and last broken slices the first and the last in which even a single broken trabecula was visible (Perilli et al., 2008). Visual inspection is by definition an operator-dependent procedure, and therefore disagreement on a few slices among operators is unavoidable. The pooled value of each first and last broken slice was used for the calculation of the RMSE. Moreover, the mean difference among operators was used as threshold in the validation against the proposed automatic registration scheme.
3. Results
Fig. 3. Flow chart of the identification of the fracture zone. Abbreviations: ROI: Region Of Interest. Overlap%: percentage of overlap of the current ROI with the ROI of the registered post-failure corresponding slice. x%: the value for which the root mean square error between the values identified by the automatic identification and those of the visual identification is minimized. Broken %: the ratio of the ROIs classified as broken, over the total number of ROIs for each slice.
The accuracy of the registration scheme was verified by its application on the TestSets. After application of the proposed registration scheme, the two datasets completely matched and no fracture zones were identified. The mean registration error was within the range of 0.002% (number of misaligned pixels over total number of pixels). Based on the analysis of the pre- and post-failure datasets (Fig. 4a), the fracture zone was visually identified and compared to the broken percentage distribution obtained by the proposed scheme (Fig. 4b). The ‘‘full 3D’’ broken region was also identified by means of the dilation procedure (Fig. 4c). The three operators reported a mean variation of 18 slices (about 350 mm) identifying the start and end of the fracture zones. This value was used as threshold for the validation of the registration scheme. The minimum RMSE (RMSE¼41) was found for a value of x%¼25% (Fig. 5a). The value was greater than the classification disagreement (18). The visual and automatic procedures were in agreement in seven out of ten cases (5 first and 5 last broken slices,
2038
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
Fig. 4. The complete identification process. (a) Micro-CT pre- and post-failure datasets. (b) The visual procedure is compared to the automatic registration scheme for fracture zone identification. On the vertical axis of the plot, the number of slices of the pre-failure dataset is displayed whereas the broken percentage of all ROIs for each slice, as obtained by the proposed registration scheme, is shown on the horizontal axis. (c) Finally a ‘‘full 3D’’ fracture zone is identified.
Fig. 5. Trend of RMSE as a function of the classification threshold x%, before (a) and after (b) the correction of the visual identification.
two measurements for every specimen). For the remaining three cases, the operators repeated their visual inspection. For all three cases, the operators found their initial findings inaccurate and they corrected them, suggesting the successful performance of the proposed automatic registration scheme. Taking into consideration the corrections from the three cases, the RMSE was re-calculated. The minimum RMSE (RMSE¼8) was found for a value of x%¼30% (Fig. 5b). The mean execution time for the application of the proposed registration procedure was two minutes (Pentium 4, 3.2 GHz, 2 GB RAM was used in this study). The final result was a substantial reduction of time for the identification of the fracture zone, compared with the 20–30 min required for the visual approach by each observer. In addition, the proposed scheme allows the automatic identification of the fracture zone in 3D.
4. Discussion The aim of the study was to present and validate a novel scheme for automatic identification of the bone fracture zone in trabecular specimens. The technique was validated and compared to a visual
approach, performed by three experienced operators. Moreover, a ‘‘full 3D’’ automatic identification of the fracture zone was introduced. A novel application of a 3D image registration scheme to identify the fracture zone in in-vitro data acquired using a micro-CT scanner was presented. Many image registration methods have been applied to micro-CT datasets. (Waarsing et al., 2004; Boyd et al., 2006; Badea et al., 2008; Hulme et al., 2008; Hardisty and Whyne, 2009; Baiker et al., 2010; Lee et al., 2010; Nishiyama et al., 2010; Xiao et al., 2010; Nagaraja et al., 2011), but to the authors knowledge no methods for the identification of a discontinuity of the trabecular structure are presented in the literature. In Hulme et al. (2008), a registration method was used for recording vertebral endplate deformations of functional spine units, under axial compression. However, the two applications are significantly different, since in Hulme et al. external markers were required for registration of the data, while the scheme proposed here is automatic and permits the study of a discontinuous region, such as the zone of trabecular fracture. No fracture zones were identified in the accuracy test and a small registration error was reported (0.002% registration error), probably due to partial volume effects and segmentation problems (Hara et al., 2002). Comparable results had been shown between the visual approach and the automatic registration scheme. Results confirmed
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
by the morphometrical analysis reported in Appendix A. Automatic registration exhibits superior reliability, based on the refined disagreement measurements presented. In addition, the automatic approach reduced the execution time and allowed for three-dimensional identification and representation of the fracture region. Nonetheless, the morphometrical findings should be verified in a larger scale analysis. A limitation of the registration scheme is that unbroken upper and lower subsets are required. Therefore, the fracture zone must not pass through the edges of the specimen. However, this situation is often avoided in in-vitro mechanical tests by the use of cement endcaps. Moreover, the proposed scheme can be applied to specimens where one of the two extremities is completely broken, by registering the opposite side only. The proposed automatic scheme allows the identification of the fracture region and its structural characteristics, thereby giving the opportunity to improve the knowledge about the fracture of trabecular bone. This knowledge may eventually lead to the prediction of the failure event.
Conflict of interest statement None of the authors have any financial or personal relationship with other people or organizations that could have inappropriately influenced this study.
Acknowledgements The authors would like to thank Luigi Lena for the illustrations, ¨ hman and Nikolaos Mouravand Francesca Particelli, Caroline O liansky for the technical support during the experiments. The datasets were available on http://www.physiomespace.com, and produced by Laboratorio di Tecnologia Medica, Istituto Ortopedico Rizzoli, Bologna, Italy, with the financial support of the EU Project Living Human Digital Library (LHDL IST-2004-026932). This work was partially supported by a Marie Curie Intra European Fellowship within the 7th European Community Framework Programme (project number: PIEF-GA-2009-253924; acronym: MOSAIC).
Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jbiomech. 2012.05.019.
References Badea, C.T., Schreibmann, E., Fox, T., 2008. A registration based approach for 4D cardiac micro-CT using combined prospective and retrospective gating. Medical Physics 35, 1170–1179. Baiker, M., Milles, J., Dijkstra, J., Henning, T.D., Weber, A.W., Que, I., Kaijzel, E.L., Lowik, C.W., Reiber, J.H., Lelieveldt, B.P., 2010. Atlas-based whole-body segmentation of mice from low-contrast Micro-CT data. Medical Image Analysis 14, 723–737. Boyd, S.K., Moser, S., Kuhn, M., Klinck, R.J., Krauze, P.L., Muller, R., Gasser, J.A., 2006. Evaluation of three-dimensional image registration methodologies for in vivo micro-computed tomography. Annals of Biomedical Engineering 34, 1587–1599. Ciarelli, T.E., Fyhrie, D.P., Schaffler, M.B., Goldstein, S.A., 2000. Variations in threedimensional cancellous bone architecture of the proximal femur in female hip fractures and in controls. Journal of Bone and Mineral Research 15, 32–40. Donaldson, M.G., Palermo, L., Schousboe, J.T., Ensrud, K.E., Hochberg, M.C., Cummings, S.R., 2009. FRAX and risk of vertebral fractures: the Fracture Intervention Trial (FIT). Journal of Bone and Mineral Research.
2039
Goldstein, S.A., Goulet, R., McCubbrey, D., 1993. Measurement and significance of three-dimensional architecture to the mechanical integrity of trabecular bone. Calcified Tissue International 53 (Suppl. 1), S127–S132, discussion S132–133. Goulet, R.W., Goldstein, S.A., Ciarelli, M.J., Kuhn, J.L., Brown, M.B., Feldkamp, L.A., 1994. The relationship between the structural and orthogonal compressive properties of trabecular bone. Journal of Biomechanics 27, 375–389. Hara, T., Tanck, E., Homminga, J., Huiskes, R., 2002. The influence of microcomputed tomography threshold variations on the assessment of structural and mechanical trabecular bone properties. Bone 31, 107–109. Hardisty, M.R., Whyne, C.M., 2009. Whole bone strain quantification by image registration: a validation study. Journal of Biomechanical Engineering 131, 064502. Hulme, P.A., Ferguson, S.J., Boyd, S.K., 2008. Determination of vertebral endplate deformation under load using micro-computed tomography. Journal of Biomechanics 41, 78–85. Lee, C.F., Li, G.J., Wan, S.Y., Lee, W.J., Tzen, K.Y., Chen, C.H., Song, Y.L., Chou, Y.F., Chen, Y.S., Liu, T.C., 2010. Registration of micro-computed tomography and histological images of the guinea pig cochlea to construct an ear model using an iterative closest point algorithm. Annals of Biomedical Engineering 38, 1719–1727. Marshall, D., Johnell, O., Wedel, H., 1996. Meta-analysis of how well measures of bone mineral density predict occurrence of osteoporotic fractures. British Medical Journal 312, 1254–1259. Matsopoulos, G.K., 2009. Medical image registration and fusion techniques: a review. In: Stergiopoulos, S. (Ed.), Advanced Signal Processing Handbook—Theory and Implementation for Radar, Sonar, and Medical Imaging Real-Time Systems. CRC Press, Taylor & Francis, Inc., Florida, pp. 148–221. Matsopoulos, G.K., Delibasis, K.K., Mouravliansky, N.A., Asvestas, P.A., Nikita, K.S., Kouloulias, V.E., Uzunoglu, N.K., 2003. CT-MRI automatic surface-based registration schemes combining global and local optimization techniques. Technology and Health Care 11, 219–232. Matsuura, M., Eckstein, F., Lochmuller, E.M., Zysset, P.K., 2007. The role of fabric in the quasi-static compressive mechanical properties of human trabecular bone from various anatomical locations. Biomechanics and Modeling in Mechanobiology. McCreadie, B.R., Goldstein, S.A., 2000. Biomechanics of fracture: is bone mineral density sufficient to assess risk? Journal of Bone and Mineral Research 15, 2305–2308. Nagaraja, S., Skrinjar, O., Guldberg, R.E., 2011. Spatial correlations of trabecular bone microdamage with local stresses and strains using rigid image registration. Journal of Biomechanical Engineering 133, 064502. Nazarian, A., Stauber, M., Zurakowski, D., Snyder, B.D., Muller, R., 2006. The interaction of microstructure and volume fraction in predicting failure in cancellous bone. Bone 39, 1196–1202. Nazarian, A., von Stechow, D., Zurakowski, D., Muller, R., Snyder, B.D., 2008. Bone volume fraction explains the variation in strength and stiffness of cancellous bone affected by metastatic cancer and osteoporosis. Calcified Tissue International 83, 368–379. Nishiyama, K.K., Campbell, G.M., Klinck, R.J., Boyd, S.K., 2010. Reproducibility of bone micro-architecture measurements in rodents by in vivo micro-computed tomography is maximized with three-dimensional image registration. Bone 46, 155–161. Ohman, C., Baleani, M., Perilli, E., Dall’Ara, E., Tassani, S., Baruffaldi, F., Viceconti, M., 2007. Mechanical testing of cancellous bone from the femoral head: experimental errors due to off-axis measurements. Journal of Biomechanics 40, 2426–2433. Ohman, C., Dall’Ara, E., Baleani, M., Van Sint Jan, S., Viceconti, M., 2008. The effects of embalming using a 4% formalin solution on the compressive mechanical properties of human cortical bone. Clinical Biomechanics (Bristol, Avon) 23, 1294–1298. Perilli, E., Baleani, M., Ohman, C., Baruffaldi, F., Viceconti, M., 2007a. Structural parameters and mechanical strength of cancellous bone in the femoral head in osteoarthritis do not depend on age. Bone 41, 760–768. Perilli, E., Baleani, M., Ohman, C., Fognani, R., Baruffaldi, F., Viceconti, M., 2008. Dependence of mechanical compressive strength on local variations in microarchitecture in cancellous bone of proximal human femur. Journal of Biomechanics 41, 438–446. Perilli, E., Baruffaldi, F., Bisi, M.C., Cristofolini, L., Cappello, A., 2006. A physical phantom for the calibration of three-dimensional X-ray microtomography examination. Journal of Microscopy 222, 124–134. Perilli, E., Baruffaldi, F., Visentin, M., Bordini, B., Traina, F., Cappello, A., Viceconti, M., 2007. MicroCT examination of human bone specimens: effects of polymethylmethacrylate embedding on structural parameters. Journal of Microscopy 225, 192–200. Serra, J., 1982. Image Analysis and Mathematical Morphology. Academic Press, London. Tassani, S., Ohman, C., Baleani, M., Baruffaldi, F., Viceconti, M., 2010. Anisotropy and inhomogeneity of the trabecular structure can describe the mechanical strength of osteoarthritic cancellous bone. Journal of Biomechanics 43, 1160–1166. Tassani, S., Ohman, C., Baruffaldi, F., Baleani, M., Viceconti, M., 2011. Volume to density relation in adult human bone tissue. Journal of Biomechanics 44, 103–108. Ulrich, D., van Rietbergen, B., Laib, A., Ruegsegger, P., 1999. The ability of threedimensional structural indices to reflect mechanical aspects of trabecular bone. Bone 25, 55–60. van den Elsen, P.A., Pol, E.J.D., Viergever, M.A., 1993. Medical image matching—a review with classification. IEEE Engineering in Medicine and Biology Magazine 12, 26–39.
2040
S. Tassani et al. / Journal of Biomechanics 45 (2012) 2035–2040
Van Sint Jan, S., Rooze, M., 1992. The thenar muscles: new findings. Surgical— Radiologic Anatomy 14, 325–329. Waarsing, J.H., Day, J.S., van der Linden, J.C., Ederveen, A.G., Spanjers, C., De Clerck, N., Sasov, A., Verhaar, J.A., Weinans, H., 2004. Detecting and tracking local changes in the tibiae of individual rats: a novel method to analyse longitudinal in vivo micro-CT data. Bone 34, 163–169.
Watts, N.B., Ettinger, B., LeBoff, M.S., 2009. FRAX facts. Journal of Bone and Mineral Research 24, 975–979. Xiao, D., Zahra, D., Bourgeat, P., Berghofer, P., Tamayo, O.A., Wimberley, C., Gregoire, M.C., Salvado, O., 2010. An improved 3D shape context based nonrigid registration method and its application to small animal skeletons registration. Computerized Medical Imaging and Graphics 34, 321–332.