A high-resolution CT system using synchrotron radiation allowed visualization of the 3D cortical bone microarchitecture and measurement of intracortical porosity of femoral neck cortical bone specimens from 19 female cadavers imaged at 10.13-μm resolution. 3D reconstruction of specimens showed osteonal system arrangement. Mean porosity was 15.88%. This technique will provide insights into the mechanisms involved in osteoporotic hip fractures.
Introduction: The purpose of this study was to show that a high-resolution CT system using synchrotron radiation (SR) allows visualization of the 3D cortical bone microarchitecture of the human femoral neck and quantification of intracortical porosity.
Materials and Methods: Bone specimens from the inferior femoral neck were obtained from 19 female cadavers with no hip fracture (mean, 86.9 ± 8.3 years). The specimens, consisting of embedded ∼7 × 7 × 12-mm cortical bone parallelepipeds, were imaged using SR at 10.13-μm resolution. Commercial software was used to visualize both the 660 × 660 × 660-voxel volumes and the 2D axial slices through each volume. Qualitative examination of 2D axial slices focused on the appearance of the vessel canal system, presence of small bright zones (fully mineralized tissue) in the osseous matrix, and presence of cracks. A method was developed to automatically measure 3D intracortical porosity after separating pure bone from pores and cortical bone from trabecular bone.
Results and Conclusions: 3D reconstruction of the specimens showed the entire structure and arrangement of the osteonal systems, parallel to the axis of the femoral neck. Bright zones were seen in the outer cortex. No cracks were observed. Porosity values varied widely from 4.96% to 38.87% (mean, 15.88 ± 9.87%). This study establishes that SR microtomography can be used to display the 3D bone microstructure of the human femoral neck cortex and to quantify intracortical porosity. This technique will provide insights into the mechanisms involved in cortical bone loss and osteoporotic hip fractures.
Osteoporotic hip fractures are a major public health problem associated with high human and economic costs. For the past 20 years, hip fracture studies have focused on trabecular bone. However, loss of compact bone may also contribute to hip fractures.(1) Mechanical testing of excised femoral necks has shown that the cortex contributes 40-60% of the overall strength of the femur.(2) Similarly, finite element modeling has suggested that cortical bone in the femoral neck region may support 50% of the stresses associated with normal gait.(3) Loss of compact bone is caused by increased porosity and decreased cortical thickness, and MacCalden et al.(4) have shown that an age-related increase in porosity accounts for 76% of the reduction in strength of the proximal femur.
Another problem is that, because of the complex anatomy of the hip and thinness of the femoral neck cortex, studies of the mass and structure of femoral cortical bone usually focused on the diaphysis (investigated by histology,(5) 2D image analysis,(6,7) or CT(8–10)). Recent data on the femoral neck has come mainly from a single research group.(11–13) Interactive 2D image analysis studies in hip fracture patients and controls have shown that loss of cortex plays a greater role than loss of trabecular bone and that cortical bone loss occurs primarily in the anterior and inferoanterior regions of the neck.(11,12)
Finally, the microstructure of the cortical bone is anisotropic. Therefore, 2D methods of analysis give only an estimate of the 3D structural properties of the cortex. Because the structural properties of the cortical shell have a marked influence on bone strength,(4) this may be an obstacle to an understanding of the risk factors for hip fracture. Available data on the 3D microstructure of the femoral cortical bone come from studies in which histomorphological structures were followed through contiguous sections,(14) or more recently, from 3D reconstruction of digitized 2D images of serial histological bone sections.(15) However, such techniques are subject to deformation artifacts; in addition, they are time-consuming and destructive, which are two drawbacks that limit their use for analyzing a large number of samples. Current 3D imaging techniques used in clinical practice do not provide sufficient spatial resolution for studying the microstructure of the cortex. To our knowledge, there are no published μCT data on the 3D appearance of human femoral cortical bone, particularly at the femoral neck.
The value of synchrotron radiation (SR) for X-ray studies of small bone specimens is based on several properties including natural collimation, extremely high intensity, and broad spectral bandwidth.(16–18) We used a 3D μCT system based on SR to study a cortical bone specimen from the inferior neck quadrant of the right femur of each of 19 female cadavers. The purpose of this study was to determine whether a high-resolution SR-based CT system is effective for observing and quantifying 3D intracortical porosity in human femoral neck cortex specimens.
MATERIALS AND METHODS
Subjects and specimens
We studied 19 specimens of human femoral neck cortex obtained, in accordance with standard bone banking procedures, from the Pathology Institute linked to the authors' university.
The specimens were harvested from the excised right femurs of 19 elderly white women who had a negative history for hip fracture. Age at death ranged from 72 to 103 years (mean age, 86.9 ± 8.3 years; Table 1). Eight women had died in a hospital, four in a nursing home, and seven at home. There was no information on causes of death or possible immobilization before death. The femurs were obtained within 10 days after death, manually cleared of soft tissue, and immediately frozen at −20°C.
Table Table 1.. Details on the Study Subjects
Preparation of the specimens (Fig. 1) included two parallel cuts at the base of the neck and at the base of the head. This produced a 1.5-cm segment of femoral neck from which the inferior sector was isolated. The periosteal side was marked, and the specimen was fixed in ethanol and embedded in polymer. A ∼7 × 7 × 12-mm parallelepiped was cut from the inferior cortex of the base of the femoral neck using an Isomet 11-1180 low-speed saw (Buehler Ltd.). The 12-mm side was aligned with the main direction of the Haversian systems. The inferior cortex was chosen as the site where the femoral neck cortex is thickest.(19) Embedding was used to ensure stable fixation of the specimen to the goniometer of the tomography system.
We used the X-ray microtomography system developed at the European Synchrotron Radiation Facility (ESRF, Grenoble, France) on the topography and high-resolution diffraction beamline (Insertion Device 19). This system has been described in detail elsewhere.(20) Only the main points are provided below.
The system uses a wide SR parallel beam emerging from the storage ring. A double crystal monochromator selects the appropriate energy from the white SR beam (range, 8-80 keV). We chose to work with an energy of 24 keV because of the size and linear attenuation coefficients of the specimens.(21)
The specimen was mounted on a goniometer that allowed high-resolution translations and rotations. This enabled us to position the sample and rotate it in the beam.
The 2D detector is composed of a scintillator screen for X-ray-to-light conversion, two optical lenses for magnification, and a charge coupled device (CCD) camera with a 1024 × 1024-element CCD chip and an element size of 19 μm. We chose the optical magnification that produced a theoretical pixel size of 10.13 μm, and the spatial resolution of the 2D detector (characterized by a global modulation transfer) was experimentally found to be 15.76 μm.(20)
Camera control, image acquisition display, and storage to the disk were done using a SUN Ultra 1 workstation linked to the camera.
For each specimen, we acquired 900 equiangularly distributed projections over 180° to obtain a complete data set for reconstruction. Tomography scanning was done automatically and involved sequentially rotating the specimen while recording an image, which was saved to the workstation hard disk. Total acquisition time was ∼15 minutes/sample, corresponding to 2 gigabytes of data.
The 3D image was reconstructed using a 3D-filtered backprojection algorithm.(20) To reduce the computation time, the reconstruction algorithm was optimized for a specific hardware (Alpha Processor DEC). The reconstruction was performed in a volume of interest of 660 × 660 × 660 isotropic voxels, with the bottom of the volume placed horizontally on the cut at the base of the femoral neck (the ∼7 × 7 × 12-mm parallelepiped was not entirely reconstructed). The duration of the reconstruction process was ∼45 minutes. The dynamics of the reconstructed linear attenuation coefficients was further reduced to a range of 256 gray levels.
Image display and analysis
Each reconstructed volume was visualized in 3D using the software VG studio (Volume Graphics, GmbH, Heidelberg, Germany) (MH, FP) and in 2D using Image J (a public domain Java image processing program) concomitantly by two of us (VB, CB). Qualitative examination of 2D axial slices focused on three points: the appearance of the vessel canal system, the presence of fully mineralized tissue in the osseous matrix, and the presence of cracks. The tissue was considered fully mineralized if it was sharply demarcated from the underlying cortex and was identified by gray-level differences.
We developed a method that automatically measures 3D intracortical porosity, defined as the percentage of pore volume (Ca.V) within the total cortical volume (including pore volume; Ct.V), in keeping with the definition used for 2D image analysis. First, we binarized the image to separate pure bone from background or pores, using simple thresholding. Because of the image contrast and high spatial resolution, the gray-level histogram was clearly bimodal (Fig. 2), and the same threshold of 98 was fixed for all samples. Second, the cortical volume (Ct.V) was defined by removing the trabecular bone present at the endosteal surface of the cortex (Fig. 3). The procedure was based on 3D mathematical morphology operations. A closing operation was first used to fill the pores within the cortical bone. Then, an opening operation eliminated the thinner trabecular bone. A 3 × 3 × 3 cubic structuring element was used. The result of the algorithm was controlled by visual inspection. An automatic run of the algorithm using 16 and 36 iterations for closing and opening, respectively, allowed successful processing of 16 of 19 volumes. For the three remaining volumes, the number of iterations was adapted manually. Porosity was finally calculated as the ratio of the number of pore voxels in the cortical volume over the number of cortical volume voxels.
In addition, for each specimen, the diameter of each pore (Po.Dm) and cortical thickness (Ct.th) were quantified. For this purpose, we used the method proposed by Hildebrand and Rüegsegger(22) to assess trabecular thickness on 3D images. This method computes the local thickness at each point of the structure as the diameter of the largest sphere containing the point and included in the structure. It therefore provides a distribution of local thickness from which statistical information can be extracted. Our implementation of the proposed method was based on discrete geometry, as detailed elsewhere.(23) The method was applied both to the 3D cortical image and to the 3D pore image to obtain the distribution of cortical thickness (Ct.th) and pore diameter (Po.Dm), respectively.
Descriptive statistics included the range, mean, and SD of age, porosity, and cortical volume.
Intraspecimen diversity in pore diameter (Po.Dm) was evaluated using the distribution of Po.Dm of each specimen to calculate the mean; the 25th, 50th, and 75th percentiles; the interquartile range (difference between the 25th and 75th percentiles); and the minimum and maximum pore sizes. The 50th percentile of cortical thickness was also obtained for each specimen.
In addition, to compare Po.Dm diversity among specimens, we divided the study population in two groups: one with lower porosity values (9 specimens) and the other with higher porosity values (10 specimens). The mean; the 25th, 50th, and 75th percentiles; and the interquartile range of Po.Dm (obtained for each specimen) were compared in the two groups using the nonparametric Mann-Whitney U-test.
Relationships between parameters were assessed by Spearman coefficients (r values). For all tests, p values smaller than 0.05 were considered significant. Statistics were obtained using Statview (Version 5.0; SAS Institute).
The 3D appearance of the vessel canal system was clearly visible. In keeping with previous descriptions of the diaphyseal cortex of long bones studied using other techniques, the vessel canal system of the femoral neck cortex appeared as a dense ramifying network. The general direction of the Haversian systems was vertical or slightly oblique along the long axis of the neck (Fig. 4). Volkmann's canals went off at a right angle or obliquely, in all directions, connecting to all the adjacent Haversian canals, whatever their diameter. Considerable variability was seen in the diameters of Haversian canals, both between specimens and within a single specimen. The Volkmann's canals also had variable sizes, which were not clearly related to the size of the canals from which they arose.
In specimens with a low porosity value examined in a vertical plane, most of the canals were slender with parallel edges. In specimens with a high porosity value, most of the canals were large. The edges of the large canals were not parallel (Fig. 5).
In some specimens, we observed numerous small zones of fully mineralized tissue within the extraosteonal matrix. These zones were located inside the cortex, predominantly in its outer region, most often directly under the periosteum. The margins of these bright zones were irregular and sharply defined, with no gradual merging into the adjacent “normal” extraosteonal matrix (Fig. 6). We did not observe cracks in any of our specimens.
3D intracortical porosity and cortical volume
Porosity ranged from 4.96% to 38.87% of the total cortical bone volume. Mean porosity and SD were 15.88% and 9.87%, respectively. Mean cortical bone volume was 136.6 ± 50.3 mm3 (Table 1).
The mean; the 25th, 50th (median), and 75th percentiles; and the interquartile range from the distribution of Po.Dm in each specimen are given in Table 1. For all specimens, minimum Po.Dm was equivalent to the size of the pixel. Maximum Po.Dm ranged between 378.2 and 1107.5 μm2. The 50th percentile of the cortical thickness is given in Table 1.
The 25th, 50th, and 75th percentiles of Po.Dm were significantly different between the two groups of porosity values (p < 0.001 in all cases).
Porosity was significantly correlated to the mean and the 25th, 50th, and 75th percentiles (r = 0.93, p < 0.0001; r = 0.96, p < 0.0001; r = 0.96, p < 0.0001; and r = 0.89, p = 0.0002; respectively). No significant correlation was observed between porosity and minimum Po.Dm, maximum Po.Dm, interquartile range, cortical volume, or cortical thickness. No significant correlation was observed between age and, porosity, Po.Dm, or cortical parameters.
To our knowledge, this study is the first report on the 3D appearance of the microstructure of the human femoral neck cortex examined using a μCT system. The system, operating with SR at a resolution of 10.13 mm, clearly visualized the 3D appearance of the vessel canal system without inducing bone destruction. We also developed original software that automatically measured intracortical porosity in 3D. We identified areas of fully mineralized tissue that are thought to be involved in bone fragility.
It has been suspected for many years that 3D reconstruction could provide spatial information on complex biological structures, such as the vessel canal system in cortical bone. Initially based on 2D serial histological sections, 3D reconstruction was time-consuming and destructive. Cohen and Harris in 1958,(24) Tappen in 1977,(14) and later Stout et al.(15) and Robling and Stout(25) described the 3D anatomy of osteons and resorption spaces by tracing them on serial cross-sections or digitizing serial cross-sections. The introduction of techniques for automating serial sectioning and for directly capturing digitized images of cut surfaces was a major advance in 3D bone imaging.(26) However the spatial resolution was not sufficient to provide information on cortex structure. Chappard et al.(27) and Arsenault(28) provided 3D images of the vessel canal system using corrosion casting methods, and Petrtyl et al.(29) studied the macroscopic spatial organization of the osteons by filling the vascular canals of undecalcified polished bone with India ink. However, these methods are destructive. Feldkamp et al.(30) pioneered the use of μCT, a time-efficient and nondestructive technique for obtaining a map of the internal structure of bone specimens. Few μCT studies have focused on cortical bone.(10) The use of SR for high-resolution tomography of small specimens was first suggested by Grodzins,(21) who pointed out the advantages of this method. To date, available data obtained using SR deal mainly with trabecular bone.(31,32) In the only study focusing on cortical bone, Rindby et al.(33) used SR (7-μm spatial resolution) to perform microdiffraction to investigate variations in size and orientation of the crystallites within an osteon. However, the 3D appearance of the cortex was not described in this SR-based study.
Direct 3D visualization of the vessel canal has three advantages over 2D images. First, it highlights the richness and complexity of the vascularization of the cortex. Second, because the network is a highly anisotropic structure orientated along the long axis of the neck, errors caused by malorientation of the section plane with 2D methods (such as overestimation of porosity) can be expected to be reduced by 3D methods of image acquisition and analysis. Third, 3D visualization of the cortical bone helps to understand what is measured under the term “porosity” and to understand the mechanisms of the age-related increase in porosity. In 2D-based studies, Bell et al.(34) and Jordan et al.(35) found that giant canals (diameter > 385 μm) were associated with the phenomenon of remodeling osteons and that remodeling osteons were not randomly distributed in the cortex but spatially distributed in clusters (called super-osteons). However, with their 2D method, these authors were not able to determine whether a cluster arose from proximate but physically separate Haversian systems or from a single complex system. Answering this question is a formidable challenge, because the images are merely snapshots of a process that unfolds over several years. Specimens with low porosity may be valuable sources of information on the development of large canals and cavities. These low-porosity specimens contained fairly large canals that seemed clustered in small groups of two or three. In 3D, the canals of these small groups appeared to be distinct entities. However, they were close together and connected by channels. The canals were large along only part of their length (i.e., 1 mm). In highly porous specimens, some canals were very large along their entire explored length, in contrast with the focal enlargement seen in low-porosity specimens; however, variations in circumference occurred along these canals. In some places, we saw two very large canals running close to each other, being separated by small amounts of pure bone or in contact intermittently. This phenomenon has been described in 2D as pore coalescence (40 years ago), or more recently, as merging of canals. Between two large canals, only small amounts of pure bone remained, possibly disconnected from the other regions of pure bone and therefore at risk for disappearing.
In some subjects, we saw small bright zones within the extraosteonal matrix, predominantly in the outer region of the cortex. Because of the technical characteristics of the SR μCT set-up, the 3D linear attenuation coefficient map accurately reflects mineral density,(18) and the reconstructed gray-scale image is related to bone mineralization at the tissue level.(36) The bright zones correspond to fully mineralized tissue, that is, tissue with a higher mineral content than the adjacent cortex.(19) Boyce and Bloebaum(19) and Vajda and Bloebaum(37) using backscattered electron imaging, observed large peripheral regions of fully mineralized tissue in cortices of human femoral neck and diaphysis. The origin of fully mineralized tissue is not clear: possibilities include subperiosteal calcification derived from the periosteum,(38) necrotic tissue where minerals continue to accumulate,(19) or calcified fibrocartilaginous tissue at the points of attachment of tendons or capsules.(19,37) In our study, the fully mineralized tissue was seen as thin bands or spots inside the cortex but was far less developed than in the subjects studied by Bloebaum and colleagues. These bright zones were not related to technical artifacts: they did not resemble target artifacts or other μCT artifacts, and with identical imaging conditions, some specimens had no bright zones, whereas others had focally distributed bright zones in the outer cortex. The presence of fully mineralized tissue may have implications for femoral neck fracture. Indeed, fully mineralized bone absorbs less energy than does non-fully mineralized bone, making formation and extension of cracks easier.(39) In our study, however, no cracks were visible in the fully mineralized regions.
We developed a computer program that automatically measures the porosity after automatic separation of the cortical bone from the trabecular bone. Considerable variations in structure were found across our human femoral neck specimens, with porosity ranging from 4.96% to 38.87% (mean, 15.88%). The results are in agreement with those of 2D studies of the cortex of human femoral diaphysis.(4-7,40) However, few data on porosity of the femoral neck cortex are available. Videodensitometry measurements showed that mean femoral neck cortex porosity was higher in 10 women with femoral neck fractures (21.6%) than in 10 controls (14.27%).(41) Mean cortical porosity at the middle of the femoral neck, as measured by backscattered electron imaging, was 11% in nine adults 60-71 years of age.(19) In another study done by backscattered electron imaging, mean porosity at the medial femoral neck was 6.1% in seven female cadavers, 7.1% in seven women treated with total hip arthroplasty for joint disease, and 12.9% in nine women treated with hemiarthroplasty for femoral neck fractures.(42) Mean porosity was 7.46% in the inferior femoral neck cortex in 12 women.(43) Mean porosity, measured using a semiautomatic image analysis system, ranged from 7.7% to 17% in femoral neck cortex biopsy specimens from 13 women with intracapsular hip fracture and 10 female controls.(11) Our mean value is slightly higher than in these earlier studies. However, all our specimens were from elderly women, and intracortical porosity is known to increase with advancing age.(19)
We determined the distribution of pore size in each specimen. The methods of image acquisition and pore analysis were 3D methods, and pore size was determined as the diameter of the largest sphere included in the structure. The smallest visible canals had a diameter as small as the voxel size, leaving open the possibility that the specimens contained canals too small to be picked up with the resolution of our system. In addition, our resolution level did not allow us to take osteocyte lacunae into account. The mean and the studied percentiles of Po.Dm were strongly correlated with porosity and were significantly smaller in low-porosity specimens than in high-porosity specimens. These two points are not surprising given that porosity is defined as the ratio of pore volume over cortical volume. In addition, we found no significant difference between low-porosity and high-porosity specimens regarding the interquartile range or minimum and maximum Po.Dm values. Thus, high-porosity specimens did not have greater variability in pore size, but rather, had larger pore size overall compared with low-porosity specimens.
Our study has several limitations. First, we had no information on the subjects, particularly regarding the causes of death and whether immobilization, a cause of increased porosity, occurred before death. The large SD of the porosity values limits the clinical application of our findings. However, the objective of our study was not to establish reference values for porosity. In addition, the porosity range and mean value were consistent with previous studies of human femoral neck cortex. A second limitation is that we did not compare 3D porosity values with 2D values from serial histological bone sections. Such comparisons have been reported for trabecular bone quantitative parameters from μCT(44,45) or high-resolution SR-based CT systems(18) and histological sections. They showed strong correlations between the techniques. Third, we obtained few quantitative results. However several decades of research have been required to obtain quantitative data on trabecular bone microarchitecture. In future work, the availability of 3D images should enable direct computation of additional quantitative parameters such as porosity distribution (from the periosteum to the endosteum), connectivity, degree of anisotropy, and matrix mineralization.
In conclusion, this study is the first report on the 3D appearance of the microstructure of the human femoral cortex evaluated using a μCT system based on SR. We studied the neck cortex, a site where osteoporotic fractures occur. The 3D arrangement of the osteonal systems was depicted. We were also able to observe changes thought to be involved in bone fragility such as zones of fully mineralized tissue. Intracortical porosity was measured in 3D. The radiation dose is too high to allow use of this technique in vivo. However, use of this technique in vitro to compare the femoral neck cortex of subjects of various ages, with and without osteoporotic hip fractures, can be expected to improve our understanding of the mechanisms leading to femoral neck fracture. In addition, cortical bone structure measurements by in vivo 3D peripheral quantitative CT could be calibrated with μCT using SR, as previously done for trabecular bone.(46)
We thank Prof Jean-Pierre Lassau for providing the excised femurs and Françoise Giannoni for helping to prepare the samples. Data were acquired at the ESRF during experiment LS1171, and we acknowledge the ESRF and CREATIS staff for helping us during the experiment, especially Stefania Nuzzo, Jean-Martial Mari, and Pierre Bleuet. This work was partly supported by the Fondation de l'Avenir (Paris, France).