Egon Perilli, Laboratorio di Tecnologia Medica, Istituti Ortopedici Rizzoli, Via di Barbiano 1/10, 40136 Bologna, Italy. Tel: +39-051-6366858; fax: +39-051-6366863; e-mail: email@example.com
X-ray microtomography permits the nondestructive investigation of trabecular and cortical bone specimens without special preparation of the sample. To do a quantitative characterization, the cross-section images have to be binarized, separating bone from nonbone. For this purpose, a widely used method is uniform thresholding. However, for commonly available microtomography scanners which use a polychromatic X-ray source, it is unclear what effect the surrounding medium (e.g. air, saline solution, polymethylmethacrylate) has on the threshold value used for the binarization. In the literature an easy procedure to find the optimal uniform threshold value for a given acquisition condition is reported. By applying this procedure, the present work investigated whether a microtomography scan of trabecular bone samples in air or embedded in polymethylmethacrylate gave the same results in terms of structural parameters. The gold standard, that is, histological sections, was used as a reference. Two fixed threshold values were found, one for the microtomography scans performed in air and one for the scans with the same samples embedded in polymethylmethacrylate. These were applied on the correspondent microtomography images for the estimation of structural parameters, such as bone volume fraction, direct trabecular thickness, direct trabecular separation and structure model index. Paired comparisons were made in bone volume fraction between histological sections and microtomography cross-sections for the same bone samples scanned first in air and then embedded in polymethylmethacrylate, by which no significant differences were found. Paired comparisons were also made in bone volume fraction, direct trabecular thickness, direct trabecular separation and structure model index for the same samples over volumes of interest of 4 × 4 × 4 mm3 between microtomography scans in air and scans with the samples embedded in polymethylmethacrylate. Neither these comparisons showed significant differences. This leads to the conclusion that structural parameters estimated by microtomography for human trabecular bone samples scanned either in air or embedded in polymethylmethacrylate are not affected by the surrounding medium (i.e. presence or absence of polymethylmethacrylate), provided that the corresponding optimal threshold value is applied for each acquisition condition.
In the field of orthopaedic research, X-ray microtomography (microCT) has become a frequently used method for the investigation of bone specimens (Sasov, 1987; Feldkamp et al., 1989; Rüegsegger et al., 1996). Trabecular and cortical bone samples can be examined nondestructively, with a spatial resolution in the micrometre range. Structural parameters (e.g. bone volume fraction, trabecular thickness, etc.) can be extracted for the characterization of the sample. To do this, the greylevel images have to be binarized, discriminating bone from nonbone. Different threshold methods are described in the literature for binarization of microCT images, and a number of studies have shown their respective validity and applications (Kuhn et al., 1990; Müller et al., 1998; Ding et al., 1999; Waarsing et al., 2004). In general, these methods can be divided into at least two groups: local and uniform threshold algorithms.
The uniform (also called global or fixed) threshold method is widely used, because of its fast and easy way of application (Hildebrand et al., 1999; van Ruijven et al., 2005). In its most straightforward use, once a threshold value is determined, it is then used over the whole dataset. The attenuation coefficients (i.e. the greylevels) in the reconstructed cross-section images depend on the intrinsic properties of the examined material (e.g. atomic number, density) and on the microCT acquisition conditions (e.g. tube voltage, energy filter). For commonly available microCT scanners which use a polychromatic X-ray source, the beam-hardening effect will change the energy spectrum of the X-rays (Kuhn et al., 1990; Dufresne, 1998). For those systems, it is still unclear what effect the surrounding medium [e.g. air, saline solution, polymethylmethacrylate (PMMA)] has on the threshold value used for the discrimination.
The easiest way to perform a microCT scan of bone specimens is simply by putting the sample in air on the rotation stage (Goulet et al., 1994). For the thresholding algorithm this should be a condition which ensures the highest contrast between bone and the surrounding medium.
Conversely, embedding in PMMA is a necessary treatment of bone samples for the comparison of microCT versus conventional histology. Embedding provides an effective long term storage and manipulation method, and preserves the original trabecular structure during the cutting procedure needed for histology (Kuhn et al., 1990; Engelke et al., 1993; Ito et al., 1998; Thomsen et al., 2005). While performing a microCT scan of embedded bone samples, the PMMA surrounding the bone structure will even change the background value in the reconstructed images with respect to bone samples acquired in air.
Ding et al. presented an easy procedure to find the optimal uniform threshold value for a given acquisition condition. In their work, the described procedure provided an unbiased estimation of bone volume fraction (Ding et al., 1999). To our knowledge, there are no studies that report comparisons between bone specimens acquired at the same microCT settings but with the samples surrounded by different media. It remains unclear what effect this will have on structural parameters such as bone volume fraction, trabecular thickness, trabecular separation and structure model index. It is interesting to investigate whether the microCT scan of bone samples in air or embedded in PMMA give the same results.
In this work we made a quantitative comparison between bone specimens microCT scanned in air and then scanned embedded in PMMA, taking their histological sections as a reference. They were binarized by using global thresholds, following the recommendations of Ding et al. The aim is, having found the appropriate threshold value for each acquisition condition, to investigate whether there are differences in the structural characterization indices.
Materials and methods
As a general overview, all the samples were first microCT scanned in air, then they were embedded in PMMA, rescanned by microCT, and finally examined by histology. The details of the whole procedure are reported as follows.
In the present study 25 human bone samples were obtained from 25 patients undergoing total hip arthroplasty. The donors were 15 women (age 63 years, range 30–82) and 10 men (age 60 years, range 43–78). The samples were bone biopsies taken from the normally removed femoral head, following the procedure reported below. Written informed consent was obtained from the patients.
Bone samples preparation
The removed femoral heads were fixed for at least 3 d in ethanol. From each head, an 8-mm-thick slice was cut in the distal region, at the level of the neck. For the bone cutting a diamond saw (Exact Apparatebau, Norderstedt, Germany) was used. From each bone slice, an approximately parallelepiped biopsy (about 15-mm side, 8-mm height) was taken in the medial region using the diamond saw. The specimens had a suitable size to fit a 15-mm-diameter glass cylinder, and each one contained both trabecular and cortical bone. After, the specimens were grinded with a grinding system (Exact Apparatebau), to ensure a flat bottom surface. As a physical landmark, a small cut was made on each sample (less than 1 mm in depth) with the diamond saw on the outside of the cortical bone. This landmark was used as a reference for placing the bone sample onto the microCT sample holder, allowing a scan of the specimen in the same position before and after the PMMA embedding.
The X-ray microCT system used in this work was a Skyscan 1072 (Skyscan, Aartselaar, Belgium). The system consisted of a microfocus X-ray source, a rotatable specimen holder and a detector system, equipped with a 1024 × 1024 pixel CCD camera. The scanning geometry of the instrument was of the cone-beam type (Sasov, 1987; Feldkamp et al., 1989; Sasov & Van Dyck, 1998). Each X-ray projection was acquired as a 12-bit greylevel image, 1024 × 1024 pixels, stored in 16-bit file format. With 0.45° rotation step, a complete acquisition over 185° lasted 2 h. The computer was equipped with two Intel Xeon 1.7 GHz processors, 1 Gbyte Ram. The cross-section reconstruction was made using the program ‘cone-rec V2.9’ (Skyscan), which is based on the cone-beam reconstruction algorithm described by Feldkamp et al. (1984). The reconstruction of 400 slices took about 2 h using the double-processor option. The total time needed for each microCT examination (acquisition and reconstruction) became about 4 h.
General microCT usage protocol:
The microCT operated at 50 kVp, 200 μA, and an aluminium filter (1-mm thick) was used for beam hardening reduction. Exposure time was 5.9 s, averaged by two frames. Magnification was 16X, the pixel size was 19.5 μm with 20 mm × 20 mm field of view. During the microCT scan, each specimen was entirely contained in the field of view. For each microCT examination a stack of 400 cross-section images was produced, with a slice separation of one pixel (19.5 μm). These images were stored in 8-bit file format (256 greylevels).
MicroCT imaging of bone in air
The bone specimens were exposed to air for 2 h at room temperature (20°C) before undergoing the microCT scan. This procedure was adopted to avoid the creation of artefacts in the microCT images, due to the partial evaporation of ethanol during scanning. The microCT examination in air was performed by applying the previously described general microCT usage protocol.
PMMA embedding of the bone samples
First the methylmethacrylate (MMA, Merck-Schuchardt, Hohenbrunn, Germany) was destabilized, then benzoyl peroxide 3.5% (Sigma-Aldrich Chemie, Steinheim, Germany) was added to start the polymerization. Each bone specimen was placed in a 15-mm-diameter glass cylinder filled with the liquid methylmethacrylate under vacuum conditions. Particular attention was used in placing the bone samples in the cylinder, maintaining the flat bottom surface of the bone sample attached to the bottom of the glass. The glass cylinder was then closed with a plastic plug and placed in an oven at 26°C for polymerization (4 d). After curing, the glass was manually broken and the bone embedded in the PMMA cylinder remained.
MicroCT imaging of embedded bone samples
The previously described general microCT usage protocol was used also for the microCT scan of the bone samples embedded in PMMA. Great care was taken in placing the embedded samples in the microCT scanner in the same position as it was during the microCT scan of the bone samples in air.
Histological examination of the samples
After microCT imaging, the embedded bone samples were grinded down to 20-μm-thick slices. One histological slice per individual was produced. 5 samples were critically damaged during the grinding procedure. Hence, for these samples it was not possible to perform the histological examination. This leaded to 20 valid samples for the comparison between histology and microCT. These were stained (light green), mounted on microscope slides and examined with an optical microscope (Leica DMR-HC, Leica Microsystems, Wetzlar, Germany). Digital images of the histological sections were acquired using a photocamera (Leica DC300, Leica Microsysytems) mounted on the microscope, with a final magnification of 72X (pixel size 4.3 μm).
Histology: calculation of bone volume fraction
For the extraction of the region of interest (ROI) and thresholding of the histological images, the software ‘Leica Qwin’ (Leica Microsystems) was used. The ROIs were square-shaped (4 mm side length), placed appropriately to contain only trabecular bone. Over each ROI, the bone area (sum of the pixels marked as bone) and tissue area (total area of the ROI) were determined using the software ‘Leica Qwin’.
MicroCT: ROI extraction, thresholding, structural parameters calculation
For each sample, the microCT cross-section image, which visually mostly corresponded to the histological image, was chosen out of the stack of microCT slices. On the selected cross-section image, a square-shaped ROI 4 × 4 mm2 (208 × 208 pixels) was accurately placed in the position resembling the one of the histological ROI. For the ROI extraction and thresholding on the 2D microCT images the ‘3D Calculator V0.9’ software (Skyscan) was used. This procedure was applied both to the bone specimens acquired in air and to the specimens embedded in PMMA.
For the binarization of the microCT images, the uniform threshold values for the acquisitions of bone in air and the embedded bone were determined following the recommendations of Ding et al., which are briefly described as follows (Ding et al., 1999).
MicroCT thresholding protocol:
1MicroCT scan of the specimens.
2Measurement of the bone volume fractions of these specimens based on an accurate external method.
3Estimation of a microCT optimal-threshold for each specimen corresponding to the externally determined bone volume fraction.
4The mean value of the optimal-thresholds becomes the fixed-optimal-threshold to be applied to the segmentation of the microCT data.
In the present case, the externally determined bone volume fraction was given by the BV/TV measured on the histological image of each bone sample (point 2 of the thresholding protocol). Then, a single microCT optimal-threshold value was found for each of the specimens scanned in air, based on comparison with the externally determined BV/TV (point 3 of the thresholding protocol). The same procedure was applied on the embedded bone samples, obtaining a second set of optimal-threshold values. For the microCT acquisitions performed in air, the fixed-optimal-threshold was then determined by taking the mean of the corresponding single microCT optimal-threshold values (point 4 of the thresholding protocol). Similarly, a fixed-optimal-threshold value for the acquisitions of the embedded bone samples was found by taking the mean of their single optimal-threshold values. These two fixed-optimal-threshold values were then applied to the segmentation of the respective microCT images.
The parameter BV/TV was estimated over the microCT slices using Eq. (1) (software ‘3D Calculator V0.9’). One histological slice for each individual was used for comparison in BV/TV between histology and the corresponding microCT sections of the bone in air and of the bone embedded in PMMA.
Then, for comparison between microCT examinations of the bone in air and of the bone embedded in PMMA, the following structural parameters were estimated over a cubic-shaped volume of interest (VOI) of 208 microCT slices (4 × 4 × 4 mm3) for each specimen (software ‘3D Calculator V0.9’):
(i) Bone volume fraction, BV/TV (%), using Eq. (1) over all 208 microCT slices.
(ii) Direct trabecular thickness, Tb.Th* (μm): Tb.Th*, also called model-independent thickness, gives the thickness in 3D of the structure, independently of an assumed structure type. It evaluates a volume-based local thickness, by fitting maximal spheres to every point contained in the 3D-structure. The mean thickness of the structure is given by the arithmetic mean value of the local thicknesses (i.e. of the diameters of the maximal spheres) taken over all points of the structure (Hildebrand & Rüegsegger, 1997a; Perilli et al., 2006).
(iii) Direct trabecular separation, Tb.Sp* (μm): Tb.Sp* is determined with the same procedure as used for Tb.Th*, but giving a volume-based estimation of the thickness of the marrow cavities (Hildebrand & Rüegsegger, 1997a). Thus, it gives a model-independent measure of the separation between the trabeculae.
(iv) Structure model index, SMI: SMI is a topological index, which gives an estimate of the characteristic form in terms of plates and rods composing the 3D-structure. Its implementation is based on a differential analysis of the triangulated surface of the structure. SMI assumes the integer values 0 and 3 for ideal plates and rods, respectively, whereas for a mixed structure containing both plates and rods the SMI-value lies between 0 and 3 (Hildebrand & Rüegsegger, 1997b; Perilli et al., 2006).
Comparisons were made in BV/TV estimated over the corresponding ROIs between histology and microCT, first for the acquisitions with the bone in air and then with the bone embedded in PMMA (20 samples). A further comparison was made in the parameters BV/TV, Tb.Th*, Tb.Sp* and SMI estimated over VOIs of 4 × 4 × 4 mm3, between microCT scans of bone in air and microCT scans of embedded bone (25 samples).
As indicated in the work of Kuhn et al., 1990, the actual differences in the structural parameters di and the percent differences di% found in the comparisons for each sample were obtained as follows:
The mean actual difference and the mean percentage difference for each structural parameter were calculated, where N is the number of the samples (N= 20 in the case ‘histology vs. microCT’, N= 25 in the case ‘microCT air vs. microCT embedded’):
The structural parameters were tested for normality by the Shapiro-Wilks test. As the data were compatible with a normal distribution, the two-tailed Student's t-test for paired samples was used in the comparisons. The differences were deemed to be statistically significant at P < 0.05.
Figure 1(a)–(d) shows the X-ray frontal images and cross-sections of a bone sample microCT scanned first in air and after being embedded in PMMA. In Fig. 1(c) and (d) can be noticed the grey halo caused by the PMMA embedding. In general, a good visual agreement between the microCT images and the histological sections was found. Figure 2(a), (b) and (c) shows a histological section of a bone sample, the corresponding microCT cross-section first with bone in air and then with the bone embedded in PMMA, respectively. The bone marrow was visible in the histological sections and in the microCT sections of the bone scanned in air (Fig. 2a and b), whereas it was indistinguishable from the PMMA in the microCT sections of the embedded bone (Fig. 2c). Figure 2(d)–(f) shows the corresponding binarized images.
The basic descriptive statistics of BV/TV estimated over the 2D sections obtained by the three methods is shown in Table 1. The mean actual difference , the mean percentage difference and the statistical significance found in the comparison between histology and microCT are shown in Table 2. There were no significant differences in BV/TV between histology and microCT.
Table 1. Basic descriptive statistics of the bone volume fraction BV/TV (%) estimated by histology and by microCT over 2D sections of the 20 bone samples (ROI: 4 × 4 mm2).
Table 2. Comparison between histology and microCT in bone volume fraction BV/TV (%) estimated over 2D sections of the 20 bone samples (ROI: 4 × 4 mm2). is the actual mean difference in BV/TV between histology and microCT, % is the mean percentage difference, P is the probability for the paired Student's t-test.
Histology – microCT air:
Histology – microCT embedded:
Figure 3(a) and (b) shows the scatter plots ‘BV/TV in air versus BV/TV histology’ and ‘BV/TV embedded versus BV/TV histology’. Both linear regressions had a high coefficient of determination (R2= 0.96 and R2= 0.94, P < 0.001, respectively).
Table 3 shows the basic descriptive statistics of the structural parameters estimated by microCT over VOIs of 4 × 4 × 4 mm3 for the bone samples scanned both in air and in PMMA. The mean actual difference , the mean percentage difference and the statistical significance found in the comparisons on these characterization parameters are reported in Table 4. No significant differences were found in BV/TV, Tb.Th*, Tb.Sp* and SMI over the VOIs between microCT scans of bone in air and microCT scans of bone embedded in PMMA. Figure 4 shows the scatter plot ‘BV/TV embedded versus BV/TV in air’ obtained by microCT over the VOIs (Fig. 5). The linear regression fitting these data had a high coefficient of determination (R2= 0.97, P < 0.001).
Table 3. Basic descriptive statistics of structural parameters estimated by microCT on bone samples scanned in air and then scanned after embedding in PMMA (25 bone samples, VOI: 4 × 4 × 4 mm3).
Table 4. Comparison in structural parameters estimated by microCT between bone samples scanned in air and then scanned after embedding in PMMA (25 bone samples, VOI: 4 × 4 × 4 mm3). is the actual mean difference in the parameters obtained by the two scan modalities, is the mean percentage difference, P is the probability for the paired Student's t-test.
MicroCT air – microCT embedded:
In the present work quantitative comparisons were made between bone samples scanned by microCT both in air and after embedding in PMMA, using their histological sections (gold standard) as a reference. As discussed by Ding et al. (1999), it is important to find the specific threshold value for each microCT acquisition condition. Their recommendations were applied, using histological BV/TV as the so-called externally determined volume fraction. The spatial resolution (pixel size 19.5 μm) used for the microCT scans and the particular attention paid throughout the process, from the embedding procedure to the histological examination, ensured close visual correspondences between microCT slices and histological sections. The BV/TV parameter, which is the most widely used for the characterization of bone samples, showed no significant differences from histology, either for acquisitions in air, or for acquisitions with the bone embedded in PMMA. Considering the structural parameters estimated over the VOIs of 4 × 4 × 4 mm3, no significant differences were found either for BV/TV, or for Tb.Th*, Tb.Sp* and SMI between bones scanned in air and bones embedded in PMMA.
The main reason for the microCT scan of bone samples in air in the present work is that it represents the easiest way of conducting a scan (Goulet et al., 1994). In this manner, attenuation of the X-ray beam is due only to the bone sample itself, no other material surrounding the bone interferes with it, and thus it provides the highest contrast in the images. Despite this, in practice the usual treatments of bone specimens undergoing microCT examination are putting the bone in a cylinder filled with a liquid solution (alcohol, formaldehyde, saline solution) (Rüegsegger et al., 1996; Müller & Rüegsegger, 1997; Müller et al., 1998) or embedding it in PMMA (Kuhn et al., 1990; Engelke et al., 1993; Ito et al., 1998; Thomsen et al., 2005). The storage method in liquid solution allows the physiological conditions of the bone sample to be restored. This treatment is widely used, thus permitting a subsequent mechanical testing of the bone specimen in wet conditions. PMMA embedding has been adopted from histology and preserves the original trabecular structure during the cutting procedure. It is a necessary treatment of bone samples for the comparison of microCT versus histology. Both methods, that is, storing in liquid solution and embedding in PMMA, allow sample storage and manipulation. At the energies used in microCT (with energy peaks of about 25–30 keV, with typical tensions of 50 kVp) the PMMA has an attenuation coefficient which is similar to water (at 30 keV: μPMMA= 0.36 cm−1, μWater= 0.38 cm−1) (Hubbel & Seltzer, 1995). Thus, in an initial approximation, storage in water or in PMMA during a microCT scan could be considered as two similar microCT acquisition conditions, if compared to the scan of bone samples in air. From these considerations, scanning bone samples in air and scanning bone samples embedded in PMMA represent two different conditions from a radiological point of view. The way of finding the fixed optimal threshold for each acquisition condition was achieved by following a simple procedure (Ding et al., 1999). By applying the corresponding fixed optimal threshold to each microCT acquisition condition, the presence or absence of the surrounding medium (PMMA) did not affect the BV/TV obtained by microCT with respect to the BV/TV determined histologically. It did not affect the structural parameters estimated by microCT over VOIs of 4 × 4 × 4 mm3 either, when comparing the bone scans in air and those embedded in PMMA. This finding supports the procedure suggested by Ding et al. for finding the fixed optimal threshold value for microCT scanning.
As a further attempt, the images of the bone samples embedded in PMMA were also binarized by using the fixed optimal threshold value found for the acquisitions in air. This led to significant overestimations of 20% in BV/TV with respect to the histological images (data not shown). This result suggests once more that it is important to find the fixed optimal threshold value for a given acquisition condition.
In conclusion, human trabecular bone samples were examined by microCT first in air and then after embedding in PMMA. Corresponding to each one of the two acquisition conditions, two fixed optimal threshold values were found. By applying these thresholds, comparison with histology showed no significant differences in BV/TV for both microCT acquisition conditions. Over VOIs of 4 × 4 × 4 mm3 obtained by microCT, the paired comparisons in BV/TV, Tb.Th*, Tb.Sp* and SMI between the bone samples scanned first in air and then after embedding in PMMA showed no significant differences either. These results show that the structural parameters determined by microCT for trabecular bone samples scanned either in air or embedded in PMMA are not affected by the surrounding medium, (i.e. presence or absence of PMMA), provided that the corresponding optimal threshold value is applied for each acquisition condition.
The authors would like to thank Roberta Fognani for the constant support during the sample manipulation, Luigi Lena for the images and Keith Smith for checking the language of the manuscript.