Quantitative Backscattered Electron Imaging of Bone Using a Thermionic or a Field Emission Electron Source

From a material point of view, bone is a composite material of a soft and tough organic collagen matrix reinforced with stiff and brittle inorganic hydroxylapatite nano-crystals [1, 2]. During an individual’s life span, bone forming and resorbing cells (osteoblasts and osteoclasts, respectively) steadily remodel bone, thus, allowing for mechanical adaptation and repairing material fatigue event like cracks. Osteoblasts lay down new bone in the form of collagenous bone matrix (osteoid) that subsequently mineralizes. This mineralization process can be roughly separated in two phases. During the phase of primary mineralization, the newly formed bone achieves up to \(70\%\) mineral content in few days, whereas it takes up to months and even years to complete mineralization in the subsequent phase of secondary mineralization [3‐8]. This interplay of bone formation and resorption leads to a specific pattern of bone packets of different age and, thus, of different mineralization contents. A quantitative assessment of bone mineralization is important as many metabolic diseases, and medication may affect mineralization. Examples include (high and low bone turnover) osteoporosis [9‐12], osteogenesis imperfecta [13‐15], melorheostosis [16, 17], hypophosphatemia [18], hypophosphatasia [19, 20] as well as bisphosphonate [21‐23] or teriparatide [24] treatment.

The clinical gold standard for characterization of the bone status of a patient is measuring the bone mineral density (BMD) via dual-energy X-ray absorptiometry (DXA). One of the major drawbacks of this technique is that the result of a DXA measurement is a 2-dimensional projection obtained from a radiography through the whole body. Thus, it is impossible to disentangle the effect of bone volume and matrix mineralization. In other words, DXA does not allow to discriminate if a larger absorption contrast stems from a larger amount of bone material in the beam path or from a higher mineralization of the bone matrix [25]. Another clinical sophisticated method is high-resolution peripheral quantitative computed tomography (HR-pQCT) [26, 27]. While it allows to measure 3-dimensional bone micro-architecture, the use of a polychromatic X-ray beam giving rise to beam hardening and its voxel resolution of approximately 80 μm limit the accuracy of the obtained volumetric BMD. The method of quantitative backscattered electron imaging (qBEI) has the potential to overcome these limitations. In contrast to x-rays used in DXA and HR-pQCT measurements, backscattered electrons provide information of mineralization density from a sample depth of only 1–2 micrometers when working with beam electron energies of 20 keV [28]. This means that instead of a 2D projection of a 3D measurement as is obtained in DXA, a qBEI analysis gives truly 2D information on bone matrix mineralization at the sample surface devoid of effects stemming from bone volume. Thus, the frequency histogram of degree in mineralization of the sample, the so-called bone mineralization density distribution (BMDD) can be obtained directly from the backscattered electron images [29]. The method was first suggested by Alan Boyde [30] and further refined by Paul Roschger who used it extensively [31, 32]. Other groups throughout the world have also used it to measure bone mineral content in pathological situations [33‐36]. Recently, evaluation protocols were developed to also assess osteocyte lacunae number, size, and shape from qBE images [15, 18, 37, 38]. However, it has to be emphasized that this method—in contrast to DXA and HR-pQCT—requires a bone biopsy sample.

The increasing popularity of the qBEI method for bone mineralization measurement makes it important to understand how results from different groups and devices with different electron sources can be compared. With respect to the electron source, existing SEMs can be classified in two groups: thermionic and field emission electron guns. The physical principle of thermionic electron guns is to heat the material to high temperatures to allow some electrons to overcome the work-function energy barrier and escape into vacuum via thermal excitation. As this type of cathodes relies on thermal excitation and, thus, has to be operated at elevated temperatures, the resulting current density and its life time are generally low. Typical thermionic electron sources are made from tungsten or Lanthanum Hexaboride (LaB\(_6\)). The second material has a lower working-function energy barrier compared to the first and, thus, can be operated at lower temperatures yielding a higher current density. In contrast to thermionic excitation, a field emission cathode uses the physical principle of electric field amplification close to sharp tips. A field emission cathode is a metal wire ending with a sharp tip with a radius of several \(100 \ \text {nm}\). If this cathode is held at constant (negative) voltage, the electric field close to the tip becomes so large (\(>10^9 \ \text {V/m}\)) that the working energy barrier is reduced and electrons can escape into vacuum. The current density is approximately a factor \(10^5\) larger compared to thermionic sources. Furthermore, the small source size of a field emission cathode results also in a small beam divergence and, thus, higher brightness (current per solid angle) compared to a thermionic source. As a field emission cathode can be operated at lower temperatures, its life time is considerably enhanced compared to a thermionic cathode (see also Fig. 1 for images of typical examples of the cathodes).

The qBEI method relies on the fact that in a scanning electron microscope (SEM), the number of backscattered electrons (BEs) is dependent on the local atomic number Z of the specimen. For low atomic numbers (\(Z \lesssim 20\)), this dependence is approximately linear [28]. The mean Z values of bone with a varying fraction of organic matrix (\(Z \approxeq 6\)) and hydroxylapatite (\(Z=14.06\)) lie well in this range, provided that in addition to the normal composition of bone, no significant amounts of other (heavier) elements are present. This allows to deduce the local mineral content in bone from the measured backscattered electron yield when an electron beam is scanned over a bone surface in a SEM that was calibrated by reference standards with known Z numbers prior to measurements [29].

In principle, the linear dependency of backscattered electron yield at low atomic numbers is actually only fulfilled if the sample material is homogeneous down to the atomic scale (randomly dispersed atoms) within the interaction volume, i.e., the sampling volume of BEs generated by the electron beam impinging the sample surface. This condition is well fulfilled for pure crystals provided that the strongly ordered arrangement of atoms on the sample surface is destroyed by proper mechanical polishing. Otherwise the BE yield would be additionally influenced by crystal orientation. In contrast, for bone, which is a nano-composite of organic molecules (mainly collagen) and nano-sized hydroxylapatite particles, the conditions of homogeneity within the sampling volume of BE are likely less strongly fulfilled. The dominant structural motif the bone material is built of, is the mineralized collagenous fibril of 50 to 200 nm in diameter impregnated by 4-nm-thick plate-like crystals. These mineralized fibrils compose a lamellar structure with a period of about 5 μm further forming the bone packets (basic structural units, BSUs). In consequence, not only the actual mineral content of the fibril, but also the amount of fibril material that is effectively sampled per pixel during scanning might influence the BE intensity. Thus, instrumental parameters like beam size, beam divergence, beam electron intensity as well as spatial pixel resolution will potentially contribute to a BE contrast additionally to that of atomic number contrast.

Nevertheless, the size and specificity of this effect are hitherto unexplored. This gap shall be closed in the current paper. We compare qBEI results from two different SEMs equipped with different electron sources. The first is a ZEISS DSM 962 with a tungsten hairpin cathode and the second a ZEISS field emission SEM SUPRA 40 (see Fig. 1 for images of the used cathodes). The former was used to validate the qBEI method and to establish reference BMDDs from healthy adults [32, 39]. Having these two different microscopes at hand and considering the issues discussed above, the question arises if measurements on the two devices are (quantitatively) comparable. In particular, this answers the question if measurements on the field emission SEM can be compared to the reference BMDDs obtained earlier with the DSM 962. The broader impact of this research question is to answer, how results from different research groups obtained with SEMs with different electron sources and operated with different experimental settings (e.g., nominal magnification), can be compared.

To answer the question, if the BMDD outcomes of both types of SEMs can be directly compared, the following analyses were performed: (i) The linear relation between backscattered electron yield and mean atomic number was verified for the two SEMs used (On the Relation Between Backscattered Electron Yield and Mean Atomic Number). (ii) The identical surface of the same sample was measured repeatedly with the same device. The time span between the measurements was even up to several years. This gave insight into the long-term stability of the devices, reproducibility of the measurement itself as well as on the stability of the resin embedded bone sample over time (Device Stability Over Time). (iii) A comparison of BMDDs obtained with the same device at different magnifications was performed to apprehend how different magnifications affect the obtained images (Measurements with the Same Device at Different Magnifications). (iv) BMDDs from the identical sample but measured with two different SEM devices at the same spatial pixel resolution were compared. This could clarify the quantitative comparability of BMDD curves obtained from different devices. The results showed that although both devices lead to qualitatively similar results, quantitatively the results were not equal. In particular, this means that BMDDs measured with the SEM SUPRA 40 cannot be used directly to compare them with the reference BMDD curves for healthy bone obtained with the DSM 962 [39] (Comparison Between Different Devices). (v) Consequently, it is described how new reference BMDDs were obtained for the new device (Reference BMDD Curves).

The measurements of the backscattered electron yield as a function of mean atomic number Z of different mineral crystals confirmed a linear dependence in the range of Z corresponding to bone (\(Z \approx 10\)). This linear dependence was found equal for different pixel resolutions and both devices with different electron sources (DSM 962 with a tungsten hairpin cathode and SEM SUPRA 40 with a field emission cathode). These results justify the use of Eq. 4 to relate measured GL and local calcium content in bone samples that lies at the heart of the qBEI method for SEMs independent of their specific electron source.

Further, we observed that there is a clear broadening of the BMDD peak as well as a GL shift in the BMDDs to higher mineralization levels when measured at similar pixel resolution by SEM SUPRA 40 compared to DSM 962. Such a shift is consistently found for all samples measured and is in the range up to 1.0 wt% Ca for the BMDD parameter CaMean. One of the main reasons for the observed discrepancies seems to be the difference in beam characteristics for the two devices. The electron source size is 100 to 1000 times smaller for the field emission compared to the tungsten cathode. The smaller source size causes a much less divergent and smaller beam diameter at the imaged sample surface for the SEM SUPRA 40 compared to the DSM 962 [28]. One particular feature of a field emission microscope is that the beam size is the same for all nominal magnifications, while for the DSM 962 the beam size has to be adjusted for each magnification. Consequently, the beam electron interaction volume in the sample material, respectively, the escape area of the backscattered electrons from the sample surface, is much smaller for the SEM SUPRA 40 compared to the DSM 962 (nevertheless, still much larger than the beam diameter for both devices [28]). Most importantly, for both devices, the escape area is smaller than the nominal magnification given by the pixel size. This means that not everything located in the pixel contributes to the measured signal. Therefore, some details of the structure (such as cement lines) may actually be under-sampled or even missed in the SEM SUPRA 40 mapping. This explains the observation of smearing of cement lines in the SEM SUPRA 40 compared to the DSM 962 presented in Fig. 6. A measurement with higher pixel resolution will, however, reveal many more details in the device with the field emission cathode (see, e.g., the gradual increase in magnification for the SEM SUPRA 40 in Fig. 5). The consequence is that the lower inherent resolution of the DSM has some advantages for (low resolution) qBEI mapping, while the field emission microscope has clear advantages for high-resolution imaging. This effect is likely leading to the described alterations in BMDD characteristics compared to the DSM 962. Noteworthy that in case of homogeneous samples like Z standards, differences in beam diameter at focal plane also lead to consistently higher BE signal outcomes for the SEM SUPRA 40 compared to the DSM 962. In contrast, differences in magnification do not change the BE signal outcomes for the SEM SUPRA 40, which are consistent with the fact that in this device, spot size (beam diameter at sample surface) does not change with magnification (see Fig. 3). Both observations support the assumption that the above-described effect of BMDD GL shift originates from the inhomogeneous hierarchically structured bone material. Consequently, a numerical rescaling of the BMDD curves of the FE-SEM is not feasible as the shift in GL between the two devices is not a function of the device alone but depends also on the ultrastructural characteristics of the sectioned bone sample surface.

In contrast, measurements with the identical device are highly reproducible over the range of years with a variation of \(\pm 1\) GL, i.e., 0.17 wt% Ca. This is much better than the variations of up to \(\pm 6\) GL (1.0 wt% Ca) found between the different devices. Consequently, the stability of the device (sample chamber, detector, and electronics) and also of the PMMA embedded sample itself, allows to compare measurements obtained from the same device over years.

These observations allow drawing the following conclusions: qBEI measurements can be quantitatively compared when they are (i) measured with the same device and (ii) measured under exactly the same conditions (especially the same magnification). If the stability of the device can be ensured the achieved relative accuracy (within one device) can be as good as 0.2 wt% Ca or \(\pm 1 \ \text {GL}\). On the other hand, the different types of electron beams used in the two devices lead to differences in measured parameters of up to 1.0 wt% Ca or \(\pm 6 \ \text {GL}\) for instance in case of CaMean.

In particular, this has implications on the trabecular reference BMDD for healthy adults published in reference [39]. In this paper, it was shown that the trabecular BMDD in healthy adults is independent of age, sex, skeletal site, and ethnicity. Consequently, measurements on arbitrary samples can be compared against these references to gain insight into the mineralization pattern, e.g., if the sample is normal-, hypo- or hyper-mineralized. As changes in BMDD parameters due to diseases, like e.g., osteoporosis or osteogenesis imperfecta, are often in the range of \(\pm 1 \ \text {wt\% Ca}\) [29] which is similar to the changes due to different electron sources, direct comparisons between measurements obtained from different devices or obtained under different settings, e.g., magnification, are impossible. As the published reference was obtained with the DSM 962, they cannot be compared against BMDDs obtained with the newer device SEM SUPRA 40. Consequently, new reference curves for the SEM SUPRA 40 had to be obtained. Additionally to the trabecular reference BMDD, also a cortical reference was established from the same individuals at the same anatomical site (iliac crest).