Introduction
In postmenopausal women, the majority of bone is lost from the cortical bone compartment as a result of both reduced cortical thickness (Ct.Th) and increased cortical porosity (Ct.Po) [
1]. Both parameters can be measured in vivo with high-resolution peripheral quantitative computed tomography (HR-pQCT) and were recently shown to be associated with a higher prevalence of hip fractures [
2]. However, this novel imaging technology is still rarely available and based on ionizing radiation. Alternatively, quantitative ultrasound (QUS) techniques are being developed, which are non-ionizing, low cost, and portable. For example, a simple ultrasonic pulse-echo measurement was proposed to predict Ct.Th, but the ultrasonic wave-speed in the cortical bone layer was assumed to be known [
3].
Modern ultrasound axial transmission (AT) measures the dispersion curves of guided waves, which propagate in the cortical shell of long bones [
4]. In early AT applications, isotropic tube models were fitted to the dispersion curves, providing Ct.Th ex vivo at the human radius [
5] and bovine tibia [
6]. Subsequently, a transverse isotropic free plate model was proposed, the use of which allowed estimating Ct.Th and four bone elastic coefficients at the same time [
7]. The unknown coefficients of this plate model were then reduced to Ct.Th and Ct.Po [
8]. To build such a model, asymptotic homogenization [
9] has been applied to estimate the effective stiffness tensor as a function of porosity, assuming an invariable stiffness of the tissue matrix. However, the mineralization of the bone tissue matrix in humans, intimately related to the stiffness, is not constant, but affected by age [
10], gender [
11], treatment, and disease [
12].
In the beginning of its development, AT has extensively been used to measure the first arriving signal velocity (
υFAS) in the cortex of the radius and tibia. The first arriving signal measured at low frequencies has a larger penetration depth than at high frequencies. Thus, it can capture features of deeper cortical bone layers in which disease-associated changes usually start to occur [
13]. Accordingly, low-frequency
υFAS (200 kHz) measured at the tibia was significantly correlated with Ct.Th (R = 0.24,
p < 0.001), whereas high-frequency
υFAS (1.25 MHz) was not [
14]. The ability of
υFAS to discriminate subjects with osteoporotic fractures from non-fractured controls was shown to be similar [
15] or inferior [
16,
17] when compared to areal bone mineral density (
aBMD) measured by dual-energy X-ray absorptiometry (DXA) at the hip or spine. DXA is considered the current standard method for osteoporosis diagnosis and fracture risk prediction.
In an attempt to provide complementary parameters to
υFAS with improved fracture discrimination ability, researchers also considered the phase velocity of the A
0 mode (
υA0) [
18,
19]. A
0 is a fundamental flexural guided wave, which propagates within the cortical bounds and is particularly sensitive to both Ct.Th and to pathological changes in the endosteal region depending on the frequency-thickness ratio regime [
20]. Following these findings, an ex vivo study at the radius showed significant correlations of
υA0 with Ct.Th (R
2 = 0.52,
p < 0.001) and with volumetric bone mineral density (
vBMD) (R
2 = 0.45,
p < 0.001) [
21]. However, when investigated in vivo at the tibia, the correlations between both
υA0 and
vBMD and
υA0 and Ct.Th were less strong [
14]. According to the authors, the correlations decreased due to interferences with the soft tissue, in which ultrasound propagates at similar velocities (~ 1500 m
.s
−1) as the A
0 mode in cortical bone [
22].
In this ex vivo study, we predicted Ct.Th and Ct.Po at the human tibia using a model-based inversion method which was previously proposed by our group for a similar 1-MHz radius probe [
23]. To account for the difference in Ct.Th between the tibia and radius, a novel probe was designed to optimize the excitation of guided waves in the Ct.Th range usually found in humans at the diaphysis of the tibia. Compared to the former radius probe, the central frequency is reduced from 1.0 to 0.5 MHz, whereas the probe dimensions are slightly increased. Cortical bone samples were extracted from the region below the receiver array for
Ct.PoμCT reference measurements using high-resolution micro-computed tomography (μCT, 7.4 μm isotropic voxel size). Site-matched reference
Ct.ThμCT and
vBMD were obtained from a larger μCT scan at lower resolution (39 μm isotropic voxel size). In addition, we assessed the acoustic impedance (a surrogate for matrix stiffness) using scanning acoustic microscopy (SAM) to evaluate the assumption of a waveguide model with invariant matrix stiffness. The ultrasonic velocities
υFAS and
υA0 were measured and compared to site-matched cortical bone properties.
Discussion
In this ex vivo study, the estimation of cortical thickness (Ct.Th) and porosity (Ct.Po) at the human tibia using full spectrum guided-wave analysis was successfully validated against site-matched high-resolution micro-computed tomography (μCT). We utilized a novel 500-kHz axial transmission (AT) transducer which was designed to optimize the excitation of guided wave modes at the diaphysis of the tibia. Furthermore, we accounted for a possible inter-specimen variation of the cortical bone matrix elasticity by incorporating the acoustic impedance from site-matched scanning acoustic microscopy (SAM). The variability of the matrix elasticity did not improve our model-based predictions of Ct.Po and Ct.Th. This result supports the concept of variations in matrix stiffness which has a minor impact on the effective elasticity tensor compared to the effect of variations in porosity [
26]. Note that our matrix stiffness measurements might have also been biased by experimental errors. For the first time, the A
0 mode velocity (
υA0) was measured in human cortical bone using SVD-enhanced 2D Fourier transforms and compared to site-matched Ct.Th and Ct.Po at the same time.
The systematic overestimation of Ct.Th (0.28 mm) by AT has twofold implications. On the one hand, the
Ct.ThμCT reference measurement is affected by the natural variability of the bone morphology, as illustrated in Fig.
5, and by the horizontal error bars of Fig.
4a. On the other hand, the exact behavior of guided waves in samples with irregular and trabecularized boundaries (Fig.
5c) has not yet been investigated. Interpreting the results for these cases is particularly challenging, since the distinction of cortical bone from the trabecular compartment in the μCT images is itself a matter of arbitrary decision, as discussed in the next paragraph. Numerical simulations of ultrasound propagation using realistic (structurally heterogeneous) cortical bone models could help in clarifying to what extent trabecularized regions participate in the waveguide. Figure
4d suggests a bias of Ct.Po that depends positively on the porosity level. This bias might be partially caused by larger partial volume effects in the estimation of reference
Ct.PoμCT for samples with higher Ct.Po. The assumption of a waveguide model with invariant matrix stiffness might also contribute to the bias. To partially correct for this effect, we accounted for variations in the axial tissue stiffness (c
33) by means of average acoustic impedance of mineralized tissue from SAM. Future ex vivo studies could incorporate the full transverse isotropic stiffness tensor of the waveguide, e.g., as experimentally obtained from resonant ultrasound spectroscopy [
39].
The prediction of Ct.Th (R
2 = 0.57) was weaker than for Ct.Po (R
2 = 0.83). This was mainly caused by one sample (indicated with a circle in Fig.
4) which had a heavily trabecularized cortex as shown in Fig.
5c. When this sample was excluded, the correlation between
Ct.ThAT and
Ct.ThμCT improved significantly (R
2 from 0. 57 to 0.94, RMSE from 0.37 to 0.16 mm). We believe that this is due to the definition used for the determination of
Ct.ThμCT, which is especially uncertain within highly trabecularized cortical bone regions. Note that a consensus on how to segment the cortical bone compartment has not yet been reached. The longitudinal μCT section of Fig.
5c (right) obtained from the outlier sample explains the Ct.Th discrepancy between μCT (green) and AT (red line). The figure suggests that guided waves also propagated in the trabecularized bone region, but our applied cortical compartment segmentation algorithm [
32] did not include this region.
We have used cortical bone samples from adults without report of metabolic bone diseases. For this reason, we cannot conclude on the general applicability of our method to subjects with considerably different matrix stiffness compared to normal adult bone (e.g., children, patients with osteogenesis imperfecta [
40], or patients on long-term bisphosphonate treatment [
41]). To overcome the assumption of invariant matrix stiffness, the elastic tensor could be derived from the plate model instead of porosity as it was previously suggested [
7,
42]. However, this approach would increase the number of unknown model coefficients and require complete resolutions of the experimental dispersion curves. Our current guided wave transducer technology is limited, particularly in spatial resolution, and therefore cannot yet provide such reconstruction quality.
The major limitation of this study was the small sample size used for statistics (
N = 17). Nevertheless, a broad range of
Ct.ThμCT (2.3–5.1 mm) and
Ct.PoμCT (5.6–22.8%) was covered, which represents what is usually found in other studies [
26,
43]. Furthermore, the dependency of
υFAS on
vBMD is consistent with previous studies at the tibia using different frequencies (200 kHz [
14], 250 kHz [
44], 400 kHz [
45], and 1.25 MHz [
43]). However, we did not find a statistically significant correlation between
υFAS and
Ct.ThμCT, as it has been observed for the tibia using 200 kHz [
14] and 400 kHz [
45]. The dependency of
υA0 on Ct.Th and
vBMD confirms the findings of an ex vivo study at the radius using 200-kHz AT [
21]. We excluded two samples due to large deviations of the ultrasonic measurements between the cycles. The one failure case (Fig.
5a) had a very thin cortical bone layer (Ct.Th < 2.0 mm) in which ultrasonic guided waves cannot sufficiently be excited using the 500-kHz probe. Alternatively, we could have used the 1-MHz probe which was originally designed for measurements at the thinner radius. The second failure case exhibited a very inhomogeneous and trabecularized cortex (Fig.
5b) which might not have guided the ultrasonic waves appropriately.
Previous studies which measured
υA0 in cortical bone extracted the wave packages of the A
0 mode in the time domain [
5,
19]. In contrast, our method isolates the A
0 dispersion curve in the frequency-phase velocity domain. We assume that this approach is more accurate since it ensures that no other signals interfere. Furthermore, we accounted for small inclination angles between the probe and bone surface using bi-directional measurements which will become more beneficial in vivo in the presence of soft tissue. However, the in vivo applicability of this novel
υA0 measurement technique remains to be demonstrated.
A former data acquisition protocol, used by our group at the radius, was based on three cycles of ten successive measurements [
42]. For the current work, we used notably longer scan times (i.e., 400 successive measurements per cycle) and slowly tilted the probe. In the post-processing, a waveguide model was then fitted to the dispersion curves of each measurement, providing estimates of Ct.Th and Ct.Po. When the dispersion curves were too noisy or incomplete, the solution to the problem was no longer unique, as indicated by several local maxima in the objective function. Therefore, we used a criterion that allowed us to exclude such problematic measurements. In the future, this automatic criterion could be evaluated in real time to retain only measurements without model ambiguities.
In conclusion, the best predictions of cortical thickness (Ct.Th) and porosity (Ct.Po) were obtained from a plate model with invariant matrix stiffness, which was fitted to the measured guided wave dispersion curves. The second best predictors of Ct.Po and Ct.Th were vBMD and υA0, respectively. No further enhancements were observed by accounting for variations in matrix stiffness. Clinical pilot studies are currently ongoing to confirm the possibility of a full-spectrum ultrasonic guided-wave analysis in vivo.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.