Calcif Tissue Int (2004) 74:60–67 DOI: 10.1007/s00223-002-2113-3
Calcified Tissue International 2003 Springer-Verlag New York Inc.
Does Quantitative Ultrasound of Bone Reflect More Bone Mineral Density Than Bone Microarchitecture? B. Cortet,1,2 N. Boutry,2,3 P. Dubois,4 I. Legroux-Ge´rot,1,2 A. Cotten,2,3 X. Marchandise4 1
Department Department 3 Department 4 Department 2
of of of of
Rheumatology, Hopital R. Salengro, University-Hospital of Lille, 59037 Lille Cedex, France Rheumatology, Unite´ de Recherche de l’Appareil Locomoteur (URAL) du CH & U de Lille, 59037 Lille Cedex, France Radiology, University Hospital of Lille, 59037 Lille Cedex, France Biophysics, University Hospital of Lille, 59037 Lille Cedex, France
Received: 20 August 2002 / Accepted: 6 May 2003 / Online publication: 2 October 2003
Abstract. Relationships among quantitative ultrasound of bone (QUS), bone mineral density (BMD) and bone microarchitecture have been poorly investigated in human calcaneus. Twenty-four specimens, from 12 men and 12 women (mean age 78 ± 10 years; range 53–93), removed from cadavers were studied. The feet were axially sectioned above the ankle. Two variables were measured for QUS (Achilles, Lunar): broadband ultrasound attenuation (BUA) and speed of sound (SOS). A third variable, the stiffness index (SI), which is a combination of both BUA and SOS, was also calculated. BMD (a lateral view) was measured on a QDR 2000 densitometer (Hologic). Bone microarchitecture was assessed by computed tomography (CT) using a conventional CT-system. Fifteen sagittal sections (1 mm in width and 2 mm apart) were selected for CT. Methods used for characterizing bone microarchitecture consisted in structural (trabecular network characterization) and a fractal analyses. The relationships between QUS and bone microarchitecture were assessed by simple linear regression analysis with and without adjustment for BMD (partial correlation) and by backward stepwise regression analysis. Strong relationships were found between BMD and QUS. Adjusted r2 values were 0.545 for SOS and 0.717 for SI. Two microarchitectural variables were also significantly correlated with both SOS and SI: apparent trabecular separation (App Tr Sp) and trabecular bone pattern factor (App TBPF). After adjustment for BMD few correlations between QUS and microarchitectural variables were always significant. Adjusted squared semipartial coefficients of correlation (rsp2) values between SOS and bone microarchitecture were 6%, 6.8%, 13.2% and 4.6% for App BV/TV, App Tr Sp, App TBPF and fractal dimension (FD), respectively. For SI, corresponding figures were 3.7%, 4.1%, 5.2% and 3.2%. Backward stepwise regression analysis using BMD and microarchitecture showed a slight increase of r2 values that varied from 8.4% for SI to 17.8% for SOS, compared with BMD alone. The current study suggests that although BMD is a major determinant of acoustic properties of human calcaneus, significant density independent relationships with bone microarchitecture should also be taken into account.
Correspondence to: B. Cortet; E-mail:
[email protected]
Key words: Quantitative ultrasound of bone — Bone microarchitecture — Computed tomography — Bone texture analysis — Bone mineral density
Quantitative ultrasound (QUS) measurement of bone is a new peripheral technique that has been well demonstrated for predicting fracture risk, especially in elderly women [1–5]. Also, QUS has been shown to be of interest in metabolic diseases other than osteoporosis and particularly for assessing bone involvement in primary hyperparathyroidism and Cushing’s syndrome [6, 7]. Compared with bone densitometry, QUS devices have several advantages. They do not use ionizing radiation, they are cheap and some are portable. Moreover, the ability of QUS to predict fracture risk is usually significant after adjustment for BMD, suggesting that QUS could provide information independent of bone mass and perhaps related to bone microarchitecture. These findings could also be the consequence of the heterogeneity of the skeleton. Indeed, the adjustment is usually done at the hip or at the spine. Nevertheless, in clinical studies BMD is not usually measured at the calcaneus. Several in vitro, studies have also shown that QUS could provide some information about bone microarchitecture [8–16]. However, one should be cautious regarding these studies for several reasons; (1) some of them have been done on animal cancellous bones [8, 9, 14]; (2) some concluded that QUS reflects bone structure because findings were different according to the three orthogonal axes taken into account, however, in clinical practice the measurements are done in a single direction; (3) very few studies assessed the relationships between QUS and bone microarchitecture at the calcaneus which is the site most commonly used for QUS measurements in clinical practice and they have drawn conflicting conclusions [11, 15, 16]. Indeed, Hans et al. [11] have shown that QUS reflects bone mass for a large part and bone microarchitecture to a lesser extent. Similar findings were also found by Ha¨usler et al. [15]. By
B. Cortet et al.: QUS and Bovine Microarchitecture
61
Fig. 1. (a) Sagittal section of the calcaneus before the selection of the region of interest (ROI). (b) Same view after the selection of the region of interest (ROI).
contrast, Nicholson et al. [16] demonstrated that by taking into account bone mass and bone microarchitecture they could explain a greater percentage of the variance in QUS (+8% compared with the use of density alone). Although the latter study is of interest, one limitation should be taken into account. The authors used excised bone samples put in a water bath and these conditions are very different from the measurements done in clinical practice. Finally, for most of these studies the variables used for characterizing bone microarchitecture were not sophisticated [15, 16]. We have developed a noninvasive tool for characterizing bone microarchitecture using a conventional computed tomography (CT) system which has been well demonstrated in postmenopausal but also male osteoporosis [17–19]. The main goal of the present study was to assess the relationships between QUS at the heel and CT-derived microarchitecture in 24 feet from cadavers.
females and 6 males and their mean age was 78 ± 10 years (ranges 53–93 years). The feet were axially sectioned above the ankle. QUS QUS measurements were performed on an Achilles (Lunar corporation, Madison, WI, USA). The entire foot, the ankle and the distal part of the leg were studied. The measurements were done in the same manner as those in clinical practice. Two variables were measured: broadband ultrasound attenuation (BUA) and speed of sound (SOS). A third variable, the stiffness index (SI), a combination of both BUA and SOS, was also calculated. SI does not reflect the homonymous mechanical property. The in vivo short-term precision expressed as the coefficient of variation (CV) and was assessed in 18 young adults measured twice using the formula: qffiffiffiffiffiffiffiffiffiffiffi Pn dffi 2n
CVð%Þ ¼ X1 þ X2 =2
Materials and Methods
where d is the difference between a pair of measurements, X1 and X2 the means values for the first and second group of measurements, and n the number of paired observations. The CVs were, respectively, 0.23%, 2.6% and 2.6% in our center for SOS, BUA and SI [20].
Sample Preparations
Bone Densitometry
Twenty-four feet taken within a few hours after death (always less than 24 hours) from 12 subjects were studied. There were 6
BMD was measured by dual-energy X-ray absorptiometry (DXA) with a Hologic QDR 2000 densitometer (Hologic Inc.,
62
B. Cortet et al.: QUS and Bovine Microarchitecture
Fig. 2. (a) Binary image related to the previous ROI. (b) Same image after dilatation for calculating the trabecular bone pattern factor.
Waltham, MA, USA). It was measured on a lateral view with the software usually used at the wrist. The ROI was selected manually and was composed only of trabecular bone. CT Scans CT scans were done on an Elite Plus (Elscint, Haifa, Israel) with a high resolution algorithm. Fifteen to 20 axial slices (1 mm in width, 2 mm apart) were acquired in standard fashion (Fig. 1a). The number of slices acquired varied according to the size of the calcaneus. For each specimen 15 sagittal central slices were selected. The following CT settings were used: 315 mAs and 140 kV; field of view 140 mm; pixel matrix 512 · 512; pixel size 200 lm leading to a maximum spatial resolution of 400 lm. The region of interest (ROI) was rectangular and was defined manually (Fig. 1b). The size of the ROI varied according to the size of the image and was composed of only trabecular bone. Images were transferred on a personal computer and bone structural analysis was performed. Algorithms used to characterize bone texture were all developed in our laboratory. For each patient we calculated the mean of the results obtained from the 15 slices. Bone Structure Analysis The procedure was extensively described elsewhere [19]. Due to the fact that the variables measured on CT sections, as com-
pared with histological sections, are not the ‘‘real parameters’’ they were preceded by ‘‘apparent parameters.’’ Trabecular Network Characterization. The boundary between cortical bone and trabecular bone was defined using an automatic contour detection algorithm. Thereafter, a segmentation process permitted separation of trabecular bone from bone marrow. The segmentation was made with an edge detection using a laplacian of gaussian filter [21, 22]. The laplacian of gaussian filter includes both a smoothing filter which convolutes the image by a gaussian filter and a 2nd order derivative filter. The tuning of this combined filter deals with the size of the smoothing window but also the variance of the convolutive gaussian. The zero-crossing detection in the resulting image provides a binary image in which the dark regions represent the bone marrow and the light region the trabeculae (Fig. 2a). The procedure that we described at the distal radius [17, 18] needed to be adapted because of the heterogeneity of the calcaneus (strong overestimation of the light regions). Therefore, we also applied to the original image a thresholding process leading to take into account the gray levels above the mean—25% of the standard deviation. The laplacian of gaussian was thereafter applied to the original image but also to the thresholded image. The 2 binary resulting images were finally combined. This procedure permitted us to keep the pixels coming from the 2 resulting images and led to exclusion of the ‘‘false positive’’ of trabeculations in the dark regions. An additional pruning step was also applied to the resulting image for removing the residual small size artifacts (below 4 pixels).
B. Cortet et al.: QUS and Bovine Microarchitecture
63
Fig. 3. Skeletonized image related to the binary image 2a. On this binary image several variables derived from histomorphometric parameters were measured [23–28]: apparent bone volume/tissue volume (App BV/TV, %) based on a 2dimensional analysis; apparent trabecular thickness (App Tb Th, mm); apparent trabecular separation (App Tb Sp, mm); apparent trabecular number (App Tb N); trabecular bone pattern factor (TBPF, mm)1) [27]; Euler number/ROI area (mm)2) [28]; and marrow space star volume (mm3) [26]. App BV/TV and other variables obtained from the binary image were measured on each slice selected. TBPF [27] was defined as (B.Pm ) B.Pm2)/(B.Ar ) B.Ar2) where B.Ar is the surface area containing bone tissue, B.Pm the perimeter of B.Ar and B.Ar2, and B.Pm2 the surface area and the perimeter of the image dilated by one pixel, respectively (Fig. 2b). The method described by Vesterby et al. [26] being time-consuming, in the present study we used the method described by Levitz and Tchoubar [29] for measuring the star volume in porous glasses and adapted by Chappard et al. [30] for the study of bone. The Euler number was measured using the method described by Compston [28]. This method consists of counting the number of enclosed cavities and connected trabeculae. The Euler number is defined by the difference between the number of particles present in the spaces occupied by trabeculae and the number of bone marrow spaces circumscribed by these trabeculae. A thinning algorithm was thereafter applied to the binary image by reduction of the thickness of the related components to one pixel (Fig. 3) On the skeletonized image, several variables derived from histomorphometric studies were also measured [24, 25]: apparent total skeleton length of trabecular network/ROI area (App TSLTN, mm)1), apparent total skeleton length of bone marrow/ROI area (App TSLBM, mm)1), apparent node count/ROI area (App N Nd, mm)2), apparent terminus count/ROI area (App N Tm, mm)2), apparent nodeto-node strut count/trabecular strut length (App Nd Nd SC, mm)1), apparent node-to-terminus strut count/trabecular strut length (App Nd Tm SC, mm)1), apparent terminus-to-terminus strut count/trabecular strut length (App Tm Tm SC, mm)1), apparent node-to-node strut length (App Nd Nd SL, mm), apparent node-to-terminus strut length (App Nd Tm SL, mm), apparent terminus-to-terminus strut length (App Tm-Tm SL, mm). Fractal Analysis. The method used in the present study was the one described by Peleg et al. [31] and modified by Lynch et al. [32]. This method for calculating the fractal dimension (FD) involved defining upper and lower surfaces (i.e., ‘‘blankets’’) that cover the initial gray level surface. We used structuring elements defined as flat discs. All the pixels of the surface area were taken into account for calculating this blanket that was
obtained by dilatation and erosion with an e · e sized structuring element. Surface area was defined as the volume of the blanket divided by 2 · e. The average time for performing CT analysis was about 10 min for each specimen. Statistical Analyses The results are expressed as mean ± standard deviation (SD). The normal distribution of the variables measured was assessed using the Kolmogorov-Smirnov test. A simple linear regression analysis assessed the relationships among QUS, BMD and bone microarchitecture first. Associations between QUS and architecture were also assessed after adjustment for BMD, i.e., using partial correlation coefficients (rp). The level of statistical significance was determined using the F statistic. When significant associations between QUS and architecture were found after adjustment for BMD, squared semipartial coefficients of correlation (rsp2) were generated. These represent the variance in QUS attributable uniquely to a given architecture independently of BMD. Finally, we performed a backward stepwise regression analysis for selecting the best models, explaining QUS properties using a combination of microarchitecture and BMD. P values lower than 0.05 were considered statistically significant. Statistical analyses were done using Statview version 5 (SAS institute, Cary, NC, USA).
Results
Raw data for BMD, QUS and bone microarchitecture are given in Table 1. Relationships among QUS, BMD and microarchitectural variables are reported in Table 2. Findings suggest that QUS variables were strongly correlated with BMD (Figs. 4 and 5), since r ranged from 0.752 for SOS (adjusted r2: 0.545, P < 0.001) to 0.854 for SI (adjusted r2: 0.717, P < 0.001). Few microarchitectural variables were also moderately correlated with QUS: App Tr Sp and App TBPF. R values for App Tr Sp ranged from )0.455 (SI) to )0.479 (SOS) and from )0.419 (SI) to )0.524 (SOS) for App TBPF. P values for all these correlations were <0.05. By contrast BMD did not correlate with the features obtained from CT image analysis, suggesting that the variables
64
B. Cortet et al.: QUS and Bovine Microarchitecture
Table 1. Descriptive statistics (mean ± SD) for clinical, ultrasonic, densitometric and bone microarchitectural variables Variables
Mean ± SD
bles did not improve the prediction of BUA compared with BMD alone.
Range Discussion
Age (years) 78 ± 10 BUA (dB/MHz) 90.06 ± 11.25 SOS (m/s) 1543.54 ± 33.14 SI 72.29 ± 15.35 BMD (g/cm2) 0.520 ± 0.17 App BV/TV (%) 39.05 ± 2.04 App Tb Th (mm) 0.465 ± 0.054 App Tb Sp (mm) 0.696 ± 0.019 App Tb N (mm)1) 0.854 ± 0.13 TBPF (mm)1) )1.287 ± 0.308 App Euler number (mm)2) 0.065 ± 0.018 App Star volume (mm3) 0.735 ± 0.106 App TSLTN (mm)1) 0.675 ± 0.11 App TSLBM (mm)1) 0.936 ± 0.119 App N Nd (mm)2) 0.264 ± 0.09 0.372 ± 0.109 App N Tm SC (mm)2) App Nd Nd SC (mm)2) 0.350 ± 0.08 App Nd Tm SC (mm)2) 0.452 ± 0.07 App Tm Tm SC (mm)2) 0.045 ± 0.009 App FD 1.58 ± 0.11 App Nd Nd SL (mm) 1.483 ± 0.23 App Nd Tm SL, mm 0.982 ± 0.18 App Tm Tm SL, mm 1.502 ± 0.27
53–93 76–121.5 1493–1620 54–114.5 0.262–0.857 34.28–43.05 0.358–0.586 0.459–0.905 0.666–1.119 )0.859/)2.215 0.036–0.119 0.562–1.024 0.497–0.942 0.736–1.204 0.131–0.471 0.214–0.642 0.220–0.504 0.347–0.614 0.032–0.070 1.33–1.73 1.131–1.963 0.688–1.423 1.013–2.014
measured on CT scans reflect more bone microarchitecture than bone mass. A borderline correlation was found between App BV/TV and BMD, however (r = 0.34, P = 0.09). After adjustment for BMD, a few correlations between QUS and microarchitectural variables were also found. Significant rp values were found between SOS and App BV/TV (rp: 0.414), App Tr Sp (rp: )0.434), App TBPF (rp: )0.567) and FD (rp: 0.429). Corresponding adjusted rsp2 values were 6.0%, 6.8%, 13.2% and 4.6% for App BV/TV, App Tr Sp, App TBPF and FD, respectively. In the same manner, significant rp values were found between SI and App BV/TV (rp: 0.411), App Tr Sp (rp: )0.426), App TBPF (rp: )0.469) and FD (rp: 0.424). Corresponding adjusted rsp2 values were 3.7%, 4.1%, 5.2% and 3.2% for App BV/TV, App Tr Sp, App TBPF and FD, respectively. We found no correlation between BUA and microarchitectural variables after adjustment for BMD. Backward stepwise regression analysis using BMD and microarchitecture showed a slight increase of r2 values that varied according to the QUS parameter selected. Indeed, BMD, App TBPF and FD explained 72.3% (adjusted r2) of the variance in SOS against 54.5% for BMD only (adjusted r2), representing an r2 increase of 17.8% compared with BMD alone. To a less extent, findings were also significant for SI. Indeed, App TBPF, FD and BMD explained 80.1% (adjusted r2) of the variance in SI against 71.7% for BMD only, representing an r2 increase of 8.4%. Multivariate regression models using BMD and bone microarchitectural varia-
The present study suggests that QUS reflects especially BMD and to a less extent, bone microarchitecture. Several studies have already been done on the relationship among QUS, BMD and microarchitecture and they led to conflicting results [8–16]. Some of them were done on animal cancellous bones [8, 9, 14]. It is not possible draw a conclusion from these studies. Indeed, bone strength is very different in animals than in human beings. In the same manner, some of these studies [8, 9] concluded that QUS reflects bone structure because findings were different according to the three orthogonal axes taken into account. These findings cannot be applicable in clinical practice where the measurements are done in a single direction. Overall, very few studies assessed the relationships between QUS and bone microarchitecture at the calcaneus, which is the site most commonly used for QUS measurements in clinical practice [11, 15, 16]. Our results are not in agreement with 2 of these studies [11, 15]. Hans et al. [11] have shown in a small sample (n = 17) of calcaneus specimens that QUS reflects especially bone quantity since trabecular thickness only explained 67%, 72% and 74% of the variance in SOS, BUA and SI, respectively. Several limitations may be drawn from the study by Hans et al. [11]: (1) the small sample size; (2) the heterogeneity of the sample since very few calcaneus (n = 5) were extracted from necropsy; (3) bone microarchitectural variables measured in this study were not the most relevant. Indeed, in the current study we showed, using stepwise regression analysis, that the 2 most relevant microarchitectural variables were App TBPF and FD, 2 parameters not measured in the study by Hans et al. [11]. The same limitations may be drawn from Ha¨usler et als’. study [15]. Moreover, the precision for BUA (CV: 12%) in the latter study was poor and therefore one should be cautious about the author’s conclusions. By contrast, our findings are very similar to those reported by Nicholson et al. [16] who found a statistical analysis very similar to our own approach, that some of microarchitectural variables correlated with QUS even after adjustment for BMD and particularly Tr Sp. In the same manner they demonstrated, by taking into account bone mass and bone microarchitecture, a greater percentage of the variance in QUS (+8% compared with the use of density alone). In the current study we showed that, by taking into account both BMD and bone microarchitecture, we increased the part of the variance in SI and SOS by 8.4% and 17.8%, respectively compared with BMD alone. Moreover, like Nicholson et al. [16], we found that the
B. Cortet et al.: QUS and Bovine Microarchitecture
65
Table 2. Correlations (simple linear regression analysis) among QUS, BMD and microarchitectural features Variables
BUA
SOS
SI
BMD
BMD (g/cm2) App BV/TV (%) App Tb Th (mm) App Tb Sp (mm) App Tb N (mm)1) TBPF (mm)1) App Euler number (mm)2) App Star volume (mm3) App TSLTN (mm)1) App TSLBM (mm)1) App N Nd (mm)2) App N Tm SC (mm)2) App Nd Nd SC (mm)2) App Nd Tm SC (mm)2) App Tm Tm SC (mm)2) App FD App Nd Nd SL (mm) App Nd Tm SL, mm App Tm Tm SL, mm
0.819*** 0.363 0.091 )0.316 0.143 )0.210 )0.124 )0.134 0.119 )0.254 0.076 )0.016 0.122 )0.005 )0.181 0.063 0.224 )0.050 )0.019
0.752*** 0.511* 0.245 )0.479* 0.356 )0.524** )0.165 )0.268 0.299 )0.223 0.230 )0.101 0.278 )0.014 )0.244 0.101 0.371 )0.176 )0.079
0.854*** 0.490* 0.191 )0.445* 0.284 )0.419* )0.164 )0.226 0.240 )0.174 0.177 )0.068 0.228 )0.012 )0.247 0.096 0.333 )0.130 )0.037
1 0.338 0.031 )0.271 0.115 )0.217 )0.116 )0.091 0.092 )0.062 0.083 )0.030 0.130 )0.069 )0.203 0.032 0.275 )0.048 )0.048
*P < 0.05, **P < 0.01, ***P < 0.001
Fig. 4. Relationship between SOS and BMD at the calcaneus (BMDc) in 24 samples removed after necropsy: adjusted r2 = 0.545.
pattern of relationships between QUS and bone microarchitecture was not consistent across the different correlation approaches used. For example, FD did not correlate with QUS variables using a simple linear regression analysis but was an independent predictor of both SOS and SI using a backward stepwise regression analysis. On the other hand, App Tr Sp correlated with both SOS and SI using a simple linear regression analysis and these correlations remained significant after adjustment for BMD. Nevertheless, App Tr Sp failed to correlate with QUS in the multivariate models. These findings indicate that some variables act by suppressing
or modifying the effect of other variables in the regression and that the interpretation of the results are difficult. One explanation is that certain calcaneus may have almost identical densities but very different structures. Finally, one limitation of the study by Nicholson et al. [16] is that the authors used excised small bone samples (4 · 4 · 4 mm3) put in a water bath and these conditions are very different than the measurement taken in clinical practice. Whereas our study has some strength, also several limitations should be noted: (1) the sample size (n = 24) is small; (2) we did not measure the real microarchitec-
66
B. Cortet et al.: QUS and Bovine Microarchitecture
Fig. 5. Relationship between SI and BMD at the calcaneus (BMDc) in 24 samples removed after necropsy: adjusted r2 = 0.717.
ture but approached it by measuring variables identical to those determined by histomorphometry. Conversely, several studies performed on human bones used a substantially superior image acquisition technique (resolution and noise) compared with the current techniques [11, 13–16]. Due to the effects of physical factors in image acquisition, the method selected in the current study does not provide absolute accuracy in measuring the dimension of the trabecular structure. Indeed, in trabeculae varying from 50–200 lm, only the larger trabeculae are depicted on CT scan whereas the smaller ones are not. By contrast, on histomorphometry, the pixel size is about 5 lm, i.e., greatly thinner than the size of trabeculae. Therefore, what is depicted on CT scan is more a texture image rather than the actual trabecular structure. For example, there was a strong overestimation of apparent BV/TV. However, due to the pixel size (200 lm, i.e., higher than the trabecular thickness), the apparent BV/TV (as compared with the real BV/TV) is necessarily overestimated. Therefore, the very real limitation of the CT-derived structure and texture parameters and their confounding by various imageprocessing steps should be emphasized. Nevertheless, it is remarkable to note that using this tool (conventional CT-system) we obtained relevant findings both in postmenopausal and male osteoporosis [17–19]. We also showed in clinical studies that some of the information obtained with this tool is independent of BMD [17, 19]. Another limitation of the current study is the type of microarchitectural analysis (i.e., 2D and not 3D), which is also a crucial issue. Finally, we studied both men and women and that may be a confounder. However, the pattern of relationships among BMD, QUS and microarchitectural variables was not different for men and women (data not shown). In conclusion, this study suggests that although BMD is a major determinant of acoustic properties of human
calcaneus, significant density-independent relationships with bone microarchitecture should also be taken into account. Therefore, a combination of both BMD and microarchitectural variables may additionally explain 8.4% (SI) to 17.8% (SOS) of the variance in QUS compared with BMD alone. Other studies are needed with large samples to confirm these findings. References 1. Hans D, Dargent-molina P, Schott AM, Sebert JL, Cormier C, Delmas PD, Pouilles JM, Breart G, Meunier PJ (1996) Ultrasonographic heel measurements to predict hip fracture in elderly women: the EPIDOS prospective study. Lancet 348:511–514 2. Bauer DC, Glu¨er CC, Cauley JA, Vogt TM, Ensrud KE, Genant HK, Black DM (1997) Bone ultrasound predicts fractures strongly and independently of densitometry in older women. Arch Intern Med 157:629–634 3. Mele R, Masci G, Ventura V, De Aloysio D, Bicocchi M, Cadossi R (1997) Three-year longitudinal study with quantitative ultrasound at the hand phalanx in a female population. Osteoporos Int 7:550–557 4. Thompson PW, Taylor J, Oliver R, Fisher A (1998) Quantitative ultrasound (QUS) of the heel predicts wrist and osteoporosis-related fractures in women aged 45–75 years. J Clin Densitom 1:219–225 5. Huang C, Ross PD, Yates AJ, Walker RE, Imose K, Emi K, Wasnich RD (1998) Prediction of fracture risk by radiographic absorptiometry and quantitative ultrasound: a prospective study. Calcif Tissue Int 63:380–384 6. Cortet B, Cortet C, Blanckaert F, Racadot A, d’Herbomez M, Marchandise X, Dewailly D (2000) Bone ultrasonometry and turnover markers in primary hyperparathyroidism. Calcif Tissue Int 66:11–15 7. Cortet B, Cortet C, Blanckaert F, D’herbomez M, Marchandise X, Wemeau JL, Decoulx M, Dewailly D (2001) Quantitative ultrasound of bone and markers of bone turnover in Cushing’s syndrome. Osteoporos Int 11:12: 117–123 8. Glu¨er CC, Wu CY, Genant HK (1993) Broadband ultrasound attenuation signals depend on trabecular orientation: an in vitro study. Osteoporos Int 3:185–191 9. Glue¨r CC, Wu CY, Goldstein SA, Genant HK (1994) Three quantitative ultrasound parameters reflect bone structure. Calcif Tissue Int 55:46–52
B. Cortet et al.: QUS and Bovine Microarchitecture
10. Nicholson PHF, Haddaway MJ, Davie MWJ (1994) Dependence of ultrasonic properties on orientation in human vertebral bone. Phys Med Biol 39:1013–1024 11. Hans D, Arlot ME, Schott AM, Roux JP, Kotski PO, Meunier PJ (1995) Do ultrasound measurements on the os calcis reflect more the bone microarchitecture than the bone mass? A two-dimensional histomorphometric study. Bone 16:295–300 12. Strelitzki R, Evans JA, Clarke AJ (1997) The influence of porosity and pore size on the ultrasonic properties of bone investigated using a phantom material. Osteoporos Int 7:370–375 13. Nicholson PHF, Mu¨ller R, Lowet G, Cheng XG, Hildebrand T, Rue¨gsegger P, Van der Perre G, Dequeker J, Boonen S (1998) Do quantitative ultrasound measurements reflect structure independently of density in human vertebral cancellous bone? Bone 23:425–431 14. Wu C, Glu¨er C, Lu Y, Fuerst T, Hans D, Genant HK (1998) Ultrasound characterization of bone demineralization. Calcif Tissue Int 62:133–139 15. Ha¨usler KD, Rich PA, Smith PC, Barry EB (1999) Relationship between static histomorphometry and ultrasound in the human calcaneus. Calcif Tissue Int 64:477–480 16. Nicholson PHF, Mu¨ller R, Cheng XG, Rue¨gsegger P, Van der Perre G, Dequeker J, Boonen S (2001) Quantitative ultrasound and trabecular architecture in the human calcaneus. J Bone Miner Res 16:1886–1892 17. Cortet B, Dubois P, Boutry N, Cotten A, Marchandise X (1999) Image analysis of the distal radius trabecular network using computed tomography. Osteoporosos Int 9:410–419 18. Cortet B, Dubois P, Boutry N, Varlet E, Cotten A, Marchandise X (2000) Does high resolution computed tomography image analysis of the distal radius provide information independent of bone mass? J Clin Densitom 3:339–351 19. Cortet B, Dubois P, Boutry N, Palos G, Cotten A, Marchandise X (2002) Computed tomography image analysis of the calcaneus in male osteoporosis. Osteoporos Int 13:33–41 20. Blanckaert F, Cortet B, Coquerelle P, Flipo RM, Duquesnoy B, Marchandise X, Delcambre B (1997) Contribution of calcaneal ultrasonic assessment to the evaluation
67
21. 22. 23.
24. 25.
26.
27.
28. 29. 30.
31. 32.
of postmenopausal and glucocorticoid-induced osteoporosis. Rev Rhum [Engl Ed] 64:305–313 Marr D, Hildreth EC (1980) Theory of edge detection. Proc R Soc Lond B 207:187–190 Hildreth EC (1983) The detection of intensity changes by computer and biological vision systems. Comput Grap Image Process 22:1–27 Parfitt AM, Mathews CHE, Villanueva AR, Kleerekoper M, Frame B, Raos DS (1983) Relationship between surface volume and thickness of iliac trabecular bone in aging and in osteoporosis: implication for the microanatomic and cellular mechanism of bone loss. J Clin Invest 72:1396–1409 Compston JE, Mellish RWE, Garrahan NJ (1987) Agerelated changes in iliac crest trabecular micro-anatomic bone in man. Bone 8:289–312 Croucher PI, Garrahan NJ, Compston JE (1996) Assessment of cancellous bone structure: comparison of strut analysis trabecular bone pattern factor and marrow space star volume. J Bone Miner Res 7:955–961 Vesterby A, Gundersen HJG, Melsen F (1989) Star volume of marrow space and trabeculae of the first lumbar vertebra: sampling efficiency and biological variation. Bone 10:7–13 Hahn M, Vogel M, Pompesius-Kempa M, Delling G (1992) Trabecular bone pattern factor. A new parameter for simple quantification of bone microarchitecture. Bone 13:327–330 Compston JE (1994) Connectivity of cancellous bone. Assessment and mechanical implications. Bone 15:463–466 Levitz P, Tchoubar D (1992) Disorder porous solids: from chord distributions to small angle scattering. J Phys I France 2:771–790 Chappard D, Legrand E, Basle MF, Fromont P, Racineux JL, Rebel A, Audran M (1996) Altered trabecular architecture induced by corticosteroids: a bone histomorphometric study. J Bone Miner Res 11:676–685 Peleg S, Naor J, Hartley R, Avnir D (1984) Multiple resolution texture analysis and classification. IEEE Trans-pattern analysis and machine intelligence. PAMI-6:661–674 Lynch JA, Hawkes DJ, Buckland-Wright JC (1991) Analysis of texture in macroradiographs of osteoarthritic knees using the fractal signature. Phys Med Biol 36:709–722