Bone 63 (2014) 7–14

Contents lists available at ScienceDirect

Bone

j ourna l homepage: www.e lsev ie r .com/ locate /bone

Original Full Length Article

Cortical measurements of the tibia from high resolution peripheralquantitative computed tomography images: A comparison withsynchrotron radiation micro-computed tomography

Agnès Ostertag a, Françoise Peyrin b,c, Sylvie Fernandez a, Jean Denis Laredo d,Marie Christine de Vernejoul a, Christine Chappard d,⁎a INSERM 606 University Paris Diderot, PRES Sorbonne Paris Cité, 75010 Paris Franceb CREATIS, INSERM U1044, CNRS 5220, Université de Lyon, 69621 Villeurbanne Cedex, Francec ESRF, X-ray Imaging Group, 38043 Grenoble Cedex, Franced B2OA, UMR CNRS7052, University Denis Diderot, PRES Sorbonne Paris Cité, 75010 Paris, France

⁎ Corresponding author at: B2OA, UMR CNRS 7052, USorbonne Paris Cité, 10 avenue de Villemin, 75010 Paris, F

E-mail address: christine.chappard@inserm.fr (C. Chap

http://dx.doi.org/10.1016/j.bone.2014.02.0098756-3282/© 2014 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 12 September 2013Revised 18 February 2014Accepted 18 February 2014Available online 26 February 2014

Edited by: Sharmila Majumdar

Keywords:Cortical boneX-rayHigh resolution peripheral quantitativecomputed tomographySynchrotron radiation micro-computedtomography

High resolution-peripheral quantitative computed tomography (HR-pQCT) measurements are carried out inclinical research protocols to analyze cortical bone. Micro-computed tomography (micro-CT) is a standard toolfor ex vivo examination of bone in 3D. The aim of this work was to evaluate cortical measurements derivedfrom HR-pQCT images compared to those from synchrotron radiation (SR) micro-CT in a distal position(4.2 cm from the distal pilon).Twenty-nine tibia specimens were scanned with HR-pQCT using protocols provided by the manufacturer. Thestandard measured outcomes included volumetric bone density (g HA/cm3) of the cortical region (Dcomp),and the cortical thickness (Ct.Th, mm). New features, such as cortical porosity (Ct.Po) and mean pore diameter(Ct.Po.Dm), were measured by an auto-contouring process. All tibias were harvested from the posterior regionand imaged with SR micro-CT (voxel size = 7.5μm). The cortical thickness, (Ct.Thmicro-CT), porosity (PoV/TV),pore diameter, pore spacing, pore number, and degree of mineralization of bone (DMB) were obtained for SRmicro-CT images. For standard measurements on HR-pQCT images, site matched analyses with micro-CT werecompleted to obtain Dcomplocal and Ct.Thlocal.Dcomp was highly correlated to PoV/TV (r =−0.84,p b 10−4) but not to DMB. Dcomplocal was correlated toPoV/TV (r =−0.72, p b 10−4) and to DMB (r = 0.40, p N 0.05). Ct.Thlocal and Ct.Thmicro-CT were moderatelycorrelated (r = 0.53,p b 0.01). Ct.Th and Ct.Po results from the autocontouring process are influenced bythe level of trabecularization of the cortical bone and need manual correction of the endosteal contour.Distal tibia is a reliable region to study cortical bonewithDcompas the best parameter because it reflects both themicro-porosity (Havers canals) and macro-porosity (resorption lacunae) of the cortical bone.

© 2014 Elsevier Inc. All rights reserved.

Introduction

Cortical bonehas a role in skeletal bone loss in appendicular skeletonand has become an indicator to be evaluated for the prevention offracture [1,2]. Bone resistance to rupture depends on 40 to 90% of corti-cal bone at femoral neck [2]. It has been shown that small changes inporosity have a significant impact on the mechanical properties ofcortical bone [3]. Increases in cortical porosity and cortical thinningare both related to aging and bone fragility [4]. Cortical porosity de-pends on bone resorption at the endosteal surface and on the surface

niversity Denis Diderot, PRESrance. Fax: +33 157278570.pard).

of Haversian canals [5]. Comparedwith trabecular bone (which consistsof a network of trabeculae), cortical bone is composed of a complexnetwork of canals with a dynamic behavior that needs to be completelycharacterized by 3 dimensional (3D) analysis methods [6].

High resolution peripheral quantitative computed tomography(HR-pQCT) with the default isotropic voxel size of 82 (μm)3 wasused for the analysis of bone at the radius and tibia with the capabilityof separately evaluating cortical bone and trabecular bone [7–16]. Thismethod is able to follow longitudinally both the trabecular micro-architecture and cortical parameters under the effect of treatment[9,10,15]. Moreover, this examination is easy and fast with a loweffective dose (4.8 μSv) [11]. The typical cortical parameters are corticaldensity and thickness; these parameters are sensitive to aging [12,13],fracture history [14,16], effects of treatment [15], and can probablygivemore information than bonemineral density acquired byDXA [16].

Fig. 1. Site locations SR-micro-CT acquisitions comparatively to HR-pQCT images.

8 A. Ostertag et al. / Bone 63 (2014) 7–14

New quantitative parameters, such as cortical porosity (Ct.Po) andmean cortical pore diameter (Ct.Po.Dm), have recently been introduced[17,18].

Trabecular parameters derived from HR-pQCT have been comparedto those obtained by conventional micro-computed tomography(micro-CT) on femoral head biopsies [19,20], at the radius [21,22]and tibia [7,23]. Cortical HR-pQCT parameters were compared to theircounterparts from micro-CT images [7,21,23].

Micro-CT has become a standard tool for 3D examination oftrabecular bone. However, this method is less often used to explorethe internal cortical bone structure [24,25]. Synchrotron radiation (SR)micro-CT, which is considered the gold standard for micro-CT (thereconstruction is not affected by the cone beam and beam hardeningartifacts), has also been proposed for the analysis of cortical bone[26–28].

With this technique, it is possible to characterize the spatial arrange-ment of the pore network [28]. The most commonly used parametersto describe the spatial distribution of cortical porosity are porediameter (Po.Dm), pore number (Po.N), and pore separation (Po.Sp)[24,25,27,28]. On SR micro-CT images, it is also possible to measurethe local degree of bone mineralization (DMB) independent of porositymeasurements [28].

The main purpose of this study was to compare the informationprovided by quantitative cortical measurements (standard measure-ments and new features) from HR-pQCT images with the microstruc-ture of the cortical tibia.

To accomplish this purpose, cortical bone parameters obtainedfrom HR-pQCT were compared to cortical parameters derived from SRmicro-CT images which were used as a reference. Bone measurementswere obtained from a distal position of the tibia with a larger amountof cortical bone than at the usual ultra-distal position.

Material and methods

Specimens

Twenty-nine tibias (19 women, 10 men; mean age: 82.5 ± 10.5)were collected at the Institute of Anatomy Paris Descartes. No datawere available regarding the cause of death, previous illnesses, ormedical treatments of these individuals. Collection of these humantissue specimens was conducted according to pertinent protocolsestablished by the Human Ethics Committee at INSERM. Soft tissueswere removed from the tibias which were stored at−20 °C.

HR-pQCT scanning

The intact tibias were scanned at 22.5 mm from the tibial pilon(ultra-distal scan) using the standard protocol provided by the manu-facturer (Xtreme-CT; Scanco Medical AG, Brüttiselen, Switzerland) forclinical evaluation. A second scan was performed at 42.5 mm from thedistal tibial pilon (distal scan) as suggested from a previous study toobtain more cortical bone [29]. The acquisition scan and quality controlare detailed in Boutroy et al. [30]. The reproducibility of the entire chainof measurements was performed on 14 tibias for both protocols (ultra-distal and distal). Two scans were performed with repositioning by thesame observer on the same day.

SR micro-CT scanning

After HR-pQCT acquisitions, bone samples including the cortex andadjacent trabecular bone of the posterior region of the tibia wereharvested (1.5 cm in width and 2 cm in height) with a handsaw, thedistal limit of the bone sample being located at a distance of 3.5 cmfrom the tibial pilon.

All bone sampleswere imaged at the European SynchrotronRadiationFacility (ESRF, Grenoble) on beamline ID19. A detailed description of the

facility has been previously reported [31]. The samples were centered toobtain a central volume of 1 cm high corresponding to the same positionof the distal HR-pQCT scan and the optical system was configured tocover the entire cortical thickness within the field of view. The sampleswere placed vertically to obtain 2 dimensional (2D) projections perpen-dicular to the Haversian system.

The detector is a 2048 × 2048 scintillator coupled to a 14-bit chargecoupled device (CCD). The energy was set to 30 keV, 3500 views wereacquired over a total angular range of 360°. The exposure time foreach view was 0.3 s with a total acquisition time of approximately30 min per sample. The 3D images were reconstructed from the set of2D radiographic projections using the filtered back-projection method.The size of the voxels in the reconstructed images was 7.5 μm in all 3dimensions. The location of the sample cores in the tibia is illustratedin Fig. 1.

Cortical bone measurements on HR-pQCT images

Standard measurementsThe imageswere processedwith the default clinical evaluation proto-

col to obtain standard cortical geometric and density measurements atboth the ultra-distal tibia and distal tibia. The images were also analyzed

Fig. 2. Examples of HR-pQCT analyses on the distal tibia: (a) autocontouring processwithout correction, (b) autocontouring process with manual correction, and (c) sitematched analysis with SR micro-CT.

9A. Ostertag et al. / Bone 63 (2014) 7–14

in a site matched area corresponding to the posterior region harvestedfor the micro-CT acquisitions. The details of the methods of measure-ments are described in the following paper [30]. The outcome variablesused in the analyses included volumetric bone density (g HA/cm3)for the entire (D100), and compact bone (Dcomp) correspondingto the cortex. Morphological parameters, such as cortical thickness(Ct.Th, mm) and cortical area (mm2), were obtained from the cortex.

Quantitative measurements of cortical bone microarchitectureTo obtain quantitative measurements of cortical bone micro-

architecture, we used the advanced cortical analysis routine from thescanner manufacturer. The system is based on an auto-contouringprocess which generated periosteal and endosteal contours [11]. Thisprocess allowed measurements of cortical pore diameter (Ct.Po.Dm),cortical pore volume (Ct.Po.V) and cortical bone volume (Ct.BV) to beobtained. Based on this segmentation, cortical porosity (Ct.Po) wascalculated as follows:

Ct:Po %ð Þ ¼ Ct:Po:V= Ct:Po:Vþ Ct:BVð Þ: ð1Þ

Ct.BMD is themean bonemineral density of the cortical bone volumeof interest (VOI)which includes the pore space. The automatically gener-ated contours were qualitatively inspected. When the contour visuallydeviated from the apparent endosteal margin, minor manual adjust-ments were performed to the affected region alone. No correctionswere performed on the periosteal contours. To test the reproducibilityof the image analysis, the images from 10 subjects were analyzed bythree observers (one skilled and two unskilled). An illustration of differ-ent VOIs is shown in Fig. 2.

Parameter calculations from SR micro-CT images

All parameters were calculated using CTAn® software Skyscan,Kontish, Belgium. Cortical thickness (Ct.Thmicro-CT) was manuallymeasured in 2 locations (top and bottom positions of the image) forevery 20 slices on the whole volume as previously described [28].

The gray levels of the images were inverted to evaluate themorphology of the pores with a unique threshold used for the wholedataset (138). The VOI selection was performed using a semi-automatic procedure based on manual contouring of the cortex on aselected number of slices followed by interpolation. All pores (primaryand secondary osteons)were included in the VOI. The total height of theVOI for SRmicro-CT analysis was 1 cm at the same location of HR-pQCTimages. The following morphological parameters were calculated: thevolume of the pore / volume of total volume (PoV / TV, %), the porediameter (Po.Dm, mm) and the pore spacing (Po.Sp, mm) withthe model-independent method described by Hildebrand [32]. Thesurface of the canals (PoS, mm−2) was obtained by triangulation ofthe surface of the mineral phase, and the volume of the pore (PoV)was calculated using tetrahedrons corresponding to the volumeenclosed within the triangulated surfaces [33]. The average number ofpores (Po.N, mm−1) was defined as the inverse of the average distancebetween the skeleton of the structure [34].

The maps of the linear attenuation coefficient at 30 keV were con-verted into degrees of mineralization of bone (DMB). The gray scalewas calibrated with the local concentration of HA in the examinedbone tissue; the data were expressed in g/cm3 using a technique de-scribed previously [31]. Examples of 3D images obtained with SRmicro-CT and binarized are shown in Fig. 3.

To assess whether the difference between HR-pQCT and micro-CTparameters could be explained by the difficulty to delineate the cortexlimit because of macro-porosity and trabecularization of the endostealsurface, a visual grading system was employed to classify the corticalbone aspect on SR micro-CT images into four classes. Class A (Cl.A)represents that the cortex was well-delineated with only insidemicro-porosity (n = 8). Class B (Cl.B) establishes that the cortex was

well-delineated with resorption cavities present within (n = 9). ClassC (Cl.C) represents a well-individualized cortex, but the endosteal con-tour is not continuous (n = 6). Finally, Class D (Cl.D) corresponds to alarge trabecularization of the cortex (n=6). An illustration is presentedin Fig. 3.

Statistical analysis

All statistical computations were performed with NCSS (2004,Kaysville, UT). To compare themethods, univariate correlation analysesby Pearson correlation coefficients replaced by Spearman correlationcoefficients in case of lack of normality were performed. The coefficientof determination was used for multivariate analysis to define therelationship between the cortical HR-pQCT measurements and the ref-erence measurements from the SR micro-CT images. A paired t-testwas used and replaced by a Wilcoxon test in case of lack of normalityto compare 1) ultra-distal and distal measurements, and 2) auto-contouring process with and without manual correction on HR-pQCT.

The root mean square coefficient of variation (RMSCV, %) was calcu-lated for reproducibility analysis [35]. The standardized CVwas derived

Fig. 3. Examples of 3D binarized images of the pore network in cortical bone from SRmicro-CT in four specimens divided into four classes (a: well-delineated cortexwith micro-porosity (n = 8), b: well-delineated cortex with macro-porosity (n = 9),c: cortex without continuous endosteal border (n = 6), d: large trabecularization of thecortex (n= 6)). The periosteal edges are at the right side and at the left side the endostealborders are delineated.

Table 2Mean ± standard deviation of cortical bone parameters:synchrotron radiation micro-CT (n = 29).

SR micro-CT

Ct.Thmicro-CT (mm) 1.55 ± 0.26PoV (mm3) 44.9 ± 29.8PoV/TV (%) 26.6 ± 13.0PoS/PoV (1/mm) 29.2 ± 14.9Po.Dm (mm) 0.28 ± 0.11Po.N (1/mm) 0.94 ± 0.23Po.Sp (mm) 0.32 ± 0.04DMB (mg/cm3) 1.32 ± 0.03

DMB = degree of mineralization.

10 A. Ostertag et al. / Bone 63 (2014) 7–14

by normalizing the CV to the variability found in the group to thefollowing relationship:

sCV %ð Þ ¼ RMSCV %ð Þ= 4SD=meanð Þ: ð2Þ

Results

The mean and standard deviations of the cortical parameters for theultra-distal and distal variables are presented in Table 1. Ct.Th wastwice as high at the distal site than at the ultra-distal site and significantlydifferent (p b 10−4). Mean and standard deviations for the distal vari-ables from the auto-contouring process performed on the distal scansare presented also in Table 1. Between the completely automatic processversus manual correction of the endosteal contouring, there was nosignificant difference. Mean ± standard deviations of the SR micro-CT

Table 1HR-pQCT results: mean ± standard deviation of the ultra-distal and distal tibia variables withcorrection of the endosteal contour at the distal tibia for the whole contour and site matched a

Standard protocol

D100(mg HA/cm3)

Dcomp(mg HA/cm3)

Ct.Th(mm)

Ultra-distal scan 237.8 ± 74.6 753.8 ± 102.4 0.70 ± 0.32Distal scan 333.4 ± 92.4 902.1 ± 63.1 1.46 ± 0.39Without correctionWith manual correctionSite matched analysis 877.9 ± 74.5 0.65 ± 0.15

are reported in Table 2. For Po.Dm, the percentage distribution accordingto size is as follow: 22.6% from 8 μm2 to 98 μm2, 22.9% from 98 μm2 to188 μm2, 14.8% from 188 μm2 to 278 μm2, 11.7% from 278 μm2 to368 μm2and 27.9%when N368 μm2. For Ct.Thmeasured in a sitematchedarea, there is a significant difference between SR micro-CT and HR-pQCTmodalities (p N 10−4). The Bland and Altman plot between Ct.ThMicroCT

and Ct.Thlocal assessed in a site matched region with the standard mea-surement (Fig. 4) shows a mean bias of −0.9 mm with a minimumbias for Cl.A specimens and a maximum bias for Cl.D specimens. Wenext assessed whether the difference between HR-pQCT without correc-tion andwithmanual correction of the auto-contouring process could beexplained by the difficulty to delineate the cortex limit because ofmacro-porosity and trabecularization of the endosteal surface by Bland and Alt-man plots (Fig. 5). The mean bias was respectively 0.006 mm for Ct.Th,−30.8 mg/cm3 for Ct.BMD and 2.4% for Ct.Po. In most of cases of Cl.Dspecimens, the Ct.Th and Ct.BMD were overestimated and Ct.Po wasunderestimated in the lack of correction of the manual contour. The var-iation between the two measurements: Ct.Th auto-contour with manualcorrection (Ct.ThAutoC_Corr) and Ct.Th auto-contour without manual cor-rection (Ct.ThAutoC) was not influenced by DMB.

The correlation analysis between all cortical measurements per-formed by the different protocols on HR-pQCT images and SR micro-CT parameters is shown in Table 3. Using the standard protocol, themost correlated parameters with PoV/TV were Dcomp (r = −0.84,p b 10−4) and Dcomplocal (r = −0.72, p b 10−4). None of the parame-ters were correlated with DMB except Dcomplocal (r = 0.40, p b 0.05).When pores below 88 μm and in a second step below 180 μm were re-moved from the SR micro-CT porosity measurements, the correlationcoefficients with Dcomplocal became r = −0.69, p b 10−4 and r =−0.65, p b 10−4, respectively. When combining PoV/TV and DMB toexplain the Dcomplocal results, we obtained: r2 = 0.64, p b 10−4.Ct.Thmicro-CT was correlated with their counterparts from HR-pQCT:r = 0.58, p N 0.01, and r = 0.53, p N 0.01 for Ct.Th and Ct.Thlocal, re-spectively. Using auto-contouring without corrections the absolutecorrelation coefficients were below 0.46 for all parameters. Usingthe manual correction improved the correlations: the correlationcoefficient between Ct.Po measured by HR-pQCT with PoV/TV mea-sured by SR-micro CT reached 0.66 (p b 0.001).

standard protocol and auto-contouring process without any correction and with manualnalysis with HR-pQCT measurements (n = 29).

Auto-contouring process

CtBMD(mg/cm3)

Ct.ThAut

(mm)CtPo(%)

CtPoV(mm3)

CtPoDm(mm2)

964 ± 63.0 1.60 ± 0.44 5.9 ± 2.6 61.1 ± 31.0 0.23 ± 0.04934 ± 82.4 1.60 ± 0.40 8.3 ± 4.0 89.6 ± 45.0 0.25 ± 0.05

Fig. 4. Bland and Altman plots with mean bias ± 2 SD between Ct.ThMicroCT and Ct.Thlocal assessed in a site matched region with the standard measurement.

11A. Ostertag et al. / Bone 63 (2014) 7–14

For HR-pQCT evaluation, the reproducibility of the whole chain ofmeasurement (two scans) for the standard protocol is reported inTable 4. The reproducibility of Dcomp was found to be better on thedistal region of the tibia than on the ultra-distal tibia 0.75%, and 4.6% re-spectively. On the contrary, the reproducibility of Ct.Th measurementswas better at the ultra-distal site than at the distal site (0.01%, and0.41% respectively). The Ct.Th standardized RMSCV(%) was better withthe standard protocol than with the auto-contouring process, 0.41%and 1.23%,respectively. For the auto-contouring process performed ontwo different scans, the standardized RMSCVs(%) were 1.6%, 2.2%and 2.9% for Ct.BMD, Ct.Po and Ct.PoDm, respectively. For the auto-contouring process correction, the inter-observer (one skilled and twounskilled) reproducibility was 3.70%, 5.3%, 6.2% and 6.1% for Ct.Th,Ct.BMD and Ct.Po and Ct.PoDm, respectively.

Discussion

Ex vivo micro-structural cortical measurements from HR-pQCT im-ages with an 82 (μm)3 isotropic voxel size were validated againstmicro-structural cortical measurements from SR-micro-CT images at7.5 (μm)3 isotropic voxel size. To the best of our knowledge, this studyis the only validation of cortical parameters from the HR-pQCT to havebeen performed using SR micro-CT images as reference. Most of thesemeasurements were correlated with their respective gold standards:Dcomp versus PoV/TV, Ct.Th versus Ct.Thmicro-CT for the standardmodality, Ct.Th versus Ct.Thmicro-CT and Ct.Po versus PoV/TV for auto-contouring modality with manual correction, which was unlike theauto-contouringmodality withoutmanual correction.We studied a dis-tal scan because Ct.Th at a distal site is almost two times thicker thanCt.Th at the ultradistal site. It was reported that the distal location, com-pared to the ultradistal site, is less affected by partial volume artifacts[29].

Our results indicated that themost related parameters to cortical po-rosity are Dcomp and Ct.BMD. The removal of pores below 180 μm didnot improve the correlation analysis, these results indirectly show theinfluence of small pores in the Dcomp measurement. These two valuesmeasure both the macro-porosity (due to osteonal remodeling) andthe micro-porosity (due to the Haversian canals) [8]. When measuring

cortical porosity on the HR-pQCT images, only large resorption cavitiesin bone are captured. In our study, 22.7% of pores have a diameterbelow 100 μm2, and consequently, they could not be assessed directlyby HR-pQCT.

This study was the first time that DMB measurements (meanvalue = 1.3 g/cm3) were performed on cortical bone of the tibia. Ina previous study, 1.3 g/cm3 corresponds to the upper bound of theDMB for the iliac crest values acquired under the same procedure[31]. The only significant correlation coefficient with DMB wasDcomplocal, moreover when DMB is added to porosity, more varianceof the measurements Dcomp assessed locally was explained. Our re-sults are concordant with those of Jepsen et al. who have found thatcortical total mineral density with a low resolution pQCT was morecorrelated with PoV/TV (r2 = 0.51) than ash density measured bygravimetric methods (r2 = 0.34) [36]. We did not find a significantinfluence of DMB on the Ct.Th results before and after the manualcorrection of the contour probably because in our study all DMBvalues were sufficiently high to allow reliable detection of the con-tours. In previous studies where the image analysis was performedin overlapped regions using a fusion method, the determination co-efficients on cortical bone parameters between HR-pQCT parametersand the morphological parameters from micro-CT were notablyhigh: r2 ranged from 0.66 to 0.98 at the radius, and r2 was 0.90 atthe tibia for Ct.Th [7,21]. In our study, the correlations were lessmarked, and the volumes of interest were manually selected andconsequently did not exactly overlap. However, this cannot explainthe entire of the discrepancywhich is due inmajority by the large dif-ference (about 10 times) of resolution between the two modalities. Weanalyzed a subsample of the tibial cortex, which represented 25% of thetotal cortical contour. The standard algorithm provided by the manufac-turer for edge detection was validated for the total contour and was notdesigned to analyze a local region. Different algorithms for the SRmicro-CT and XtremeCT data were used, the quality of SR micro-CT images al-lows us to perform a single and simple thresholding for the whole data-base which separates, unambiguously, bone from pores. In addition, SRmicro-CT images do not require any post-treatment contrary to XtremCTdata. Consequently, the correlations we obtained were not notably highcompared to those found in the literature.

Fig. 5. Bland and Altman plots with the mean bias and ±2 SD to compare theautocontouring process with and without manual correction for Ct.Th, Ct.BMD, and Ct.Po.

12 A. Ostertag et al. / Bone 63 (2014) 7–14

For the cortical thicknessmeasurements, protocols use either a fixedattenuation threshold or either an advanced segmentation recently de-scribed by Buie et al. In the standard protocol, the cortexwas segmentedfrom the grayscale image with a Gaussian filter and secondarilythresholded, cortical thickness was derived from the Ct cross-sectionalarea (mm2) divided by the circumference [17]. In the advanced segmen-tation method, Ct.ThAutoC was measured from the VOI defined by theauto-contour. No correction was made in the periosteum because ofits high visibility, as observed in Fig. 1. In all cases, regardless of class,the periosteumwas clearly individualized with no surrounding soft tis-sue. Consequently, there was a good contrast at the interface, but thissituation may be different in vivo in the presence of soft tissue. TheCt.Thlocal parameter was surprisingly underestimated comparatively toCt.ThmicroCT whereas the Ct.Th parameter of the whole contour whatev-er the method of measurements was of the same order, the underesti-mation being maximum for Cl.D specimens and minimum for Cl.A.This error probably occurred mostly when analyzing small local areas.

The explanation is probably due to the algorithm itself whose perfor-mances depend on the quantity of analyzed cortical bone.

Most bone loss is cortical and occurs after age 65 due to intracortical,rather than endocortical or trabecular, remodeling [1]. In our study, only31% of the tibial cortices had only intra-cortical resorptionwith an intactendosteal border seen on SR micro-CT images. However, for specimensclassified Cl.B presenting resorption lacunae inside the cortex butmain-ly located close the endosteal border, there is possibly awrong detectionof the endosteal border in HR-pQCT, which is clearly visible only in theSR micro-CT images. The measurements of cortical bone porosity onlyquantified intra-cortical pore volume; the measurements excluded theperiosteal and endosteal voids with the aim of only taking osteonalremodeling into account. These measurements sought to exclude theeffects of endosteal remodeling and other artifacts due to determinationof endosteal and periosteal contours.

In Cl.D where the porosity is more than 30%, with a discontinuity ofthe endosteal border, the correction of the auto-contour has a realimpact on the cortical porosity results. When resorption cavities arelargely open in the trabecular zone, these cavities are not comprised inthe Ct.Po.V measurements. Manual correction of the contour had theeffect of smoothing the contour and integrated most of endostealvoids. The effectiveness of the autocontouring process is affected bythe level of trabecularization of the cortical bone. In the case of a poorlydefined endosteal border, only a high resolution imaging systemby syn-chrotron radiation was able to localize the real limitation and visualizethe osteonal shape that characterizes cortical bone contrary to theconventional micro-CT for which the endosteal border is more difficultto define.

The combination of DMB and PoV/TV increased the determinationcoefficient of Dcomplocal to r2 = 0.64 from a correlation of r = 0.72between Dcomp and PoV/TV. The coefficient of determination wasprobably underestimated due to an imperfect site matching betweenthe analyzed areas in HR-pQCT and SR micro-CT.

TheCt.Th standardized RMSCVvalueswere for the standardprotocoland the auto-contour process without any manual correction were0.41% and 1.46%, respectively, these results demonstrate that thestandard protocol seems to be the most reliable. Despite a relativelylarge inter-observer reproducibility of the manual correction, the preci-sion results remain favorable compared to the biological variability,which was 20 times for Ct.Th and 90 times for Ct.Po. Regardless of theresults, analysis from a skilled observer is recommended.

Previous studies only used conventional micro-CT to validateHR-pQCT measurements [7,19–23]. Compared to conventionalmicro-CT, SR micro-CT has many advantages: the X-ray is monochro-matic, does not diverge and has a high X-ray of intensity. The contrastand signal-to-noise ratio are high and facilitate thresholding of theimage. In addition, because the beam has a parallel geometry, thereare no distortions due to geometric magnification and tomographicreconstruction. Size measurements from the system are exact. Thethresholding process is able to separate bone from pores (fromHaversian system and resorption lacunae) with no ambiguity [28].This process is the method of choice for studying cortical bone porosityand cortical bone arrangement. In addition, all comparisonswithmicro-CT images as a reference were performed with a resolution rangingfrom 19 to 25 μm3. It was later suggested that the spatial resolutionmust be better than 10 μm to obtain consistent porosity data [7]. Thevoxel size used in this study was 7.5 μm, which allowed us to measurethe complexity of the pore network and to see the osteonal border,the latter is important to visualize with the aim to segment correctlytrabecular bone to cortical bone [28].

Conclusion

HR-pQCT measurements are carried out in clinical research proto-cols to analyze cortical bone from which cortical bone measurementsprovide consistent results with good correlations of Dcomp, Ct.Th and

Table 3Correlation analysis between SR-micro-CTmeasurements andHR-pQCTmeasurements using Pearson correlation coefficients (in italic: Spearman coefficient) in the case of non-normality(n = 29).

PoV(mm3)

PoV/TV(%)

PoS/PoV(1/mm)

Po.Dm(μm)

Po.N(1/mm)

Po.Sp(μm)

Ct.Thmicro-CT

(mm)DMB(g/cm3)

Standard protocol (site matched)Dcomplocal(mg HA/cm3)

−0.60¥ −0.72¥ 0.51¥ −0.50¥ −0.69¥ 0.60¥ ns 0.40⁎

Ct.Thlocal(mm)

−0.58¥ −0.72¥ 0.60⁎ −0.58⁎⁎ −0.55⁎⁎ 0.61¥ 0.53⁎⁎ ns

Standard protocol (whole contour)Dcomp(mg HA/cm3)

−0.74¥ −0.84¥ 0.66¥ −0.64¥ −0.75¥ 0.74¥ ns ns

Ct.Th(mm)

−0.52¥ −0.68¥ 0.56⁎ −0.53⁎⁎ −0.52⁎⁎ 0.66¥ 0.58⁎⁎ ns

Auto-contouring without correction (whole contour)Ct.Thauto(mm)

ns ns ns ns ns ns ns ns

CtBMD(mg/cm3)

ns ns ns ns ns ns ns ns

CtPoV(mm3)

−0.42⁎ −0.45⁎ 0.40⁎ ns −0.40⁎ ns ns ns

CtPo(%)

−0.42⁎ −0.42⁎ ns ns −0.38⁎ ns ns ns

CtPoDm −0.46⁎ −0.38⁎ 0.42⁎ ns ns ns ns ns

Auto-contouring with manual correction (whole contour)Ct.Thauto

(mm)−0.59¥ −0.67⁎⁎ 0.62¥ −0.55⁎⁎ −0.50⁎⁎ 0.54⁎⁎ 0.56⁎⁎ ns

CtBMD(mg/cm3)

−0.67¥ −0.75¥ 0.56⁎⁎ −0.56⁎⁎ −0.67¥ 0.71¥ ns ns

CtPoV(mm3)

0.51⁎⁎ 0.45⁎ −0.41⁎ 0.37⁎ 0.37⁎ ns ns ns

CtPo(%)

0.69¥ 0.66¥ −0.63¥ 0.54⁎⁎ 0.50⁎ −0.40⁎ ns ns

CtPoDm(mm2)

0.51⁎⁎ ns −0.50⁎⁎ 0.45⁎ ns ns ns ns

DMB: degree of mineralization.⁎ p b 0.05.⁎⁎ p b 0.01.¥ p b 0.001.

13A. Ostertag et al. / Bone 63 (2014) 7–14

Ct.Po with their respective gold standards from SR micro-CT. The distaltibia instead of the ultra-distal tibiamust be considered as an interestingsite for cortical bone analysis with HR-pQCT. Using the standard proto-col, Dcomp is a parameter of choice highly related to cortical boneporosity because it considers both the micro-porosity (Haversiansystems) and macro-porosity (resorption lacunae) of the corticalbone. This later parameter is related in a lesser extend to the degree ofmineralization. Porosity measurements from HR-pQCT images were

Table 4Reproducibility of thewhole chain ofmeasurements on two acquisitions for both protocols (n=are in italics.

RMSCV(%) Acquisition

Standard protocol

Ultradistal Distal

D100(mg HA/cm3)

0.37/0.29 0.33/0.30

Dcomp (mg HA/cm3) 2.58/4.62 0.21/0.75Ct.Th(mm)

0.02/0.01 0.44/0.41

Ct.BMD(mg/cm3)

CtPo(%)

CtPo.V(mm3)

CtPo.Dm(mm2)

low correlated than Dcomp with PoV/TV assessed on SR micro-CT im-ages. Using the auto-contouring process to measure Ct.Po and Ct.Th atthe distal tibia which requires a visual checking of the endostealcontouring should always be verified and corrected if necessary.

Conflict of interest statement

None.

14). Image analysiswas performed by 3 observers (n= 10). Standardized RMSCV (%)

Image analysis

Auto-contour without correction Auto-contour with correction

Distal Distal

1.2 3.7

1.6 5.3

2.2 6.2

4.6 7.9

2.9 6.1

14 A. Ostertag et al. / Bone 63 (2014) 7–14

Acknowledgments

We would like to thank Stéphanie Boutroy for these advices.

References

[1] Zebaze RM, Ghasem-Zadeh A, Bohte A, et al. Intracortical remodelling and porosityin the distal radius and post-mortem femurs of women: a cross-sectional study. Lan-cet 2010;15(375):1729–36.

[2] Burr DB. Cortical bone: a target for fracture prevention. Lancet 2010;15(375):1672–3.[3] Yeni YN, Norman TL. Fracture toughness of human femoral neck: effect of microstruc-

ture, composition, and age. Bone 2000;26:499–504 [Erratum in: Bone. (2000) 27:327].[4] Seeman E. Pathogenesis of bone fragility in women and men. Lancet

2002;25(359):1841–50.[5] Parfitt AM. Osteonal and hemi-osteonal remodeling: the spatial and temporal frame-

work for signal traffic in adult human bone. J Cell Biochem 1994;55(3):273–86.[6] Stout SD, Brunsden BS, Hildebolt CF, et al. Computer-assisted 3D reconstruction of

serial sections of cortical bone to determine the 3D structure of osteons. Calcif TissueInt 1999;65:280–4.

[7] Liu XS, Zhang XH, Sekhon KK, et al. High-resolution peripheral quantitative comput-ed tomography can assess microstructural and mechanical properties of humandistal tibial bone. J Bone Miner Res 2010;25:746–56.

[8] Burghardt AJ, Kazakia GJ, SodeM, et al. A longitudinal HR-pQCT study of alendronatetreatment in postmenopausal women with low bone density: relations amongdensity, cortical and trabecular microarchitecture, biomechanics, and bone turnover.J Bone Miner Res 2010;25:2282–95.

[9] Burghardt AJ, Kazakia GJ, Ramachandran S, et al. Age- and gender-related differencesin the geometric properties and biomechanical significance of intracortical porosityin the distal radius and tibia. J Bone Miner Res 2010;25:983–93.

[10] Rizzoli R, LarocheM, KriegMA, et al. Strontium ranelate and alendronate have differ-ing effects on distal tibia bone microstructure in women with osteoporosis.Rheumatol Int 2010;30:1341–8.

[11] Burghardt AJ, Buie HR, Laib A, Majumdar S, Boyd SK. Reproducibility of direct quan-titative measures of cortical bone microarchitecture of the distal radius and tibia byHR-pQCT. Bone 2010;47:519–28.

[12] Dalzell N, Kaptoge S, Morris N, et al. Bone micro-architecture and determinants ofstrength in the radius and tibia: age-related changes in a population-based study ofnormal adultsmeasuredwith high-resolution pQCT. Osteoporos Int 2009;20:1683–94.

[13] Macdonald HM, Nishiyama KK, Kang J, Hanley DA, Boyd SK. Age-related patterns oftrabecular and cortical bone loss differ between sexes and skeletal sites: apopulation-based HR-pQCT study. J Bone Miner Res 2011;26:50–62.

[14] Vico L, Zouch M, Amirouche A, et al. High-resolution pQCT analysis at the distal ra-dius and tibia discriminates patients with recent wrist and femoral neck fractures. JBone Miner Res 2008;23:1741–50.

[15] Seeman E, Delmas PD, Hanley DA, et al. Microarchitectural deterioration of corticaland trabecular bone: differing effects of denosumab and alendronate. J Bone MinerRes 2010;25:1886–94.

[16] Sornay-Rendu E, Boutroy S, Munoz F, Delmas PD. Alterations of cortical and trabec-ular architecture are associated with fractures in postmenopausal women, partiallyindependent of decreased BMD measured by DXA: the OFELY study. J Bone MinerRes 2007;22:425–33.

[17] Buie HR, Campbell GM, Klinck RJ, MacNeil JA, Boyd SK. Automatic segmentation ofcortical and trabecular compartments based on a dual threshold technique forin vivo micro-CT bone analysis. Bone 2007;41:505–15.

[18] Burghardt AJ, Buie HR, Laib A, Majumdar S, Boyd SK. Reproducibility of direct quan-titative measures of cortical bone microarchitecture of the distal radius and tibia byHR-pQCT. Bone 2010;47:519–28.

[19] Laib A, Rüegsegger P. Calibration of trabecular bone structure measurements ofin vivo three-dimensional peripheral quantitative computed tomography with 28-microm-resolution microcomputed tomography. Bone 1999;24:35–9.

[20] Laib A, Häuselmann HJ, Rüegsegger P. In vivo high resolution 3D-QCT of the humanforearm. Technol Health Care 1998;6:329–37.

[21] Nishiyama KK, Macdonald HM, Buie HR, et al. Postmenopausal women withosteopenia have higher cortical porosity and thinner cortices at the distal radiusand tibia than women with normal aBMD: an in vivo HR-pQCT study. J BoneMiner Res 2010;25:882–90.

[22] Krug R, Carballido-Gamio J, Burghardt AJ, et al. Assessment of trabecular bone struc-ture comparingmagnetic resonance imaging at 3 Tesla with high-resolution periph-eral quantitative computed tomography ex vivo and in vivo. Osteoporos Int2008;19:653–61.

[23] Tjong W, Kazakia GJ, Burghardt AJ, Majumdar S. The effect of voxel size on high-resolution peripheral computed tomography measurements of trabecular and corti-cal bone microstructure. Med Phys 2012;39:1893–903.

[24] Cooper DML, Turinsky AML, Jensen CW, Hallgrimsson B. Quantitative 3D analysis ofthe canal network in cortical bone by microcomputed tomography. Anat Rec2003;274:169–79.

[25] Basillais A, Bensamoun S, Chappard C, et al. Three-dimensional characterization ofcortical bone microstructure by microcomputed tomography: validation with ultra-sonic and microscopic measurements. J Orthop Sci 2007;12:141–8.

[26] Bousson V, Peyrin F, Bergot C, et al. Cortical bone in the human femoral neck: three-dimensional appearance and porosity using synchrotron radiation. J Bone Miner Res2004;19:794–801.

[27] Chappard C, Peyrin F, Rachidi M, et al. Assessment of cortical bonemicroarchitecturefrom synchrotron radiation and desktop X-ray micro tomography: a comparativestudy on human femur. 18th International Bone Densitometry Workshop. J ClinDensitom 2009;12(1):110.

[28] Chappard C, Bensalah S, Olivier C, et al. 3D characterization of pores in the corticalbone of human femur in the elderly at different locations as determined by synchro-tron micro-computed tomography images. Osteoporos Int 2013;24:1023–33.

[29] Engelke K, Stampa B, Timm W, et al. Short-term in vivo precision of BMD and pa-rameters of trabecular architecture at the distal forearm and tibia. Osteoporos Int2012;23:2151–8.

[30] Boutroy S, Bouxsein ML, Munoz F, et al. In vivo assessment of trabecular bonemicroarchitecture by high-resolution peripheral quantitative computed tomogra-phy. J Clin Endocrinol Metab 2005;90:6508–15.

[31] Nuzzo S, Lafage-Proust MH, Martin-Badosa E, et al. Synchrotron radiationmicrotomography allows the analysis of three-dimensional microarchitecture anddegree ofmineralization of human iliac crest biopsy specimens: effects of etidronatetreatment. J Bone Miner Res 2002;17:1372–82.

[32] Hildebrand T. A new method for the model independent assessment of thickness inthree dimensional images. J Microsc 1997;18(15):67–75.

[33] Lorensen WE, Cline HE. Marching cubes: a high resolution 3D surface constructionalgorithm. Comput Graph 1987;1(21 Suppl):7–12.

[34] Ulrich D, Van Rietbergen B, Laib A, Ruegsegger P. The ability of three dimensional struc-tural indices to reflect mechanical aspects of trabecular bone. Bone 1999;25:55–60.

[35] Glüer CC, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate assessment of pre-cision errors: how to measure the reproducibility of bone densitometry techniques.Osteoporos Int 1995;5:262–70.

[36] Jepsen KJ, Centi A, Duarte GF, et al. Biological constraints that limit compensation ofa common skeletal trait variant lead to inequivalence of tibial function amonghealthy young adults. J Bone Miner Res 2011;26:2872–85.