Quantitative 3D analysis of the canal network in cortical bone by micro-computed tomography

FEATURE ARTICLE

Quantitative 3D Analysis of the Canal Network inCortical Bone by Micro-Computed TomographyD.M.L. COOPER,* A.L. TURINSKY, C.W. SENSEN, AND B. HALLGRIMSSON

Cortical bone is perforated by an interconnected network of porous canals that facilitate the distribution ofneurovascular structures throughout the cortex. This network is an integral component of cortical microstructureand, therefore, undergoes continual change throughout life as the cortex is remodeled. To date, the investigation ofcortical microstructure, including the canal network, has largely been limited to the two-dimensional (2D) realm dueto methodological hurdles. Thanks to continuing improvements in scan resolution, micro-computed tomography(�CT) is the first nondestructive imaging technology capable of resolving cortical canals. Like its application totrabecular bone, �CT provides an efficient means of quantifying aspects of 3D architecture of the canal network. Ouraim here is to introduce the use of �CT for this application by providing examples, discussing some of the parametersthat can be acquired, and relating these to research applications. Although several parameters developed for theanalysis of trabecular microstructure are suitable for the analysis of cortical porosity, the algorithm used to estimateconnectivity is not. We adapt existing algorithms based on skeletonization for this task. We believe that 3D analysisof the dimensions and architecture of the canal network will provide novel information relevant to many aspects ofbone biology. For example, parameters related to the size, spacing, and volume of the canals may be particularlyuseful for investigation of the mechanical properties of bone. Alternatively, parameters describing the 3D architectureof the canal network, such as connectivity between the canals, may provide a means of evaluating cumulativeremodeling related change. Anat Rec (Part B: New Anat) 274B:169–179, 2003. © 2003 Wiley-Liss, Inc.

KEY WORDS: micro-computed tomography; �CT; Haversian canals; cortical porosity; bone architecture; virtual realityenvironment

INTRODUCTIONHuman cortical bone has a complexand dynamic microstructure that un-dergoes continual change throughoutlife. An integral component of this mi-crostructure is the network of porouscanals that facilitate the distribution

of neurovascular structures through-out the cortex. With the exception of asmall minority of studies, the analysisof cortical bone microstructure, in-cluding the canal network, has largelybeen restricted to the two-dimen-sional (2D) realm. Although 2D histo-

morphometric techniques have andcontinue to provide a wealth of infor-mation regarding bone tissue dynam-ics (Frost, 1969; Parfitt, 1983; Ott,2002), they cannot yield a completepicture of the three-dimensional (3D)microstructure of cortical bone. Cor-tical microstructure exists and remod-els in 3D; therefore, a full understand-ing of its architecture will require 3Danalysis (Stout et al., 1999).

To date, the key factors that havelimited 3D investigation are method-ological. For example, reconstructionfrom serial sections is tedious (De-Hoff, 1983), whereas ink perfusionand casting techniques provide largelyqualitative data regarding 3D struc-ture. Automated, computer-assistedreconstruction of serial sections doesoffer a more efficient alternative(Odgaard et al., 1990), however, suchtechniques have largely been re-placed, for trabecular bone at least, bymicro-computed tomography (�CT).

Mr. Cooper is a Ph.D. candidate in the In-terdisciplinary Ph.D. program in the De-partments of Archaeology and MedicalScience at the University of Calgary (U ofC). His research interests include biologi-cal anthropology and bone biology. His on-going research involves the use of imagingtechnology to study bone morphology. Dr.Turinsky is currently a Postdoctoral Re-search Associate with the Sun Center ofExcellence for Visual Genomics at U of C.He holds a Ph.D. in Mathematical Com-puter Science from the University of Illi-nois at Chicago. His research areas in-clude applied mathematics, data mining,and bioinformatics. Dr. Sensen is a ten-ured Full Professor for Bioinformatics atthe U of C’s Faculty of Medicine. He is thedirector of the Sun Center of Excellencefor Visual Genomics, which features theworld’s first Java 3D-enabled CAVE im-

mersive virtual environment. Dr. Hall-grımsson is an Associate Professor in theDepartment of Cell Biology and Anthropol-ogy and Assistant Dean (USE) in the Fac-ulty of Medicine at the U of C. He is abiological anthropologist and evolutionarybiologist who combines developmentalgenetics with morphometrics to addressthe developmental basis as well as evolu-tionary and biomedical significance ofcomponents of phenotypic variability.*Correspondence to: Benedikt Hallgrıms-son, Department of Cell Biology and Anat-omy, Health Sciences Center, University ofCalgary, 3330 Hospital Drive NW, Calgary,Alberta T2N 4N1, Canada. Fax: 403-210-9747; E-mail: [email protected]

DOI 10.1002/ar.b.10024Published online in Wiley InterScience(www.interscience.wiley.com).

THE ANATOMICAL RECORD (PART B: NEW ANAT.) 274B:169–179, 2003

© 2003 Wiley-Liss, Inc.

�CT offers several advantages, includ-ing the preservation of the specimenfor correlative analysis and a digitaldata format, that facilitates quantita-tive 3D architectural analysis.

With continuing improvements in�CT scan resolution, increasinglysmaller structures have been broughtwithin the reach of this technology,and the ability of �CT to resolve thecanals within human cortical bone(Wachter et al., 2001a) and recon-struct them in 3D (Cooper and Hall-grımsson, 2002) has recently beendemonstrated. Therefore, �CT repre-sents a new, relatively unexplored,tool for 3D analysis of the cortical ca-nal network that we believe has thepotential to provide novel insightsinto the overall microstructure of cor-tical bone. Our aim in this article is tobriefly review what is known about

the 3D microstructure of corticalbone, introduce �CT as an efficienttechnique for quantitative 3D analysisof the cortical canal network, and todiscuss the potential of such analysisin future research.

BRIEF HISTORY

One of the earliest descriptions of thenetwork of canals within cortical bonewas made in 1691 by Clopton Havers,who noted the existence of longitudi-nal and transverse pores within com-pact bone and speculated that thesecanals form an interconnected net-work, providing a route for the con-duction of medullary oils throughoutthe cortex (Havers, 1691). Longitudi-nal canals associated with secondarybone still bear his name (Haversiancanals). However, in an earlier study,

Antonie van Leeuwenhoek (1677–1678)described the composition of compactbone as a collection of “pipes” andincluded what are likely the first 3Drenderings of cortical bone micro-structure (Figure 1). Like Havers’ laterdescription, van Leeuwenhoek alsonoted that the longitudinally andtransversely arranged pores form aninterconnected network. Notably, theidentification of the canals within cor-tical bone preceded that of other mi-crostructural features. It was not untilthe mid-19th century that Todd andBowmann (1845) first described Hav-ersian systems (1845) and Tomes andDe Morgan (1853) discerned the na-ture of their formation within local-ized Haversian (resorption) spaces.

More than 300 years after their firstdescription, the canals within corticalbone are viewed in very much thesame way. The longitudinally orientedcanals are now divided into two types,non-Haversian and Haversian, corre-sponding to primary osteons and sec-ondary osteons, respectively, andtransversely oriented canals areknown as Volkmann’s canals in honorof a 19th century German physiolo-gist. However, evidence from a smallnumber of studies has revealed agreater degree of complexity than wassuggested by conventional descrip-tions. These investigations have pri-marily used two approaches, direct vi-sualization of canals by staining orcasting and reconstruction from serialhistological sections, to evaluate the3D microstructure of cortical bone.Staining techniques range from su-perficial observation of abraded bonesurfaces to determine the general ori-

entation of vascular canals (Hert etal., 1994; Petrtyl et al., 1996) to moreinvolved preparations, which includestaining followed by clearing, to per-mit the visualization of deeper canals(Ruth, 1947). These techniques havebeen particularly effective at demon-strating general patterns of vascularsupply within bones (Dempster andEnlow, 1959; Albu and Georgia,1984), the distribution of resorptivespaces associated with remodeling(Vasciaveo and Bartoli, 1961), and dif-ferences in the canal network in thebones of different species (Georgia etal., 1982; Georgia and Albu, 1988). Al-though effective at visualization of thecanal network, casting and stainingtechniques do not provide a directmeans of quantifying 3D structure;therefore, architectural observationsare largely qualitative.

Serial sectioning, while tedious,provides a means of visualizing thevascular canal network in relation tononporous structures. Cohen andHarris (1958), for example, examinedthe paths of Haversian systems in 3D,and Tappen (1977) examined the 3Dstructure of resorption spaces associ-ated with the formation of Haversiansystems. Before the availability ofcomputers, reconstruction from serialsections required the resourceful useof materials such as cardboard andwires to create 3D models (Amprino,1948; Cohen and Harris, 1958; Krag-strup and Melsen, 1983). The use ofcomputers in conjunction with histo-logical serial sections has allowedsuch innovations as real-time manip-ulation of semitransparent models,permitting the visualization of thespatial relationship between cells andextracellular matrix in 3D (Schnapperet al., 2002).

Direct analysis of serial sections, or3D reconstructions derived fromthem, has provided a means of testinghypotheses that have been inferredfrom analysis of sections. For exam-ple, 3D reconstruction of osteons inthe canine femur by Stout et al. (1999)demonstrated a “complex pattern,dominated by branching” and lead tothe conclusion that several morpho-logic types of osteons identified in 2Dsections may not actually exist “butrather are artifacts of the relative ori-entation of the plane of sectioning, orwhere along the length of the haver-

Figure 1. van Leeuwenhoek’s diagrams ofcortical bone structure (van Leeuwenhoek,1677–1678).

One of the mostimportant advantages of

�CT-based analysis oftrabecular bone is that it

provides an efficientmeans by which 3Darchitecture can be

quantified.

170 THE ANATOMICAL RECORD (PART B: NEW ANAT.) FEATURE ARTICLE

sian canal the section was made”. Aspecific example that was provided isthat dumbbell-shaped osteons are ac-tually artifacts of sectioning through abifurcating osteon. Such findings un-derline the need for further 3D analy-sis of cortical bone microstructure.Investigation of the vascular canalnetwork by �CT may provide an im-portant stepping stone.

APPLICATION OF �CT IMAGINGTO CORTICAL BONE

Computed microtomography imaginghas been growing in use since its de-scription as a tool for the direct anal-ysis of 3D trabecular bone structure in1989 (Feldkamp et al., 1989). Themost common application of thistechnology has been the in vitro quan-tification of osteoporotic change intrabecular bone architecture (Borahet al., 2001). After its introduction,technical developments have resultedin the progressive improvement ofscan resolution from 70 microns toresolutions on the order of a singlemicron (Jiang et al., 2002). The high-est resolutions are generally achievedwith synchrotron radiation (SR)sources and provide an extraordinarylevel of detail. For example, osteocytelacunae are visible in tomographic im-ages with 1.4 �m spatial resolution(Peyrin et al., 1998). The developmentof higher spatial resolution is ongo-ing, and sub-micron tomography (0.6�m) has recently been achievedthrough the combination of �CT andx-ray microscopy technology (Takeu-chi et al., 2002). For commerciallyavailable �CT scanners, resolutions of30 to 15 �m are typical (Borah et al.,2001), reflecting the range of resolu-

tions commonly used for trabecularbone imaging. However, higher reso-lutions are increasingly available andwe use a scanner that is capable ofspatial resolutions ranging from 20 to5 �m.

Clearly, with the ability to resolveosteocyte lacunae, �CT technologyhas developed beyond the resolutionnecessary for the analysis of corticalcanals. This finding leads to the ques-tion of what resolution is needed fortheir analysis, and the answer re-quires some knowledge of the scale ofthe actual canals. Based on data from2D histological studies, it is knownthat the size of pores (canals viewed incross-section) within cortical bonevaries both within and among species(Jowsey, 1966). Additionally, due tothe formation by of Haversian sys-tems by the infilling of resorptionspaces, Haversian canal diameterscan vary greatly within a single bonedue to the simultaneous presence ofsystems at various states in the re-modeling process. The size of resorp-tion spaces in human bone have beenreported to be on the order of 200 to300 microns (Dempster and Enlow,1959; Johnson, 1964; Jaworski et al.,1972); therefore, Haversian canal di-ameters can be expected to range be-tween these values and their “mature”size. Human cortical pore sizes varyconsiderably across age, sex, and sam-pling site (Stein et al., 1999; Boussonet al., 2001), making this “mature”size difficult to define. Pfeiffer (1998)reported that human Haversian ca-nals at the femoral midshaft havelarger cross-sectional areas than thosein ribs, with (assuming circular canalshape) mean diameters of 33 and 27�m, respectively. However, consider-

ably smaller Haversian canals werealso found with the 5th percentiles forfemora and ribs being 17 �m and 12�m, respectively (Pfeiffer, 1998). Sim-ilarly, in the human mandible, meanHaversian canal diameter has been re-ported to be 45 �m; however, 7% ofcanals measured were 15 �m or less indiameter (Dempster and Enlow, 1959).

The various factors that influencecanal size must be kept in mind whenchoosing scan resolution. If only per-cent porosity and bone volume frac-tion are required, lower resolutionsmay provide adequate results. For ex-ample, Wachter et al. (2001a) re-ported that a 30 �m scan resolutionwas effective at predicting cortical po-rosity (%) when compared with con-ventional histomorphometry in asample of older adults. However, ashuman bones contain many canalswith diameters well below 30 �m,analyses of 3D architectural charac-teristics require considerably higherresolutions.

QUANTITATIVE ASSESSMENT OFTHE CORTICAL CANALNETWORK

One of the most important advantagesof �CT-based analysis of trabecularbone is that it provides an efficientmeans by which 3D architecture canbe quantified. A host of quantitativeparameters have been developed tomeasure different aspects of trabecu-lar bone architecture, and the readeris referred to a recent review by Borahet al. (2001) for an excellent summary.In general terms, these parameterscan be divided into two groups, (1)those that describe the amount of tra-becular bone and average characteris-tics of individual trabeculae, and (2)those that describe the more complexarchitecture of the trabeculae collec-tively. Provided that they are mea-sured directly and not based on astructural model, many parametersfrom the first group are transferableto the analysis of the cortical canals(see Table 1). For example, corticalporosity (Ca.V/TV) reflects the relativevolume of porous canals (Ca.V) withinthe sample tissue volume (TV). Thisvalue is the compliment of bone vol-ume fraction (BV/TV) and is analo-gous to marrow volume fraction intrabecular bone (Ma.V/TV). Likewise,

TABLE 1. Analogous morphological parameters fortrabecular and cortical bone*

Trabecular bone Cortical bone

Tissue Volume (TV) Tissue Volume (TV)Bone Volume (BV) Canal Volume (Ca.V)Bone Surface (BS) Canal Surface (Ca.S)Bone Volume Fraction (BV/TV) Cortical Porosity (Ca.V/TV)Bone Surface to Tissue Volume (BS/TV) Canal Surface to Tissue Volume

(Ca.S/TV)Trabecular Thickness (Tb.Th) Canal Diameter (Ca.Dm)Trabecular Separation (Tb.Sp) Canal Separation (Ca.Sp)

*All abbreviations are based upon standard nomenclature (Parfitt et al., 1987).

FEATURE ARTICLE THE ANATOMICAL RECORD (PART B: NEW ANAT.) 171

trabecular thickness (Tb.Th) and sep-aration (Tb.Sp), can be translated intomeasures of cortical canal diameter(Ca.Dm) and canal separation (Ca.Sp).A particularly attractive aspect ofthese parameters is that many can bemeasured using existing software de-veloped for trabecular bone analysis.

The application of parameters usedto describe the architecture of trabec-ular bone to cortical canal analysis isless straightforward. For example,structural model index (SMI), a mea-sure that describes the relative contri-butions of plate-like and rod-likestructures (Hildebrand and Ruegseg-ger, 1997b), is likely to be of limitedutility. The canals within corticalbone are approximately cylindrical,and in terms of SMI, very rod-like. Onthe other hand, measurement of theconnectivity within the network of ca-nals is of interest, particularly withrespect to the investigation of remod-eling-related change. Connectivity, asdefined for trabecular bone analysis,essentially reflects the number ofclosed loops (holes) in the structure(Odgaard and Gundersen, 1993). Be-cause trabecular bone may be viewedas a tightly packed mesh of loopsformed by holes in the bony tissue,this topological approach is appropri-ate. On the other hand, the canalswithin cortical bone vary in 3D struc-ture from rectangular meshes tohighly branched, complex networksand, therefore, are not well character-ized by their topology alone. Indeed, atree-like network of cortical poreswith a large number of branches but

no closed loops would have the sameconnectivity measure as a single pore!Therefore, we suggest defining corti-cal canal connectivity (Ca.ConnD) ingeometric rather than topologicalterms, as the number of canal inter-sections per unit volume.

To measure canal connectivity andother architectural characteristics, wehave developed custom image analy-sis software. There exist several stan-dard pattern recognition approachesfor image analysis that relate to suchtasks. Many of them, however, are in-herently stochastic, and their accu-racy is at best approximate (Jain et al.,2000). We have chosen a deterministicskeletonization technique that thinsstructures until only their framework,called a skeleton, remains (Lam et al.,1992). Skeletonization has several im-portant characteristics, including thepreservation of the topological prop-erties and a deterministic nature thatis well suited for proof-of-correctnessanalysis. Both of these factors are es-sential for meaningful analysis ofparameters, such as connectivity.Skeletonization has been used success-fully for several practical applicationsincluding the analysis of biomedical(Selle et al., 2000), and geological(Lindquist and Venkatarangan, 1999)structures. This approach works par-ticularly well with tube-like imagesegments whose central axes are eas-ily identifiable and, therefore, is wellsuited to the 3D structure of the cor-tical canals.

In terms of their implementation,popular 3D thinning algorithms ex-

tract the skeleton of an image by peel-ing subsequent layers of voxels (avoxel, or volume element, is a 3D an-alogue of a pixel, or picture element).The process stops when no voxel canbe removed without altering the topo-logical properties of the remaining 3Dstructure. The Euler characteristic,which is commonly used to quantifytopology, is used to regulate the voxelpeeling process. For a 3D object, it isdefined as E � number of Compo-nents � number of Holes � number ofCavities., where Components refers tothe number of distinct disconnectedelements, Holes refers to holes thatpierce the object, and Cavities are en-closed spaces that are not connectedto the outside of the object. The thin-ning procedure is designed so that thevalue E remains constant at all times.Several additional rules are developedto protect important geometric fea-tures of the structural boundaries,such as the end points of thebranches. A peeling skeletonizationalgorithm must be able to identifyboundary points that can be removedwithout altering the topology of theobject or violating the geometricalconstraints. In our application, we im-plement a thinning algorithm devel-oped by Lee et al. (1994), which isbased on a parallel voxel peelingschema that is fast, robust, and con-tains several safeguards to preservethe geometric properties of the image.After this process, architectural pa-rameters such as canal connectivitydensity (Ca.ConnD) and total canallength (Ca.Le) can be measured fromthe skeletons.

DEMONSTRATION OF THETECHNIQUE

Box 1 details the methods and mate-rials used for this 3D �CT analysis oftwo human femora to quantify spe-cific aspects of the architecture of thecanal network. Examples of 2D cross-sectional images, 3D surface render-ings, and 3D representations of theskeletonized networks are provided inFigures 2, 3, and 4, respectively. The2D cross-sectional images of the sam-ples demonstrated marked differencesin the cross-sectional size of the ca-nals (pores) and overall porosity. 3Drenderings demonstrated qualitativedifferences in the canal networks of

Figure 2. Two-dimensional cross-sectional micro-computed tomography images from sam-ples A (left) and B (right) acquired at 10 �m spatial resolution. The region of interestrepresents the region of interest (2.5 mm � 2.5 mm) targeted for skeletonization andquantitative analysis. Scale bar � 1 mm.

172 THE ANATOMICAL RECORD (PART B: NEW ANAT.) FEATURE ARTICLE

the two samples and reflected thesame marked difference in overallcortical porosity and canal size thatwere evident in the 2D cross-sections.After skeletonization, it was clear thatsample A had a simpler canal networkwith many longitudinal canals andrelatively fewer canal intersections.On the other hand, sample B had amore complex appearance with manyperpendicular and oblique intersect-ing branches. Quantitative analysis,presented in Table 2 and Figures 5and 6, revealed that sample B yieldedlarger values than sample A for nearlyall parameters, despite that the twosamples had similar numbers of ca-nals (Ca.N). The exception to this pat-tern was canal separation (Ca.Sp),which was lower in sample B.

DISCUSSION OF LIMITATIONS

The �CT technique we used (Box 1)offers an efficient means of quantify-ing 3D parameters related to the canalnetwork within human bone. How-ever, as with its application to trabec-ular bone architecture, �CT is limitedto aspects of porosity and volumefraction. �CT does not provide infor-mation related to many facets of bonedynamics that can be assessed by his-tological analysis. For example, it can-not differentiate between resorbing,forming, or resting surfaces nor can itprovide information regarding theorientation of collagen fibers (Uchi-yama et al., 1997; Muller et al., 1998).With regard to cortical bone, our 2D�CT images acquired at 10 �m reso-

lution showed no evidence of nonpo-rous microstructural details, such aslamellae and cementing lines. As aconsequence, no differentiation be-tween pore types (Haversian or non-Haversian) was possible.

Similar findings were reported byEngelke et al. (1993), who comparedmicroradiographs and �CT imagedata and found that, at 50 �m resolu-tion, osteons were not discernible inthe �CT images. An additional limita-tion of the 10 �m resolution was evi-dent from the 3D reconstruction ofsample A. The “break-up” of the integ-rity of the smaller canals indicatesthat the resolution was insufficient toconsistently resolve them. This cer-tainly must have negatively affectedthe accuracy of the quantitative pa-

Box 1. �CT Analysis: Materials and Methods

To illustrate the use of �CT foranalysis of the cortical canal network,we applied the technique to anteriormidshaft samples from two humanfemora (5 mm � 10 mm blocks). Thepreviously macerated bones were ac-quired from teaching materials andwere of unknown age and sex. De-spite that chronological age was un-known, histological examination re-vealed marked differences in tissueage. The cortex of the first sample (A)contained many non-Haversian ca-nals, and remodeling-related changewas not extensive throughout thecortex, therefore indicating a rela-tively young tissue age (Kerley, 1965).Alternatively, the second sample (B)was extensively remodeled, showedlittle sign of primary bone (osteons orlamellae), and had a highly porousappearance, all signs that indicatemore advanced tissue age. The sam-ples were scanned at the Universityof Calgary 3D Morphometrics Labo-ratory, using a SkyScan 1072 (Aartse-laar, Belgium) scanner. A spatial res-olution of 10 �m was used, based onpast experience (Cooper and Hall-grımsson, 2002). The scan protocolincluded rotation through 180 de-grees at a rotation step of 0.45 de-grees, x-ray settings standardized to100 kV and 100 �A, and an exposuretime of 5.9 s per frame. Four-frameaveraging was used to improve the

signal to noise ratio. Scan times wereapproximately 4 hr per sample.

After scanning, a 2D reconstructionstage using a cone–beam algorithmwas used to produce serial cross-sectional images. These consisted ofmatrices of 1,024 � 1,024 pixels,which collectively composed a vol-ume of isotropic 10 �m3 voxels. A 3Dmedian filter (5 � 5 � 5 kernel) wasused to further improve the signal tonoise ratio in the images. Globalthresholding was used to segmentthe eight-bit grayscale images into bi-nary black and white images to facil-itate quantitative analysis and 3D vi-sualization of the cortical canals. Afterthis, a 2.5 mm � 2.5 mm region ofinterest (ROI) from the midcortical re-gions of each of the samples wascropped from 500 serial images rep-resenting 5 mm along the length ofthe sample. This strategy producedmatching 31.25 mm3 volumes of in-terest from which all measurementswere acquired. Surface renderingsand measurement of the parametersanalogous to trabecular bone analy-sis (see Table 1) were performed us-ing SkyScan software (3D Creator,version 2.2d). Canal diameter (Ca.Dm)and spacing (Ca.Sp) were measureddirectly, using a model-independent3D method (Hildebrand and Rueg-segger, 1997a). Canal number (Ca.N)

was measured two-dimensionally asthe average number of distinct canalsper mm2 in the serial image stacksusing the particle analysis function ofImageJ 1.27z (http://rsb.info.nih.gov/ij/). 3D visualization of the skeleton-ized canal networks and quantitativeanalysis of architectural parameterswere performed using the high-per-formance computing environment atthe University of Calgary’s Sun Cen-ter of Excellence for Visual Genomics(http://www.visualgenomics.ca/).

The 3D models of the cortical ca-nals were converted into Java 3D ob-jects. The Java 3D models could thenbe displayed and manipulated in theCAVE immersive virtual reality envi-ronment (Sensen, 2002) with 12 de-grees of freedom (6 for the head-tracking system and 6 for thejoystick-like wand). The large displaysize and the high resolution of theCAVE, in combination with the uniquehuman-machine interface, allowedrapid study of the complex 3D canalmesh. Using the Java 3D ConfiguredUniverse tool, the same Java codecan be used on various display envi-ronments, including regular computerscreens. A demonstration of the canalmodel, configured for monographictwo-dimensional computer displays,and a screenshot are available athttp://www.visualgenomics.ca/cave.

FEATURE ARTICLE THE ANATOMICAL RECORD (PART B: NEW ANAT.) 173

rameters, particularly those related tocanal length (Ca.Le) and diameter(Ca.Dm). On the other hand, overallporosity was likely impacted less, be-cause only the smallest branches werelost. The effect on the measurement ofcanal intersections is unclear but alsolikely to be minor, as the canalstended to be enlarged around suchpoints and, therefore, well preservedin the 3D renderings and skeletons.

In light of these limitations, we havebegun exploring the use of 5 �m spa-tial resolution (Figure 7). At thishigher resolution, we have found thatcanal continuity in 3D renderings isdramatically improved. Additionally,some differences in mineral densityassociated with different osteons arediscernible, but these differences tendto be diffuse and inconsistent alongserial images. �CT images acquired at1.4 �m using SR do show localizeddifferences in mineral density, al-though distinct lamellae do not ap-pear to be visible (Peyrin et al., 1998).Therefore, further improvements inscan resolution may allow the identi-fication of additional structural detail.However, it should be noted that, as

spatial resolution increases, the fieldof view decreases (Peyrin et al., 2000);thus, it is not necessarily beneficial topursue the highest resolution possi-ble. Ultimately, a compromise be-tween scan resolution, the size of theregion/volume of interest, and the de-sired parameters is necessary. Furtherinvestigation is required to determinethe impact of resolution on quantita-tive parameters, a relationship knownas resolution dependency (Muller etal., 1996).

The availability of the CAVE im-mersive virtual reality environment atthe Sun Center of Excellence for Vi-sual Genomics has greatly facilitatedour work on connectivity measure-ments. In particular, fine-tuning the3D skeletonization algorithm and test-ing its correctness may have beenrather problematic without the abilityto interactively visualize the resulting3D models (Figure 8). Although gen-eral characteristics of the corticalpore network may be assessed by ex-amining its 2D projections, (such asthose in Figures 3 and 4), the entan-gled nature of the canal networkmakes it difficult to discern finer de-tails. The CAVE environment proved

Figure 3. Three-dimensional renderings of the subregions from samples A (left) and B (right).The volumes were acquired from the region of interest (2.5 mm � 2.5 mm) outlined in Figure2 and carried through 500 serial images representing 5 mm along the length of the sample.

Figure 4. Three-dimensional skeletons of the subregions from samples A (left) and B (right),showing the canal skeletons (blue) and canal intersections (red dots). The volumes wereacquired from the region of interest (2.5 mm � 2.5 mm) outlined in Figure 2 and carriedthrough 500 serial images representing 5 mm along the length of the sample.

174 THE ANATOMICAL RECORD (PART B: NEW ANAT.) FEATURE ARTICLE

to be an indispensable research tool,allowing the visualization of the 3Dstructure of the canal network, or itsskeleton, in its entirety at any timeduring the development and testing,thereby ensuring the correctness andthe quality of the results.

POTENTIAL APPLICATIONS OF3D ANALYSIS

Cortical bone microstructure existsand remodels in 3D, and section-based analysis is incapable of provid-ing a complete picture of its 3D struc-ture. The complex skeletonizedmeshes produced by our analysis il-lustrate this point. However, an im-portant question to address is how,specifically, can 3D analysis of thecortical canal network contributenovel information relevant to bone bi-ology research? Whereas we believethat a variety of applications will befruitful, we focus our following dis-cussion on two general areas of re-search, cortical bone mechanics andcortical remodeling, to illustrate somekey potentials.

The mechanical properties of boneare the product of many factors, in-

cluding porosity, tissue density, min-eral content, and the orientation ofthe collagen fibers within the extracel-lular matrix. Numerous studies havenoted the importance of porosity withrespect to cortical bone strength andelasticity (Currey, 1988; Schaffler andBurr, 1988; Martin and Ishida, 1989).McCalden et al. (1993) reported thatage-related increase in cortical poros-ity is a major contributor to the con-current decrease in mechanical prop-erties, accounting for 76% of the lossof bone strength with age. In additionto overall measures of porosity, thespatial arrangement of microstruc-tural features and size of pores havebeen implicated in mechanical prop-erties such as fracture toughness(Yeni et al., 1997). Due to its clinicalsignificance, the femoral neck hasbeen the focus of numerous studies,which have examined the relationshipbetween cortical microstructure andstrength. Cortical porosity is an im-portant factor influencing the fracturetoughness of this structure (Yeni andNorman, 2000). Differences in the dis-tribution of cortical porosity in thefemoral neck, with increased porosityconcentrated in the anterior regionhave been associated with increasedfracture risk (Bell et al., 1999). In ad-dition to the percentage of porosityoverall, the dimensions of the individ-ual canals and their spatial arrange-ment have also been implicated. Largevoids (diameter � 385 �m) termed“giant canals” (Bell et al., 1999) havebeen associated with spatial cluster-ing of remodeling osteons (Jordan et

al., 2000) and have been linked withincreased fracture risk.

As has been demonstrated, �CT iswell suited to measuring cortical po-rosity (Ca.V/TV) as well as the number(Ca.N), size (Ca.Dm), and spacing(Ca.Sp) of cortical canals. In additionto these parameters, quantification ofcanal orientation, by using measure-ments such as degree of anisotropy (ameasure of preference in orientation),may be useful because Haversian sys-tems, and hence their central canals,lie in tracts that are related to loadinghistory (Hert et al., 1994; Peyrin et al.,2000). The ability to acquire suchquantitative parameters in 3D, and todo so nondestructively, presents sev-eral major advantages. First, the anal-ysis of a larger tissue volume may pro-vide a better picture of regionalvariation in the size and spatial ar-rangement of canals. By using themethods outlined above, at 10 �mspatial resolution, a sample up to 1 cmlong can be analyzed from a singlescan. Second, the sample is preservedso further analysis such as mechani-cal testing or histological sectioningcan be conducted on the same tissuevolume (Wachter et al., 2001b).

Figure 5. Distribution of canal diameters(Ca.Dm) from samples A and B.

Figure 6. Distribution of canal separation(Ca.Sp) from samples A and B.

TABLE 2. Quantitative Results for Samples A and B*

Parameter Sample A Sample B Units

Tissue Volume (TV) 31.25 31.25 mm3

Canal Volume (Ca.V) 1.45 6.50 mm3

Canal Surface (Ca.S) 75.89 167.18 mm2

Cortical Porosity (Ca.V/TV) 4.64 20.80 %Canal Surface to Tissue Volume (Ca.S/TV) 2.43 5.35 mm�1

Canal Number (Ca.N) 9.39 9.80 mm�2

Canal Diameter (Ca.Dm) 55 115 �mCanal Separation (Ca.Sp) 344 115 �mCanal Length (Ca.Le) 329.3 441.1 mmCanal Length to Tissue Volume (Ca.Le/TV) 10.5 14.1 mm�2

Canal Intersections 190 462 —Canal Connectivity Density (Ca.ConnD) 6.1 14.8 mm�3

*All abbreviations are based upon standard nomenclature (Parfitt et al., 1987).

FEATURE ARTICLE THE ANATOMICAL RECORD (PART B: NEW ANAT.) 175

Recently, it has been proposed thatcortical remodeling may have a higher3D level of organization, termed “su-per-osteons,” that account for the phe-nomena of remodeling clusters and,possibly, giant canals (Bell et al.,2001). This model hypothesizes thatspatial clustering of canals may be theresult of a single remodeling entitythat branches, producing the appear-ance of clustered osteons in section.Similarly, giant canals (�385 �m)may be the product of these clustersmerging through excessive resorption.This hypothesis effectively links themechanically oriented investigationoutlined above with the physiologicalprocess of bone remodeling.

This link leads to the second areawhere we believe 3D �CT-based anal-ysis will be particularly useful: quan-tification of remodeling associatedbone dynamics. As has been noted,�CT is limited to volume fraction andporosity; thus, it is unable to assessmany aspects of cortical microstruc-tural dynamics that have been inves-tigated histologically. However, be-cause the network of cortical canals isan intrinsic component of the cre-ation of Haversian systems, it followsthat the structure of the canal networkundergoes continual change through-

out life. This raises the possibility that3D morphologic analysis of the canalnetwork may provide novel insightsinto bone physiology that are not pos-sible with 2D approaches. To illus-trate this point, it is important to un-derstand the canal network in thecontext of the processes responsiblefor bone formation and turnover. Be-fore the mid-1960s, the cell popula-tions that carry out bone formation(osteoblasts) and resorption (oste-oclasts) were believed to operate inde-pendently of one another. Over thepast four decades, an alternative pic-ture of highly coordinated cellular ac-tivity has emerged with two generalprocesses recognized: modeling andremodeling (reviewed in Frost, 2001).Modeling is the process by whichbone surfaces are altered, either byformative or resorptive drifts. It isthus the process by which bone shape,thickness, and position in “tissuespace” are regulated. Formative mod-eling drift produces new bone, re-ferred to as primary bone, which in-corporates primary vascular spaces

Figure 7. Micro-computed tomography cross-section and three-dimensional rendering ofcanal network of a sample of human femoral bone (third specimen, unrelated to quanti-tative analyses) using 5 �m spatial resolution. Scale bar � 1 mm.

Figure 8. Simultaneous three-dimensional visualization of the canal network (transparent),its skeleton (blue), and canal intersections (red dots) in the CAVE immersive virtual realityenvironment at the Sun Center of Excellence for Visual Genomics.

176 THE ANATOMICAL RECORD (PART B: NEW ANAT.) FEATURE ARTICLE

that subsequently form primary os-teons with non-Haversian canals. The3D structure of primary osteons and,thus, their canals is determined by thepattern of the vasculature that theyassimilate. Subperiosteal bone forma-tion incorporates the longitudinallyoriented rectangular meshes thatserve as the template for primary os-teon structure and as a framework forsubsequent changes associated withremodeling (Marotti and Zallone,1980). These rectangular meshes cor-respond well with conventional de-scriptions of the canal network involv-ing primarily longitudinally andtransversely oriented canals.

Remodeling creates secondary os-teons, and involves the close coordi-nation of osteoclasts and osteoblastsin what is known as the basic multi-cellular unit (BMU; Frost, 1966). Theporous component of the BMU takesthe form of a “cutting cone,” whereosteoclasts resorb a roughly cylindri-cal packet of bone, followed by a“closing cone,” where osteoblasts de-posit the concentric lamellae withinthe resorption space (Johnson, 1964)(Figure 9). The large resorptionspaces created by BMU cutting conesare readily identifiable using resolu-tions of 10 �m and lower (unpub-lished results) and it is potentiallypossible to count the number of activeBMUs per unit volume in 3D recon-struction. The 3D structure of BMUsis also of interest, as they do not al-ways have a simple configuration. 2Dsection-based studies have found thatBMUs can be double ended (Tappen,1977), branched, or both (Johnson,1964). It is the more complex form ofBMU that is likely responsible for the

production of “super-osteons” that re-sult in clustering of remodeling os-teons (Bell et al., 2001). New BMUsare initiated at bone surfaces includ-ing the periosteal, endosteal, and en-docortical walls of existing canals.Tappen (1977), by using serial sec-tions to examine the courses of re-sorption spaces, found that thesestructures are the continuations of de-veloping osteons whose canals are, inturn, always connected to previouslyexisting vascular canals by trans-versely oriented canals (Figure 10).The transverse canals were identifiedas the points of origin for new second-ary osteons that formed through thecreation of “breakout zones” in thelateral walls of existing channels inthe bone (Tappen, 1977). Therefore,in a general sense, remodeling pro-duced by BMU activity continuallymodifies the canal network throughthe creation of new canals that branchoff of existing canals.

We propose that quantification ofcanal intersections in the canal net-work (Ca.ConnD) may, in essence,provide a measure of cumulativeBMU activity. The increase in thenumber of canal intersections in sam-ple B relative to sample A is consistentwith this hypothesis. However, it wassomewhat curious that sample B didnot have considerably more canals(Ca.N) when viewed in cross-section.This finding was likely the product ofmeasuring Ca.N in a 2D manner,where pores that are interconnectedin the sampling plane are collectivelycounted as a single pore. Due to this,sample B, which had far more trans-versely oriented interconnectingbranches, much larger canal diame-ters, and much smaller canal spacing,was biased toward a lower canalcount. This phenomenon may help toaccount for the results of previous 2Dstudies of femoral cortical porosity,which have found that pore numbersincrease early in life, level off in the5th to 6th decade, and then decreasewith advancing age (Stein et al., 1999;Bousson et al., 2001). This findinghighlights a limitation of the 2D ap-proach and suggests that 3D analysismay be particularly useful in examin-ing older tissue ages. In cross-section,the accumulation of new Haversiansystems reaches an asymptote as theaddition of new systems eventually

obscures the evidence of former sys-tems (Frost, 1987).

A similar asymptote likely existswith respect to 3D change in the canalnetwork. However, as suggested byour results with samples A and B, thislimit may be later in 3D than 2D. Ad-mittedly, with such a small samplesize, caution should be exercised be-fore drawing inferences about generalcharacteristics of remodeling. BMUscan have complex, branched mor-phologies; thus, it should not be ex-pected that a simple one-to-one rela-tionship exists between BMU activationand canal intersections. However,qualitatively and quantitatively, thereappears to be a shift in the canal net-work from a relatively simple rectan-gular mesh in younger bone, to amore complex, highly branched struc-ture with advancing age. While wehave focused upon canal interconnec-tions, other parameters, existing andnew, will undoubtedly prove useful inanalyzing cortical canals in 3D. Forexample, the addition of transverselyoriented branches to a predominantlylongitudinally oriented network shouldact to decrease the degree of anisotropy.

CONCLUSION

It has been noted for some time that�CT has the potential for the analysisof cortical bone structures (Feldkampet al., 1989). Until now, however, thispotential has remained relatively un-explored. The inability of �CT to re-solve fine details, such as lamellae andcementing lines in cortical bone,means for the time being that �CT islimited to the analysis of the corticalcanal network. This limitation, com-bined with the notion that corticalbone has a regular structure that is

Figure 9. Schematic view of Haversian re-modeling: the structure of the basic multi-cellular unit in longitudinal and cross-sec-tion. A: Completed Haversian system. B:Location of the closing cone. C: Haversian(resorption) space adjacent to the advanc-ing cutting cone.

Figure 10. Canal-based view of the forma-tion of a new Haversian system/canal. A:Cutting cone. B: Closing cone. C: Locationof break-out zone.

FEATURE ARTICLE THE ANATOMICAL RECORD (PART B: NEW ANAT.) 177

well defined by histological analysis(McCreadie et al., 2001), has resultedin the limited exploration of �CT ap-plications for cortical microstructuralresearch.

As demonstrated by our results, thecortical canal network is a complex3D structure that cannot be fully ap-preciated with 2D sections. Ulti-mately, nondestructive 3D histomor-phometric analysis, comparable toexisting histological techniques, maybe possible through further techno-logical developments. For now, 3Danalysis of the cortical canal networkserves as an important stepping stonethat we believe will provide numerousinsights into the overall 3D micro-structure of cortical bone. Achieving abetter understanding of the 3D archi-tecture of cortical canal network willenhance our current understanding ofthe biomechanical properties of boneand how bone remodels in 3D. This, inturn, may improve our understandingof pathological processes that affectbone. The use of �CT for trabecularbone analysis has proven to be a fast,precise method for quantitative mor-phologic analysis, while at the sametime, providing a less-destructive al-ternative to conventional histomor-phometry. We believe that the samebenefits can be extended to the analy-sis of cortical bone microstructure.

ACKNOWLEDGMENTSWe thank the three anonymous re-viewers for their constructive com-ments. This research was supportedby Alberta Innovation and Science,the Canadian Foundation for Innova-tion (grant 3923), the National Sci-ence and Engineering Research Coun-cil (grant 238992-01 to B.H.), and theUniversity of Calgary. C.S. is sup-ported by the Alberta Science and Re-search Authority, Western EconomicDiversification, the Alberta Networkfor Proteomics Innovation, GenomePrairie, and Genome Canada. B.H. re-ceived funding from the National Sci-ence and Engineering Research Coun-cil.

LITERATURE CITED

Albu I, Georgia R. 1984. Some aspects ofthe system of Haversian canals in thediaphysary compact bone in man. Mor-phol Embryol (Bucur) 30:17–20.

Amprino R. 1948. A contribution to thefunctional meaning of the substitutionof primary by secondary bone tissue.Acta Anat 5:291–300.

Bell KL, Loveridge N, Power J, GarrahanN, Meggitt BF, Reeve J. 1999. Regionaldifferences in cortical porosity in thefractured femoral neck. Bone 24:57–64.

Bell KL, Loveridge N, Reeve J, Thomas CD,Feik SA, Clement JG. 2001. Super-os-teons (remodeling clusters) in the cortexof the femoral shaft: Influence of age andgender. Anat Rec 264:378–386.

Borah B, Gross GJ, Dufresne TE, SmithTS, Cockman MD, Chmielewski PA,Lundy MW, Hartke JR, Sod EW. 2001.Three-dimensional microimaging (MR-microI and microCT), finite elementmodeling, and rapid prototyping provideunique insights into bone architecture inosteoporosis. Anat Rec 265:101–110.

Bousson V, Meunier A, Bergot C, Vicaut E,Rocha MA, Morais MH, Laval-JeantetAM, Laredo JD. 2001. Distribution of in-tracortical porosity in human midfemo-ral cortex by age and gender. J BoneMiner Res 16:1308–1317.

Cohen J, Harris W. 1958. The three-dimen-sional anatomy of haversian systems.J Bone Joint Surg Am 40:419–434.

Cooper DML, Hallgrımsson B. 2002. Theapplication of 3D �CT to the analysis ofcortical bone porosity. Am J Phys An-thropol 117:56–57.

Currey JD. 1988. The effect of porosity andmineral content on the Young’s modulusof elasticity of compact bone. J Biomech21:131–139.

DeHoff RT. 1983. Quantitative serial sec-tioning analysis: Preview. J Microsc 131:259–263.

Dempster WT, Enlow DH. 1959. Patternsof vascular channels in the cortex of thehuman mandible. Anat Rec 135:189–205.

Engelke K, Graeff W, Meiss L, Hahn M,Delling G. 1993. High spatial resolutionimaging of bone mineral using com-puted microtomography. Comparisonwith microradiography and undecalci-fied histologic sections. Invest Radiol 28:341–349.

Feldkamp LA, Goldstein SA, Parfitt AM,Jesion G, Kleerekoper M. 1989. The di-rect examination of three-dimensionalbone architecture in vitro by computedtomography. J Bone Miner Res 4:3–11.

Frost HM. 1966. Relation between bonetissue and cell population dynamics, his-tology and tetracycline labeling. Clin Or-thop 49:65–75.

Frost HM. 1969. Tetracycline-based histo-logical analysis of bone remodeling. Cal-cif Tissue Res 3:211–237.

Frost HM. 1987. Secondary osteon popula-tion densities: An algorithm for estimat-ing the missing osteons. Yearb Phys An-thropol 30:239–254.

Frost HM. 2001. From Wolff’s law to theUtah paradigm: Insights about bonephysiology and its clinical applications.Anat Rec 262:398–419.

Georgia R, Albu I. 1988. The Haversiancanal network in the femoral compact

bone in some vertebrates. Morphol Em-bryol (Bucur) 34:155–159.

Georgia R, Albu I, Sicoe M, Georoceanu M.1982. Comparative aspects of the densityand diameter of Haversian canals in thediaphyseal compact bone of man anddog. Morphol Embryol (Bucur) 28:11–14.

Havers C. 1691. Osteologia nova, or, somenew observations of the bones and theparts belonging to them, with the man-ner of their accretion, and nutrition,communicated to the Royal Society inseveral discourses (reprinted 1977). AnnArbor: University Microfilms Interna-tional.

Hert J, Fiala P, Petrtyl M. 1994. Osteonorientation of the diaphysis of the longbones in man. Bone 15:269–277.

Hildebrand T, Ruegsegger P. 1997a. A newmethod for the model-independent as-sessment of thickness in three-dimen-sional images. J Microsc 185:67–75.

Hildebrand T, Ruegsegger P. 1997b. Quan-tification of bone microarchitecture withthe structure model index. ComputMethods Biomech Biomed Eng 1:15–23.

Jain A, Duin P, Mao J. 2000. Statisticalpattern recognition: A review. IEEETrans PAMI 22:4–37.

Jaworski ZF, Meunier P, Frost HM. 1972.Observations on two types of resorptioncavities in human lamellar cortical bone.Clin Orthop 83:279–285.

Jiang Y, Zhao J, Genant HK. 2002. Macro-and microimaging of bone architecture.In: Bilezikian JP, Raisz LG, Rodan GA,editors. Principles of bone biology. 2nded. New York: Academic Press. p 1599–1623.

Johnson LC. 1964. Morphologic analysis ofpathology. In: Frost HM, editor. Bonebiodynamics. Boston: Little, Brown, andCompany. p 543–654.

Jordan GR, Loveridge N, Bell KL, Power J,Rushton N, Reeve J. 2000. Spatial clus-tering of remodeling osteons in the fem-oral neck cortex: A cause of weakness inhip fracture? Bone 26:305–313.

Jowsey J. 1966. Studies of Haversian sys-tems in man and some animals. J Anat100:857–864.

Kerley ER. 1965. The microscopic determi-nation of age in human bone. Am J PhysAnthropol 23:149–164.

Kragstrup J, Melsen F. 1983. Three-dimen-sional morphology of trabecular boneosteons reconstructed from serial sec-tions. Metab Bone Dis Relat Res 5:127–130.

Lam L, Lee SW, Suen CY. 1992. Thinningmethodologies: A comprehensive survey.IEEE Trans PAMI 14:869–885.

Lee TC, Kashyap R, Chu CN. 1994. Build-ing skeleton models via 3-D medial sur-face/axis thinning algorithms. CVGIP:Graphical Models Image Processing 56:462–478.

Lindquist WB, Venkatarangan A. 1999. In-vestigating 3D geometry of porous me-dia from high resolution images. PhysChem Earth A24:593–599.

178 THE ANATOMICAL RECORD (PART B: NEW ANAT.) FEATURE ARTICLE

Marotti G, Zallone AZ. 1980. Changes inthe vascular network during the forma-tion of Haversian systems. Acta Anat 106:84–100.

Martin RB, Ishida J. 1989. The relative ef-fects of collagen fiber orientation, poros-ity, density, and mineralization on bonestrength. J Biomech 22:419–426.

McCalden RW, McGeough JA, Barker MB,Court-Brown CM. 1993. Age-relatedchanges in the tensile properties of cor-tical bone. The relative importance ofchanges in porosity, mineralization, andmicrostructure. J Bone Joint Surg Am75:1193–1205.

McCreadie BR, Goulet RW, Feldkamp LA,Goldstein SA. 2001. Hierarchical struc-ture of bone and micro-computed to-mography. Adv Exp Med Biol 496:67–83.

Muller R, Koller B, Hildebrand T, Laib A,Gianolini S, Ruegsegger P. 1996. Resolu-tion dependency of microstructuralproperties of cancellous bone based onthree-dimensional mu-tomography. Tech-nol Health Care 4:113–119.

Muller R, Van Campenhout H, VanDamme B, Van Der Perre G, Dequeker J,Hildebrand T, Ruegsegger P. 1998. Mor-phometric analysis of human bone biop-sies: A quantitative structural compari-son of histological sections and micro-computed tomography. Bone 23:59–66.

Odgaard A, Gundersen HJ. 1993. Quantifi-cation of connectivity in cancellousbone, with special emphasis on 3-D re-constructions. Bone 14:173–182.

Odgaard A, Andersen K, Melsen F, Gun-dersen HJG. 1990. A direct method forfast three-dimensional serial reconstruc-tion. J Microsc 159:335–342.

Ott SM. 2002. Histomorphometric analysisof bone remodeling. In: Bilezikian JP,Raisz LG, Rodan GA, editors. Principlesof bone biology. 2nd ed. New York: Aca-demic Press. p 303–319.

Parfitt AM. 1983. The physiologic and clin-ical significance of bone histomorpho-metric data. In: Recker RR, editor. Bonehistomorphometry: Techniques and in-terpretation. Boca Raton: CRC Press. p143–222.

Parfitt AM, Drezner MK, Glorieux FH, etal. 1987. Bone histomorphometry: Stan-

dardization of nomenclature, symbols,and units. Report of the ASBMR Histo-morphometry Nomenclature Commit-tee. J Bone Miner Res 2:595–610.

Petrtyl M, Hert J, Fiala P. 1996. Spatialorganization of the haversian bone inman. J Biomech 29:161–169.

Peyrin F, Salome M, Cloetens P, Laval-Jeantet AM, Ritman E, Ruegsegger P.1998. Micro-CT examinations of trabec-ular bone samples at different resolu-tions: 14, 7 and 2 micron level. TechnolHealth Care 6:391–401.

Peyrin F, Salome M, Nuzzo S, Cloetens P,Laval-Jeantet AM, Baruchel J. 2000. Per-spectives in three-dimensional analysisof bone samples using synchrotron radi-ation microtomography. Cell Mol Biol(Noisy-le-grand) 46:1089–1102.

Pfeiffer S. 1998. Variability in osteon sizein recent human populations. Am J PhysAnthropol 106:219–227.

Ruth EB. 1947. Gross demonstration of thevascular channels. Anat Rec 98:59–66.

Schaffler MB, Burr DB. 1988. Stiffness ofcompact bone: Effects of porosity anddensity. J Biomech 21:13–16.

Schnapper A, Reumann K, Meyer W. 2002.The architecture of growing compactbone in the dog: Visualization by 3D-reconstruction of histological sections.Ann Anat 184:229–233.

Selle D, Spindler W, Preim B, Peitgen HO.2000. Mathematical methods in medicalimaging: Analysis of vascular structuresfor liver surgery planning. In: Math-ematic Unlimited — 2001 and beyond.New York: Springer. p 1039–1059.

Sensen CW. 2002. Using CAVE® Technol-ogy for functional genomics studies. Di-abetes Technol Ther 4:867–871.

Stein MS, Feik SA, Thomas CD, ClementJG, Wark JD. 1999. An automated anal-ysis of intracortical porosity in humanfemoral bone across age. J Bone MinerRes 14:624–632.

Stout SD, Brunsden BS, Hildebolt CF,Commean PK, Smith KE, Tappen NC.1999. Computer-assisted 3D reconstruc-tion of serial sections of cortical bone todetermine the 3D structure of osteons.Calcif Tissue Int 65:280–284.

Takeuchi A, Uesugi K, Takano H, Suzuki Y.2002. Submicrometer-resolution three-di-mensional imaging with hard x-ray imag-ing microtomography. Rev Sci Instrum 73:4246–4249.

Tappen NC. 1977. Three-dimensional stud-ies of resorption spaces and developingosteons. Am J Anat 149:301–332.

Todd RB, Bowmann W. 1845. The physio-logical anatomy and physiology of man.Philadelphia: Blanchard and Lea.

Tomes J, de Morgan C. 1853. Observationson the structure and development ofbone. Philos Trans R Soc Lond B BiolSci 143:109–139.

Uchiyama T, Tanizawa T, Muramatsu H,Endo N, Takahashi HE, Hara T. 1997. Amorphometric comparison of trabecularstructure of human ilium between mi-crocomputed tomography and conven-tional histomorphometry. Calcif TissueInt 61:493–498.

van Leeuwenhoek A. 1677–1678. Micro-scopical observations of the structure ofteeth and other bones: Made and com-municated, in a Letter by Mr. Antonievan Leeuwenhoek. Philos Trans R SocLond B Biol Sci 12:1002–1003.

Vasciaveo F, Bartoli E. 1961. Vascularchannels and resorption cavities in thelong bone cortex the bovine bone. ActaAnat 47:1–33.

Wachter NJ, Augat P, Krischak GD, Ment-zel M, Kinzl L, Claes L. 2001a. Predictionof cortical bone porosity in vitro by mi-crocomputed tomography. Calcif TissueInt 68:38–42.

Wachter NJ, Augat P, Krischak GD, SarkarMR, Mentzel M, Kinzl L, Claes L. 2001b.Prediction of strength of cortical bone invitro by microcomputed tomography.Clin Biomech (Bristol, Avon) 16:252–256.

Yeni YN, Norman TL. 2000. Fracturetoughness of human femoral neck: Ef-fect of microstructure, composition, andage. Bone 26:499–504.

Yeni YN, Brown CU, Wang Z, Norman TL.1997. The influence of bone morphologyon fracture toughness of the human fe-mur and tibia. Bone 21:453–459.

FEATURE ARTICLE THE ANATOMICAL RECORD (PART B: NEW ANAT.) 179