ARTHRITIS & RHEUMATISMVol. 46, No. 2, February 2002, pp 385–393DOI 10.1002/art.10108© 2002, American College of RheumatologyPublished by Wiley-Liss, Inc.

Magnetic Resonance Imaging of Normal andOsteoarthritic Trabecular Bone Structure in the Human Knee

Olivier Beuf,1 Srinka Ghosh,2 David C. Newitt,1 Thomas M. Link,3 Lynne Steinbach,1

Michael Ries,2 Nancy Lane,1 and Sharmila Majumdar2

Objective. To use high-resolution magnetic reso-nance imaging (MRI) to evaluate the trabecular bonestructure in the distal femur and the proximal tibia andits to correlate the findings with different stages ofosteoarthritis (OA) of the human knee.

Methods. Axial images of the distal femur andproximal tibia were obtained at 1.5 T in patients withoutand with mild OA and with severe OA. The spatialresolution was 195 � 195 �m2 with a 1-mm slicethickness. Apparent measures of trabecular bone vol-ume fraction (BV/TV), trabecular number (Tb.N), tra-becular separation (Tb.Sp), and trabecular thickness(Tb.Th) were calculated.

Results. Significant differences existed in the tra-becular bone structure of the femur and tibia. Differ-ences in trabecular bone structure between the tibia andthe femur decreased with the degree of OA. The appar-ent BV/TV, Tb.N, and Tb.Sp in the femoral condylescould be used to differentiate healthy patients or pa-tients with mild OA from patients with severe OA (P <0.05). Among individuals, the structural variation of thelateral and medial femoral condyle was indicative of theextent of the disease.

Conclusion. High-resolution MRI of the kneejoint can provide a noninvasive assessment of trabecu-lar bone structure. Trabecular bone structure, deter-

mined by high-resolution MRI, shows significant varia-tion in patients with varying degrees of OA. The impactof OA on trabecular bone is different in the tibia than inthe femur, and this difference depends on the extent ofthe disease.

Osteoarthritis (OA) is a multifactorial diseasecharacterized by the progressive loss of articular hyalinecartilage and the development of altered joint congru-ency, subchondral sclerosis, intraosseous cysts, and os-teophytes. It affects about 14% of the adult population(1) and is the second most common cause of permanentdisability among subjects over the age of 50 years (2). Inaddition to changes in articular cartilage that occur inOA, it has been suggested that early changes are seen inthe adjoining subchondral and trabecular bone (3).

In a guinea pig model, Layton et al (4) observedwith microscopic computed tomography (CT) that aninitial loss of trabecular bone volume fraction andthinning of trabeculae was followed, in the advancedstages of OA, by an increase of trabecular bone volumefraction via eventual thickening of trabeculae. In acanine OA model induced by anterior cruciate ligament(ACL) transection, Dedrick et al (5) demonstrated anincrease in subchondral bone thickness, accompanied bya decrease in trabecular thickness. Using magnificationradiographs of humans, Lynch et al (6) showed anincrease in the horizontal trabecular thickness of thetibia in early OA, followed by an increase in verticalconnectivity of trabeculae in advanced disease. Morerecently, Buckland-Wright and colleagues (7) showedthat the structural changes in the trabecular bone micro-architecture in patients with ACL ruptures are detectableby fractal analysis of radiographs, well before joint spacenarrowing and other radiologic changes are detectable.

Imaging and assessment of OA are based primarilyon plain radiography (8). While radiographic changesreflect the pathologic changes in the bone and joint space,

Supported by NIH grants R03-AG-16388, R01-AR-46905,and P01-AR-43584.

1Olivier Beuf, PhD, David C. Newitt, PhD, Lynne Steinbach,MD, Nancy Lane, MD: University of California, San Francisco;2Srinka Ghosh, MS, Michael Ries, MD, Sharmila Majumdar, PhD:University of California, San Francisco, and University of California,Berkeley; 3Thomas M. Link, MD: Technische Universitat Munchen,Munich, Germany.

Address correspondence and reprint requests to SharmilaMajumdar, PhD, Department of Radiology, Magnetic ResonanceScience Center, University of California, San Francisco, 1 IrvingStreet, San Francisco, CA 94143-1290. E-mail: sharmila.majumdar@mrsc.ucsf.edu.

Submitted for publication May 29, 2001; accepted in revisedform October 1, 2001.

385

these changes do not generally correlate with the sever-ity of pathologic joint destruction. Furthermore, conven-tional radiographs only depict gross osseous changesdirectly, and such changes tend to occur late in thedisease. Early changes in the articular cartilage andother articular tissues are not directly visible. Cartilageloss can only be inferred from the development of jointspace narrowing. CT provides cross-sectional images ofthe affected joint, and thus diminishes the problem ofoverlapping structures. However, CT is limited in termsof soft-tissue contrast and imaging planes.

Magnetic resonance imaging (MRI) is ideal formonitoring OA. It is nonionizing, offers multiplanarcapabilities, has high spatial resolution, and providessuperior depiction of soft tissue detail. In addition toevaluation of articular cartilage volume, thickness, anddegeneration (9–14), recent advances have made itpossible to use MRI to assess bony and soft tissuechanges. In a guinea pig model, results showed that MRimages of trabecular bone accurately reflected the de-gree of osteopenia and the development of subchondralsclerosis and osteophytes (15). In vitro studies performedon small cubic specimens have shown that some of theparameters derived from high-resolution MR images con-tributed to the prediction of biomechanical properties ofbone (16). In the last decade, numerous studies have alsodemonstrated the feasibility of MR micro-imaging of tra-becular structure in humans in vivo at a resolution ade-quate for visualizing trabecular bone (17,18).

The assessment of human trabecular bone struc-ture using MRI techniques showed that it was possible toquantify the trabecular architecture and derive suchmeasures as trabecular width, trabecular bone volumefraction, and mean intercept length, as well as quantita-tive measures of texture, such as the fractal characteris-tics of the trabecular bone network (19–21). Majumdaret al (20) derived structural parameters in premeno-pausal healthy women and in postmenopausal womenwith osteoporosis and correlated the data with peri-pheral quantitative CT bone mineral density (BMD) andspinal fracture status. Although MR techniques mayresult in partial volume effects because the spatialresolution is comparable with trabecular dimensions,Kothari et al (22) have shown that these effects may beminimized when the lower resolution of the slice isselected along the direction of primary trabecular ori-entation. In this study, dedicated MR techniques wereused in vivo to characterize and quantify variations inthe trabecular bone structure along the distal femur andproximal tibia of OA patients and normal subjects.Differences of structure between the tibia and femur in

joints of patients with different stages of OA were alsoinvestigated.

PATIENTS AND METHODS

The distal femur and the tibial plateau of 28 subjectsdivided into 3 groups were investigated. Group I included 10young healthy subjects (6 men and 4 women; age 129 � 4.9years) with no knee impairment. The other 2 groups consistedof patients with OA, classified according to the radiography-based Kellgren/Lawrence (K/L) scale (23). Group II included8 patients with mild OA (1 man and 7 women; age 68 � 9.1years) with K/L scores of 1–2, and group III included 10patients with severe OA (4 men and 6 women; age 70 � 6.3years), with K/L scores of 3–4. Reproducibility was assessed in4 healthy patients (2 men and 2 women; age 32 � 5.8 years),with repositioning between 3 measurements.

The examinations were performed in accordance withthe rules and regulations of the University of California, SanFrancisco Human Research Committee. Informed consent wasobtained from all patients after the nature of the examinationshad been fully explained.

Imaging. The images were acquired on a Signa 1.5T-echo-speed system (General Electric Medical Systems, Mil-waukee, WI) equipped with gradients operating at 2.2 G/cmand a rise time of 184 �sec. To ensure accuracy in subjectpositioning as well as to reduce motion artifacts, the leg of eachsubject was placed on a dedicated holder (manufacturedin-house) and immobilized using Velcro straps. A bilateraldual-phased array coil (USA Instruments, Cleveland, OH)consisting of 4 elements was used to image the knee joint. Coilpositioning was critical to maximize the signal-to-noise ratio. Asagittal localizer was used to identify the 2 regions of interest,the proximal tibia and the distal femur, in the axial plane(Figure 1A).

High-resolution MR images were obtained using a3-dimensional (3-D) fast gradient-echo sequence (17) with apartial echo acquisition (echo time [TE] 4.5 msec, repetitiontime [TR] 30 msec, 40° flip angle, �15.6 kHz bandwidth). Atotal of 60 images (1 mm thick) were obtained in the axialplane with a field of view (FOV) of 10 cm and an imagingmatrix size of 512 � 384 pixels, corresponding to a recon-structed spatial resolution of 195 � 195 �m2. Two sets of 60slices were collected to cover the knee joint, as shown in Figure1. The scan time was �12 minutes per acquisition, amountingto a total of 24 minutes for the entire anatomic coverage.

In addition, routine clinical sequences were used forradiologic grading purposes. Coronal 3-D T1-weighted images(TR 25 msec, TE 3 msec, FOV 10 cm, slice thickness 2 mm,matrix 512 � 256 � 32) were used to detect osteophytes andsclerosis. Sagittal T2-weighted (TR 3 sec, TE 70 msec) andT1-weighted (TR 500 msec, TE 11 msec) images were obtainedto detect the presence of edema, subchondral cysts, and jointeffusion, and to assess the state of ligaments and menisci.Based on these clinical scans, 2 radiologists (TML and LS)identified by consensus whether the lateral or medial side wasmost affected by the disease. The total time of a typicalimaging session was 60 minutes.

Image analysis. The reconstructed volumetric datawere transferred to a Sun workstation (Sun Microsystems,

386 BEUF ET AL

Mountain View, CA). Images were analyzed using processingsoftware developed at our laboratory using IDL (ResearchSystems, Boulder, CO) programming language.

Since the images were acquired with a set of surfacecoils, the reception profile was inhomogeneous, leading to anintensity inhomogeneity in the image (Figures 1B and D).Prior to a quantitative analysis of the trabecular bone struc-ture, a 3-D low-pass (LP) filter–based correction algorithm wasapplied (24). The corrected image was obtained by dividing theinitial image by an LP-filtered image obtained from theoriginal using Gaussian k-space filtering. In the ideal case,choice of an appropriate filter bandwidth allowed us to removevariations due to the coil reception profile without loss ofanatomic details (Figures 1C and E).

The first (proximal) and last (distal) 5 slices of thevolume were eliminated from the analysis to minimize artifactsfrom slice selection profile imperfection (Figure 1A). Regionsof interest (ROIs) fitting the trabecular bone and marrowregions in both the distal femur and the proximal tibia weredrawn manually over 35–45 slices of each data set (Figures 2Aand B). Individual ROIs were clustered in 4 distinct groups forstatistical analysis. Region R1 included ROIs selected in theproximal tibia, starting at the level of the subchondral bone.Region R2 included ROIs in the femur, starting from the pointwhere the medial and lateral condyles merged and going alongthe shaft. Regions R3 and R4 included ROIs drawn in themedial femoral condyle and lateral femoral condyle, respec-tively (Figures 2C and D).

In order to use the standard stereologic techniques andquantify the trabecular bone network, images were thresh-olded and segmented into bone and marrow phases. Theproblem of segmenting 3-D images, where the image resolu-tion approximates that of the trabeculae, is a critical issue. The

threshold depends significantly on the imaging modality andmay vary depending on the operator. Based on a dual refer-ence limit and assuming a biphasic model, as previouslyvalidated (20), a global threshold was calculated and theimages were binarized into a bone phase and a marrow phase.The segmented binary image was used to compute (per sliceand ROI) trabecular bone parameters, such as the apparenttrabecular bone volume fraction (BV/TV), apparent trabecular

Figure 1. Images of regions of interest acquired by high-resolution (HR) magnetic resonanceimaging (MRI). A, Sagittal localizer image showing prescribed volumes for the axial HR images inthe proximal tibia (solid line) and distal femur (dashed line). An overlap of 10 mm along thesuperior-inferior direction was used to ensure good continuity between the 2 scans. B and D,Representative HR MR images from the proximal tibia and the distal femur. C and E, Results afterlow-pass filter–based correction.

Figure 2. Axial high-resolution images. Regions of interest are out-lined in A, the proximal tibia and B, the distal femur. C and D,Analyzed regions R1, R2, R3, and R4 on reformatted coronal images.

TRABECULAR ARCHITECTURE OF THE KNEE USING HIGH-RESOLUTION MRI 387

number (Tb.N), apparent trabecular separation (Tb.Sp), andapparent trabecular thickness (Tb.Th) analogous to standardbone histomorphometry (25). (All parameters discussed in thisstudy are apparent parameters.) The average time to performthe structure analysis of each volume was 15–20 minutes on anUltra-Sparc 1 (Sun).

For each region, spatial variations of the structureparameters along the slice direction were obtained to deter-mine localized differences and regional variations, starting atthe joint line and extending into the shaft. The average valuesfor regions R1–R4 were calculated for statistical analysis. Allstatistical computations were performed using JMP software(SAS Institute, Cary, NC). Differences between the individualgroups were evaluated using Student’s 2-tailed t-tests of signif-icance. P values less than 0.05 were considered significant.

RESULTS

As shown in Figures 1C and E, the MR imagesclearly depicted the trabecular bone microarchitecturein the femur and the tibia. Coil intensity correctionimproved the signal homogeneity in the plane of theimages. Due to coil sensitivity dropoff on the axial plane,this was a crucial first step to quantitative analysis.

Trabecular structure variations as a function ofthe distance from the joint line. For each volunteer, theanalysis of the trabecular structure varied substantiallywithin each ROI and between the femur and the tibia,especially when evaluated in progression from the meta-physis to the epiphysis. Representative graphs of trabec-ular spacing, thickness, and number in a young, healthyvolunteer are shown in Figure 3, from the shaft to thesubchondral region toward the joint line for the tibia andfrom the subchondral region to the shaft away from thejoint line for the femur. In the tibia, the shaft showed adecrease in the Tb.Sp up to the growth plate and anincrease in the Tb.N, while the Tb.Th remained approx-

Figure 3. Representative data for a young, healthy subject, showingmeasured structural parameters in region R1 of the proximal tibia andin region R2 of the distal femur. Tb.Sp � apparent trabecularseparation; Tb.Th � apparent trabecular thickness; Tb.N � apparenttrabecular number.

Figure 4. Mean values of morphologic parameters calculated in the tibia (region R1) and in thefemur (region R2). Group I � healthy subjects; group II � patients with mild osteoarthritis (OA);group III � patients with severe OA. � � P � 0.01; # � P � 0.05. Values are the mean and SD.BV/TV � apparent trabecular bone volume fraction; Tb.Sp � apparent trabecular separation;Tb.N � apparent trabecular number; Tb.Th � apparent trabecular thickness.

388 BEUF ET AL

imately constant. A valley in the Tb.Sp and a peak in theTb.N marked the growth line. In contrast, in the femur,a peak in the Tb.Sp and a valley in the Tb.N wereobserved at the growth line. Progression from the growthline to the shaft yielded a decrease in the Tb.Sp and anincrease in the Tb.N.

Difference in trabecular bone structure betweenthe tibia and the femur. Mean values were calculatedper patient group for each morphologic parameter in thetibia (region R1) and the femur (region R2); these areshown in Figure 4. The BV/TV, Tb.N, and Tb.Th wereall typically higher in the femur, and the Tb.Sp was lowerthan in the tibia.

The absolute differences in the mean parametervalues between the tibia and the femur are summarizedin Table 1. The absolute value was chosen so that themagnitude of the difference could be evaluated. Thesignificance of the differences observed between thetibia and the femur was established using Student’spaired t-test. When compared with groups I and II,patients with severe OA (group III) had substantiallylower absolute differences in mean values for all para-meters, with the exception of Tb.Th. There were signif-icant differences (P � 0.05) between the tibia and thefemur for patients from groups I and II for all parame-ters. For group III, significant difference was obtainedonly for BV/TV.

Differences in trabecular bone structure betweencontrols and patients with OA. The morphologic para-meter variations as a function of group are shown inFigure 4. Based on data reported in Table 2, a significantnonmonotonic variation was observed in the tibia forBV/TV with disease progression, while the other para-meters showed no specific trend. Based on a group ofhealthy patients, no correlation was found betweenmeasures of the structure and age.

Absolute differences and significance of mean

parameter values between the 3 groups are presented inTable 2. The analysis showed that mean values forBV/TV and Tb.Sp were significantly different (P � 0.05)between groups I and II in both the tibia and the femurand between groups I and III in the femur. Additionally,BV/TV differences between groups II and III in the tibiaand Tb.N differences between groups I and III in thefemur were significant. No significant differences inTb.Th were seen between any groups.

Differences in trabecular structure between thelateral and medial femoral condyles. In regions R3 andR4, corresponding respectively to the medial and lateralfemoral condyles (Figure 2), the mean values of mor-phologic parameters were also calculated. For eachderived trabecular bone structure parameter, the 3subject groups were compared using the average andabsolute difference of the population mean, calculatedin the 2 condyles. Graphs illustrating the results areshown in Figure 5.

For patients from group III, the BV/TV in thecondyle with the most severe degenerative changes washigher than the average femoral value (region R2), whilethe BV/TV was lower in the least diseased condyle.However, as seen from Figures 4 and 5, the averagevalue of the 2 femoral condyles and the mean valuesobtained in region R2 of the femur were nearly thesame. As seen from the graphs, the absolute differencefor all morphologic parameters was found to be signifi-cantly higher (P � 0.05) for group III compared withgroups I and II, which had no significant structuraldifferences between the condyles.

In addition to analyses of the mean and differ-ence, the variations in bone structure when progressingfrom the joint line to the shaft of the femur were alsostudied in regions R3 and R4. In a normal femoralcondyle (as graded by radiologists [TML and LS]), with

Table 1. Absolute difference in the mean parameter values betweenthe tibia and the femur in each group of subjects*

Group BV/TV Tb.N Tb.Th Tb.Sp

I 0.051† 0.087† 0.024† 0.075†II 0.047‡ 0.102‡ 0.021† 0.099‡III 0.023† 0.005 0.020 0.022

* BV/TV � apparent trabecular bone volume fraction; Tb.N �apparent trabecular number; Tb.Th � apparent trabecular thickness;Tb.Sp � apparent trabecular separation. Group I � healthy subjects;group II � patients with mild osteoarthritis (OA); group III � patientswith severe OA.† P � 0.01.‡ P � 0.05.

Table 2. Absolute difference of the mean parameter values for thetibia (region R1) and the femur (region R2) between different groupsof subjects*

Region, group BV/TV Tb.N Tb.Th Tb.Syp

R1I and II 0.040† 0.106 0.012 0.087†II and III 0.032† 0.048 0.018 0.053I and III 0.006 0.058 0.007 0.033

R2I and II 0.041† 0.093 0.015 0.064†II and III 0.007 0.048 0.018 0.023I and III 0.034† 0.140† 0.003 0.087†

* See Table 1 for definitions.† P � 0.05.

TRABECULAR ARCHITECTURE OF THE KNEE USING HIGH-RESOLUTION MRI 389

increasing distance from the joint line, an increase inBV/TV and a decrease in Tb.Sp was observed. A dis-eased femoral condyle, however, showed a decrease or avery small increase in BV/TV and an increase or a verysmall decrease in Tb.Sp (Figure 6).

Using linear regression, the slope of the variation

in each bone parameter with respect to the distancefrom the joint line was determined. The results fordetermined parameters from all 10 patients from groupIII patients are presented in Figure 6. For BV/TV, allpatients with mild OA had a positive slope in bothcondyles, whereas greater differences (positive and neg-

Figure 5. Absolute difference and average value calculated from mean values obtained in themedial femoral condyle (region R3) and in the lateral femoral condyle (region R4) for all derivedtrabecular bone structure parameters. Values are the mean and SD. BV/TV � apparent trabecularbone volume fraction; TbSp � apparent trabecular separation; TbN � apparent trabecularnumber; TbTh � apparent trabecular thickness.

Figure 6. Calculated slopes for all derived trabecular bone structure parameters on the lateral(region R4) and medial (region R3) sides of femoral condyles in patients with severe osteoarthritis(OA) (group III). The side most affected by OA, according to the radiologist’s reading, is indicated.BV/TV � apparent trabecular bone volume fraction; TbSp � apparent trabecular separation;TbN � apparent trabecular number; TbTh � apparent trabecular thickness.

390 BEUF ET AL

ative slopes) in the condyles of the same patient weremeasured in the severe OA group. As can be seen inthese graphs, there was a correlation between the slopecharacterizing the BV/TV rate of variation and the sidemost affected by disease, as shown by the results of theclinical readings of the MR images. Thus, the femoralbone structure variations close to the joint line, asquantified by the slope direction, appeared to be indic-ative of the disease extent in advanced OA patients.

Reproducibility of the technique, including scanand quantification processes, was assessed using thecoefficient of variation (26). The reproducibility of theimage acquisition and segmentation with quantificationof bone structure parameters in the femur (region R2)was 5.4, 2.9, 2.7, and 5.2% for BV/TV, Tb.N, Tb.Th, andTb.Sp, respectively, and 4.0, 3.3, 1.4, and 4.6% in thetibia (region R1).

DISCUSSION

In this study, using high-resolution images cou-pled with image processing, we evaluated the trabecularbone architecture in the proximal tibia and the distalfemur in healthy subjects as well as in patients with mildand severe OA. As discussed earlier, knowing that thepredominant trabecular orientation in the femur andtibia is along the shaft as a result of the loading functionof these long bones, we acquired images in the axialdirection in order to reduce the partial volume effects ofthe slice dimension being thicker than the in-planeresolution (22).

Due to spatial resolution limitations, parametersrelated to trabecular bone structure suffer from partialvolume effects. The structure parameters assessed differfrom those derived using histomorphometry, and there-fore are considered “apparent” structure parameters.Majumdar et al (27) have shown in an experimentalmodel that as the resolution decreases, there is anoverestimation of the bone volume fraction and trabec-ular thickness as well as an underestimation of trabecu-lar separation. Although absolute MR-derived measuresdiffer from histomorphometry measures, it has beendemonstrated that measures of structure derived fromMR images contribute to the assessment of trabecularbone strength with good correlations (22).

All quantified morphologic parameters werefound to be significantly different between the tibia andthe femur in both groups I and II (Table 1). The BV/TV,Tb.N, and Tb.Th are all typically higher in the femur,and the Tb.Sp is lower than in the tibia. We hypothesizethat this is due to differences in loading function. The

convexity of the distal femur, and the relative flatness ofthe proximal tibia have specific implications on bothlocal loading environments. Trabecular bone is moredense in the distal femur, where loading forces areconcentrated in the 2 condyles, in contrast with the tibia,where loading forces are evenly distributed on the wholesurface of the tibial plateau.

For group III, however, only BV/TV was signifi-cantly different between the tibia and the femur (Table1). Except for the Tb.Th, the differences in bone struc-ture between the tibia and the femur decreased withincreased severity of OA. From group I to group III, thiswas mainly due to a reduction in the BV/TV and anincrease in the trabecular spacing in the femur (Figure 4).This finding indicates a change in the loading function withthe progression of OA, as a result of which, in late OA, thefemoral trabecular bone is lost, and this loss is akin toosteopenic changes compared with the baseline status.

Variations in mean structure parameters betweensubject groups indicated that the MR-derived structuremeasures BV/TV and Tb.Sp were the most sensitiveparameters (Table 2), with significant differences in thetibia as well as in the femur. These parameters weresignificantly different in both the tibia and the femurbetween groups I and II, and in the femur betweengroups I and III (P � 0.05). Studies correlating biome-chanical strength of bone cubes from the spine withMR-derived structure measures (28) yielded similarfindings, with the best correlations found for BV/TV andTb.Sp.

The difference in the mean parameter valuesbetween groups I and II was 12.3, 6.6, 6.1, and 13.2% forBV/TV, Tb.N, Tb.Th, and Tb.Sp, respectively, in thefemur (region R2) and 13.3, 8.2, 5.3, and 15.6% in thetibia (region R1). Accordingly, differences betweengroups I and III were 10.2, 10.0, 1.2, and 18.0%,respectively, in the femur and 2.2, 4.4, 3.1, and 6.0% inthe tibia. These differences are overall a factor of 2 ormore greater than the reproducibility of the measureswe have demonstrated.

Comparing the absolute differences of the sam-ple mean between the medial and lateral femoral con-dyles permits us to differentiate between subjects with-out OA versus those with severe OA as well as betweensubjects with mild versus severe OA. These results havebeen obtained with all of the structure parameters, butthe differences were most significant for BV/TV, Tb.Sp,and Tb.Th (Figure 5).

Thus, in summary, we found that as a result ofloading differences and biomechanical loads, the struc-ture of trabecular bone is different in the tibia compared

TRABECULAR ARCHITECTURE OF THE KNEE USING HIGH-RESOLUTION MRI 391

with the femur. The bone is denser in the femoralcondyles and femur in general; however, with diseaseprogression to severe OA, the trabecular bone structureis lost in the femur compared with the tibial structure.The significance of these results will lie in the manage-ment of patients with OA, and once the relationshipbetween bone changes and cartilage changes is estab-lished, modification of biomechanical loading and gaitcould ameliorate the course of the disease.

Another finding of this study was that the axialvariation of the bone structure in the femoral condyleswas indicative of disease extent. There was an asymme-try of the medial and lateral condyles related to theextent of disease in subjects with severe OA. Indeed, inpatients from group III, the slope direction as deter-mined by a linear fit of the morphologic parametersBV/TV and Tb.Sp, with respect to the distance from thejoint line, was found to be predictive of the side mostaffected by the disease (Figure 6). Given that the BV/TVvalues at the point where the 2 condyles merge (regionR2) are the same, a negative slope of BV/TV in 1condyle corresponds to an increased mean BV/TV onthat side (most affected) and vice versa. This indicatesthat a modification of the trabecular structure leading tobone sclerosis appears in the most affected condyle. Thisresult confirms the findings of a previous study per-formed by Layton et al (4), in which femoral heads froma guinea pig model of OA showed a highly significantincrease in bone volume fraction.

Using macroradiographs, Buckland-Wright et al(29) have recently shown in human tibia that ACLrupture leads to thickening of subchondral horizontaltrabeculae in the medial tibial compartment of knees.No significant changes were detected in the lateralcompartment. But whether the bony changes detectedprecede those of early OA is still under investigation.Unlike other imaging modalities, MRI offers multipla-nar capabilities, is capable of directly visualizing all partsof the joint simultaneously, and provides informationregarding the relationship between bone and cartilage.MRI was previously used to monitor bony changes inOA in a guinea pig model (30). The investigators foundthat trabecular bone accurately reflected the degree ofosteopenia. Trabecular thinning was also noted aroundthe cruciate ligament insertion.

Radin and Rose (31) postulated that increasedstiffness in the subchondral bone was responsible for theinitiation of cartilage damage in OA. A sharp gradient instiffness in the bone under articular cartilage increasesstress in the cartilage layer. In patients with severe OA,the bone volume fraction values were highest in the most

diseased femoral condyle. For non-OA patients, theBV/TV was lower near the subchondral bone, in theepiphysis, and increased toward the shaft, while in OApatients, this gradient was reversed. One explanation forthe reversal of the gradient in patients with severe OA isinitial trabecular bone hypertrophy, which then in-creases cartilage stresses, leading to OA.

Seven of the 10 subjects with severe OA consti-tuting group III were predominantly affected on the medialside. This is consistent with the mean loading forces in theknee, which do not act in the middle of the joint but aremore centered in the medial compartment (32).

The results presented here provide an insight intothe amplitude and trend of trabecular bone structuremodifications with disease progression. A potential lim-itation of this study, however, is that the subjects ingroup I, which was used as the reference group, werehealthy, with no knee impairment, but were not age-matched with groups II and III.

Future studies will extend these MRI and analysismethods to both cross-sectional and longitudinal studiesfor assessment of the progression of OA. The former isnecessary to increase the patient population per group,and thereby increase the statistical relevance of theresults. Currently both trabecular bone and cartilagedata are being acquired on all subjects for the purpose ofinvestigating the dynamics between changes in the tra-becular bone architecture and degeneration of cartilage.Establishing the magnitude of these changes and theirinterrelationships and chronologic progression with car-tilage degeneration may assist in understanding themechanism for postinjury cartilage and joint degenera-tion. The objective of this study was to establish thathigh-resolution MRI could provide an efficient tech-nique for early detection of changes in periarticularbone relative to OA.

MRI is an emerging clinical modality that allowsinvestigation of the clinical relationship between boneand cartilage and the onset and progression of OA. Inthe future, high-resolution MRI of trabecular bonestructure may become a very powerful tool in providingquantitative and objective parameters in the early stagesof joint disease. More specifically, derived structureparameters from femoral condyles could be a potentialmarker for evaluating disease progression. Adjustingbiomechanical loads, either by modification of gait or byosteotomies, may potentially modulate the changes in-duced by the progression of OA. From a clinical per-spective, the ability to monitor cartilage and bonechanges in 3 dimensions would thus have a tremendousimpact on overall patient management. Furthermore, if

392 BEUF ET AL

the intrinsic connection between bone and articularcartilage is established, it may be possible to preservearticular cartilage by moderating bone turnover andquality. These applications, although speculative, doindicate the tremendous potential of MR as a means tomonitor not only trabecular and subchondral bone andcartilage, but also the whole joint, in OA.

ACKNOWLEDGMENTS

The authors would like to thank Dr. T. K. Tran, Dr. A.Laib, and Dr. A. Shimakawa for helpful discussions, as well asDr. Vikas Patel, Andrew Burghardt, and Cynthia Frazier forassistance with data analysis.

REFERENCES

1. Forman MD, Malamet R, Kaplan D. A survey of osteoarthritis ofthe knee in the elderly. J Rheumatol 1983;10:282–7.

2. Peyron JG. The epidemiology of osteoarthritis. In: Moskowitz RWet al, editors. Osteoarthritis: Diagnosis and Medical/Surgical Man-agement, 1st ed. Philadelphia: W. B. Saunders; 1984. p. 9–27.

3. Dedrick DK, Goulet R, Huston L, Goldstein SA, Bole GG. Earlybone changes in experimental osteoarthritis using microscopiccomputed tomography. J Rheumatol 1991;Suppl. 27:44–5.

4. Layton MW, Goldstein SA, Goulet RW, Feldkamp LA, KubinskiDJ, Bole GG. Examination of subchondral bone architecture inexperimental osteoarthritis by microscopic computed tomography.Arthritis Rheum 1988;31:1400–5.

5. Dedrick DK, Goldstein SA, Brandt KD, O’Connor BL, GouletRW, Albrecht M. A longitudinal study of subchondral plate andtrabecular bone in cruciate-deficient dogs with osteoarthritis fol-lowed up for 54 months. Arthritis Rheum 1993;36:1460–7.

6. Lynch JA, Hawkes DJ, Buckland-Wright JC. Analysis of texture inmacroradiographs of osteoarthritic knees using the fractal signa-ture. Phys Med Biol 1991;36:709–22.

7. Buckland-Wright JC, Lynch JA, Macfarlane DG. Fractal signatureanalysis measures cancellous bone organisation in macroradio-graphs of patients with knee osteoarthritis. Ann Rheum Dis1996;55:749–55.

8. Altman RD, Fries JF, Bloch DA, Carstens J, Cooke TD, GenantH, et al. Radiographic assessment of progression in osteoarthritis.Arthritis Rheum 1987;30:1214–25.

9. Disler DG, McCauley TR, Kelman CG, Fuchs MD, Ratner LM,Wirth CR, et al. Fat-suppressed three-dimensional spoiledgradient-echo MR imaging of hyaline cartilage defects in the knee:comparison with standard MR imaging and arthroscopy. AJRAm J Roentgenol 1996;167:127–32.

10. Eckstein F, Adam C, Sittek H, Becker C, Milz S, Schulte E, et al.Non-invasive determination of cartilage thickness throughout jointsurfaces using magnetic resonance imaging. J Biomech 1997;30:285–9.

11. Bashir A, Gray ML, Burstein D. Gd-DTPA2� as a measure ofcartilage degradation. Magn Reson Med 1996;36:665–73.

12. Kim DK, Ceckler TL, Hascall VC, Calabro A, Balaban RS.Analysis of water-macromolecule proton magnetization transfer inarticular cartilage. Magn Reson Med 1993;29:211–5.

13. Morris GA, Freemont AJ. Direct observation of the magnetizationexchange dynamics responsible for magnetization transfer contrastin human cartilage in vitro. Magn Reson Med 1992;28:97–104.

14. Stammberger T, Eckstein F, Englmeier KH, Reiser M. Determi-nation of 3D cartilage thickness data from MR imaging: compu-tational method and reproducibility in the living. Magn ResonMed 1999;41:529–36.

15. Watson PJ, Carpenter TA, Hall LD, Tyler JA. Cartilage swelling andloss in a spontaneous model of osteoarthritis visualized by magneticresonance imaging. Osteoarthritis Cartilage 1996;4:197–207.

16. Majumdar S, Kothari M, Augat P, Newitt DC, Link TM, Lin JC,et al. High-resolution magnetic resonance imaging: three-dimen-sional trabecular bone architecture and biomechanical properties.Bone 1998;22:445–54.

17. Jara H, Wehrli FW, Chung H, Ford JC. High-resolution variableflip angle 3D MR imaging of trabecular microstructure in vivo.Magn Reson Med 1993;29:528–39.

18. Wehrli FW, Ford JC, Haddad JG. Osteoporosis: clinical assess-ment with quantitative MR imaging in diagnosis. Radiology 1995;196:631–41.

19. Majumdar S, Genant HK. Magnetic resonance imaging in osteo-porosis. Eur J Radiol 1995;20:193–7.

20. Majumdar S, Genant HK, Grampp S, Newitt DC, Truong VH, LinJC, et al. Correlation of trabecular bone structure with age, bonemineral density, and osteoporotic status: in vivo studies in thedistal radius using high resolution magnetic resonance imaging.J Bone Miner Res 1997;12:111–8.

21. Majumdar S, Genant HK. Assessment of trabecular structureusing high resolution magnetic resonance imaging. Stud HealthTechnol Inform 1997;40:81–96.

22. Kothari M, Keaveny TM, Lin JC, Newitt DC, Genant HK,Majumdar S. Impact of spatial resolution on the prediction oftrabecular architecture parameters. Bone 1998;22:437–43.

23. Hart DJ, Spector TD. The classification and assessment of osteo-arthritis. Baillieres Clin Rheumatol 1995;9:407–32.

24. Wald LL, Carvajal L, Moyher SE, Nelson SJ, Grant PE, BarkovichAJ, et al. Phased array detectors and an automated intensity-correction algorithm for high-resolution MR imaging of the hu-man brain. Magn Reson Med 1995;34:433–9.

25. Parfitt AM, Drezner MK, Glorieux FH, Kanis JA, Malluche H,Meunier PJ, et al. Bone histomorphometry: standardization of no-menclature, symbols, and units. J Bone Miner Res 1987;2:595–610.

26. Gluer CC, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK.Accurate assessment of precision errors: how to measure thereproducibility of bone densitometry techniques. Osteoporos Int1995;5:262–70.

27. Majumdar S, Newitt DC, Mathur A, Osman D, Gies A, Chiu E, etal. Magnetic resonance imaging of trabecular bone structure in thedistal radius: relationship with x-ray tomographic microscopy andbiomechanics. Osteoporos Int 1996;6:376–85.

28. Link TM, Majumdar S, Lin JC, Newitt D, Augat P, Ouyang X, etal. A comparative study of trabecular bone properties in the spineand femur using high resolution MRI and CT. J Bone Miner Res1998;13:122–32.

29. Buckland-Wright JC, Lynch JA, Dave B. Early radiographic featuresin patients with anterior cruciate ligament rupture. Ann Rheum Dis2000;59:641–6.

30. Watson PJ, Hall LD, Malcolm A, Tyler JA. Degenerative jointdisease in the guinea pig: use of magnetic resonance imaging tomonitor progression of bone pathology. Arthritis Rheum 1996;39:1327–37.

31. Radin EL, Rose RM. Role of subchondral bone in the initiationand progression of cartilage damage. Clin Orthop 1986;213:34–40.

32. Bohringer ME, Beyer WF, Weseloh G. Comparative histomor-phometry of subchondral bone density and articular cartilagethickness in the tibial head in early human arthritis. Z Orthop IhreGrenzgeb 1995;133:291–302.

TRABECULAR ARCHITECTURE OF THE KNEE USING HIGH-RESOLUTION MRI 393