Received: 23 July 2003Revised: 3 February 2004Accepted: 5 March 2004Published online: 19 May 2004© Springer-Verlag 2004

Abstract The validity of a non-fluo-roscopic fixed-flexion radiographicacquisition and analysis protocol formeasurement of joint space width(JSW) in knee osteoarthritis is deter-mined. A cross-sectional study of165 patients with documented kneeosteoarthritis participating in a mul-ticenter, prospective study of chon-droprotective agents was performed.All patients had posteroanterior,weight-bearing, fixed-flexion radi-ography with 10° caudal beam angu-lation. A specially designed frame(SynaFlexer) was used to standardizethe positioning. Minimum medialand lateral JSW were measured man-ually and twice by an automatedanalysis system to determine inter-technique and intra-reader concor-dance and reliability. A random sub-sample of 30 patients had repeatknee radiographs 2 weeks apart toestimate short-term reproducibilityusing automated analysis. Concor-

dance between manual and automat-ed medial JSW measurements washigh (ICC=0.90); lateral compart-ment measurements showed some-what less concordance (ICC=0.72).There was excellent concordance between repeated automated JSWmeasurements performed 6 monthsapart for the medial (ICC=0.94) andlateral (ICC=0.86) compartments.Short-term reproducibility for thesubsample of 30 cases with repeatacquisitions demonstrated an aver-age SD of 0.14 mm for medial JSW(CV=4.3%) and 0.23 mm for lateralJSW (CV=4.0%). Fixed-flexion radi-ography of the knee using a position-ing device provides consistent, reli-able and reproducible measurementof minimum JSW in knee osteo-arthritis without the need for con-current fluoroscopic guidance.

Keywords Knee · Osteoarthritis ·Radiography · Joint space width

Eur Radiol (2004) 14:1568–1573DOI 10.1007/s00330-004-2312-6 M U S C U L O S K E L E TA L

Manish KothariAli GuermaziGabriele von IngerslebenYves MiauxMartine SieffertJon E. BlockRandall StevensCharles G. Peterfy

Fixed-flexion radiography of the knee providesreproducible joint space width measurementsin osteoarthritis

Introduction

A broad-based research and development effort is under-way to test and commercialize a variety of pharmacolog-ical agents aimed at inhibiting the pathogenetic process-es related to the breakdown of articular cartilage and/orunderlying subchondral bone associated with knee osteo-arthritis [1–4]. Consequently, a critical need has emergedfor accurate and precise radiographic outcome measuresto evaluate the comparative effectiveness of these agentsamong large patient populations participating in multi-center clinical trials. A variety of methods have been

proposed, but standard radiographic assessment of jointspace width (JSW) is generally accepted as the most rel-evant [2]. Indeed, Buckland-Wright et al. [5] demonstra-ted that JSW measurements made in the weight-bearingtunnel view reliably reflected medial compartment carti-lage thinning as determined by double contrast macro-arthography.

Longitudinal changes in JSW measured on serialfilms are small relative to inherent biological or treat-ment response variability. Thus, maintaining a high de-gree of JSW measurement precision is critical. A numberof studies have verified that knee radiographs taken in

M. Kothari · M. Sieffert · J. E. BlockC. G. PeterfyDepartment of Scientific Client Services,Synarc, Inc.,575 Market Street, 17th Floor, San Francisco, CA 94105, USA

A. Guermazi (✉) · G. von IngerslebenY. MiauxDepartment of Reading Services, Synarc, Inc.,575 Market Street, 17th Floor, San Francisco, CA 94105, USAe-mail: ali.guermazi@synarc.comTel.: +1-415-8178992Fax: +1-415-8178999

R. StevensHoffmann-La Roche,Nutley, NJ, USA

1569

the traditional weight-bearing position with full knee ex-tension provide inadequate JSW reproducibility for theevaluation of chondroprotective agents in multi-centertrials of knee osteoarthritis [6–11]. Most of this inherentmeasurement error can be attributed to irreproducibleimage acquisition, particularly with respect to consistentpositioning of the knee [12].

Several authors have confirmed that even minor alter-ations in image acquisition such as the amount of kneeflexion or foot rotation, non-standardized X-ray beamalignment and variable distance between the knee andthe X-ray cassette can have a substantial influence onJSW measurement precision [7–9, 12–15]. Therefore,standardized image acquisition is essential, and somestudies have shown that automated image analysis mayprovide additional improvements in reproducibility [7,16, 17]. Importantly, imaging the knee in approximately20–30° of flexion engages the central posterior aspect ofthe femoral articular cartilage, the region most markedlythinned in the arthritic disease process [18, 19]. How-ever, achieving reproducible joint flexion even with theassistance of fluoroscopic guidance remains a formidablechallenge in knee radiography.

The current paper introduces a technique for imageacquisition and analysis in knee osteoarthritis. Thismethod employs a specially designed positioning frameand calibration phantom that fixes knee flexion and posi-tions the patient’s feet reproducibly. To assure that the X-ray beam is tangential with the floor of the medial tibial plateau, a 10° caudal X-ray beam angulation isused in all cases. These standardized radiographic acqui-sition parameters guarantee that the most affected region

of articular cartilage is imaged reproducibly on initialand follow-up films. This study estimates the short-termJSW reproducibility of this technique and assesseswhether the addition of an automated image analysissoftware method (SynaVu) that corrects for magnifica-tion and parallax effects offers further improvements inprecision compared with manual JSW measurements.

Patients and methods

Patients

Four hundred and fifty seven patients were screened for entranceeligibility to participate in a randomized (Hoffman-La Roche trial),prospective multicenter (58 centers) trial of a chondroprotective osteoarthritis drug. Informed consent and IRB approval were ob-tained for all patients and all sites. Two hundred and thirty eightpatients qualified for inclusion based on clinical and radiographicevidence of osteoarthritis including:

– Chronic knee pain– An Osteoarthritis Research Society International (OARSI)

grade of 1 or 2 [20]– A minimum JSW measurement in the medial compartment of

between 2.0 and 4.0 mm determined by manual measurement

Approximately 70% (165/238) of these patients from 47 centerselected to participate and were monitored prospectively with serialradiography of both knees and follow-up clinical assessment. Theaverage age of these patients was 60.5 years (±8 years), and mostpatients were female (119/165, 72%). The baseline radiographs of these patients represent the study material for the current in-vestigation and include a random sample of 30 knee radiographs(30 patients) obtained from 14 geographically dispersed centerswhere standardized knee radiography was repeated twice within a 2-week period to estimate short-term JSW measurement preci-

Fig. 1 Schematic illustration ofthe non-fluoroscopic fixed-flexion radiographic knee pro-tocol with 10° caudal beam angulation to ensure alignmentof the beam with the medialtibial plateau. A standardizeddegree of knee flexion (20°)and external foot rotation (5°)are achieved with use of theSynaFlexer calibration and positioning frame

sion. The mean age of this patient subsample was 59.4 years (±6.8 years) and 53% (16/30) of the cases were female.

Radiographic technique

A standardized radiographic technique was utilized to image theknee in all cases. Weight-bearing, posteroanterior (PA) radiogra-phy was undertaken without concurrent fluoroscopic guidance.Uniform anatomical alignment of the knee joint was facilitated byuse of a proprietary positioning frame and calibration phantom(SynaFlexer) that places the patient’s feet reproducibly in 5° exter-nal rotation. The beam was centered on the back of the knees atthe level of the joint line, defined by the horizontal skin crease ofthe popliteal fossa. Use of this positioning device requires that thethigh, patella and pelvis remain flush with the frame, coplanarwith the tips of the great toes, resulting in a fixed knee angulationof approximately 20° flexion (i.e., “fixed-flexion”). Importantly,10° caudal X-ray beam angulation provides correct projection ofthe tibial spines between the femoral condyles and superimposi-tion of the anterior and posterior tibial rims. Fig. 1 illustrates thestandardized patient positioning achieved with the SynaFlexerframe. A representative fixed-flexion PA radiograph showing con-current imaging of magnification markers in the positioning frameis provided in Fig. 2.

Analytical methods

Minimum JSW was determined by a trained radiologist from allradiographs using a manual measurement technique with a gradu-ated magnifying lens. Medial and lateral compartment measure-ments were made separately in all cases. JSW measurements werealso conducted by a second trained radiologist using an automatededge detection and analysis software system (SynaVu). This auto-mated image processing engine has been previously validated [21,22] and was licensed from the University of California at SanFrancisco (UCSF). This automated analysis was repeated by thesame radiologist blinded to the initial results 6 months later. Theanalysis software suite utilizes calibration information obtainedfrom the SynaFlexer phantom during radiography to verify beamangulation and knee side (right or left) and to correct for magnifi-cation differences and parallax errors between films. In all cases,automated results are presented with magnification correction.This image acquisition and analysis system provides automatedfilm digitization and the software edge enhances the digital imagesusing unsharp masking to maximize the delineation of cortical andtrabecular bone (Fig. 3). Image quality also can be optimized us-ing window and level adjustment, zoom and panning functions, arobust edge-detection algorithm to demarcate the weight-bearingarticular surfaces as well as to calculate minimum JSW, averageJSW and joint space area. To enhance methodological integrityduring clinical trials, the system provides reader and chronologicalblinding in addition to automated databasing and archiving of im-ages and JSW results.

Statistical methods

Utilizing the baseline films from the 165 patients participating inthe prospective treatment study of knee osteoarthritis, the concor-dance and association between manual JSW measurements and theinitial automated measurements with magnification correctionwere estimated using the intra-class correlation coefficient (ICC)[23, 24] and Pearson’s correlation coefficient [25], respectively,for medial and lateral compartments separately. Using the samestudy material, the initial and repeat automated measurementswere compared using similar statistical procedures as above, andthe reproducibility was estimated by calculating the corresponding

1570

coefficient of variation (CV). The short-term reproducibility ofminimum JSW by the magnification-corrected automated analysisprocedure was estimated by calculating the average standard devi-ation (SD) between measurements from the subset of paired radio-graphs (n=30) taken 2 weeks apart. The corresponding CV wascalculated as an estimate of the magnitude of between-subjectvariability in JSW relative to the mean of JSW.

Results

A high degree of concordance and association was dem-onstrated between manual and magnification-correctedautomated baseline JSW measurements in the medialcompartment among the study group of 165 patients

Fig. 2 Typical PA fixed-flexion radiograph showing simultaneousimaging of magnification markers incorporated in the SynaFlexercalibration and positioning frame. Note the proper alignment ofthe tibial plateau with superimposition of the anterior and posteri-or tibial rims

Fig. 3 The SynaVu image analysis system showing automatedJSW measurement. In addition to automatically verifying kneeside and beam angulation and correcting magnification and paral-lax, the software automatically traces the articular surfaces of thefemorotibial joints and localizes and calculates minimum medialand lateral JSW as well as joint space area. The three vertical linesrepresent the capability of the software to restrict the analysis tothe desired areas of interest

By standardizing radiographic acquisition parameterssuch as the amount of knee flexion, degree of foot rota-tion and X-ray beam angulation and by controlling eval-uation factors including magnification errors, the currentautomated technique either corrects or attenuates theknown variability associated with many of the factorsidentified to degrade measurement precision [9, 14]. Thefixed-flexion technique is essentially similar to the LyonSchuss view [10], but without the necessity of fluoro-scopic guidance to assist in aligning the X-ray beam.Piperno et al. [10] reported a short-term JSW repro-ducibility with a corresponding CV of 3.5% for the LyonSchuss view, an estimate that compares favorably withthe precision demonstrated herein using the fixed-flexionmethod. However, the fixed-flexion method does not require a dedicated technologist trained in fluoroscopicexamination and spares the patient the added cost and radiation exposure associated with fluoroscopy.

Other radiographic methods have been developed tomeasure minimum JSW and to allow for some degree ofknee flexion. A semi-flexed (i.e., 7–10°) anteroposteriorview with fluoroscopic guidance has demonstrated rea-sonable short-term reproducibility (CV=5.5%) in 25 ar-thritic cases imaged at a single center [7], but consider-ably worse measurement precision (CV=8.7%) when thesame technique was employed across five clinical cen-ters [13]. Center-to-center variability was common withalmost 30% of paired radiographs judged unsatisfactorywith respect to the degree of knee flexion or foot rotation[13].

A non-fluoroscopically assisted protocol utilizing thesame semi-flexed position with a PA horizontal beamhas been proposed [8, 26]. In this method, the first meta-tarsophalangeal (MTP) joints are positioned directly be-neath the front surface of the X-ray film cassette; patel-lae remain in contact with the cassette and aligned verti-cally with the first MTP joints. External foot rotation isstandardized at 15° and maintained reproducibly with theaid of individual foot maps. Short-term reproducibilityhas been reported to be excellent (CV=1.6%) in a singlecenter series of cases [8] and considerably worse(CV=8.0%) in a field test of the procedure [26]. In bothreports of this technique, parallel alignment of the tibialplateau and the X-ray beam occurred in only about 30%of cases [8, 26]. This problem, coupled with the fact thatthe slight degree of flexion achieved (i.e., 7–10°) fails to capture radiographically the most affected articular region, raises some concern about its use in the serial assessment of chondroprotective agents.

We noted somewhat worse inter-technique and intra-reader concordance and reliability for lateral compared tomedial compartment JSW measurements. However, short-term reproducibility estimates for the subsample of repeatknee radiographs demonstrated similarly low imprecisionfor medial (CV=4.3%) and lateral (CV=4.0%) compart-ment measurements. Buckland-Wright et al. [7] likewise

1571

(330 knees). The ICC and Pearson’s correlation coeffi-cient were both 0.90 for this comparison. A lesser degreeof concordance and association was observed for lateralcompartment JSW measurements with an ICC of 0.72and a Pearson’s coefficient of 0.74, respectively. Fig. 4illustrates superimposed frequency distributions of medi-al JSW measurements for manual and automated mea-surements separately. Inspection of these distributionsshows that only four baseline measurements (2.4%) were<2.0 mm by automated measurement, and these caseswould not have qualified for study participation based onthe study inclusion criterion of minimum medial JSW of2.0–4.0 mm by manual measurement.

There was excellent concordance and association be-tween automated JSW measurements repeated 6 monthsapart by the same reviewer using the SynaVu softwaresystem with magnification correction. For the medialcompartment, the ICC and Pearson’s coefficient were0.94 and 0.95, respectively. The corresponding CV was3.9%. For the lateral compartment, the ICC and Pear-son’s coefficient were 0.86 and 0.89, respectively, andthe corresponding CV was 5.6%.

Among the subsample of 30 cases with fixed-flexionknee radiography repeated twice in a 2-week period, theshort-term reproducibility in the medial compartment us-ing automated analysis was estimated by an average SDof 0.14 mm and a corresponding CV of 4.3%. In the lat-eral compartment, differences measured between repeatradiographs resulted in an average SD of 0.23 mm and acorresponding CV of 4.0% using automated analysis.

Discussion

The standardized radiographic technique described in thecurrent study fixes the knee reproducibly in approxi-mately 20° flexion, thereby engaging the articular aspectof the medial compartment most severely affected inknee osteoarthritis, the central posterior region [18, 19].

Fig. 4 Superimposed frequency distributions for 165 patients (330knees) of minimum medial JSW by manual and automated mea-surement

1572

noted inconsistencies with respect to JSW measurementprecision between the medial and lateral compartment,with the medial compartment generally offering somewhatbetter reproducibility overall. Nevertheless, minimum me-dial JSW is uniformly considered the standard measure-ment parameter for clinical study eligibility and for moni-toring longitudinal articular changes [12].

Using 2.0-mm medial JSW as the lower cutoff eligi-bility criterion for study participation, we found that onlyfour out of 165 patients that were eligible by manualmeasurement would have been excluded by the higherprecision automated SynaVu analysis system with magni-fication correction. The results of our correlation analysisindicated that manual measurements of JSW, when per-formed by an experienced radiologist, were concordantand associated with automated measurement methods. Incontrast to longitudinal assessments of joint space nar-rowing (i.e., change in JSW), in which measurement pre-cision is critical, the rate of incorrect inclusion or exclu-sion in patient selection was not substantially impactedby the choice of measurement technique in this study.

Conclusion

The current study provides initial validation results forthe fixed-flexion radiographic technique in determiningJSW in knee osteoarthritis. A high degree of concor-dance was demonstrated between manual and automatedmeasurement techniques as well as when the automatedmeasurements were repeated by the same reader. More-over, short-term measurement reproducibility was high.These findings were particularly impressive in that thestudy material of knee radiographs was obtained from 47clinical centers overall and from 14 centers for the sub-sample of cases with repeat radiographs. This suggeststhat the SynaFlexer positioning frame aids in eliminatingmany of the image acquisition errors that plague non-standardized radiographic protocols used to evaluateJSW.

Acknowledgment This image acquisition for this study wasfunded by Hoffman-La Roche. Authors thank Melissa M. Ta fromSynarc for providing the drawings.

References

1. Hochberg MC, Altman RD, BrandtKD, Moskowitz RW (1997) Designand conduct of clinical trials in osteo-arthritis: preliminary recommendationsfrom a task force of the OsteoarthritisResearch Society. J Rheumatol24:792–794

2. Lequesne M, Brandt K, Bellamy N et al (1994) Guidelines for testing slowacting drugs in osteoarthritis. JRheumatol [Suppl] 41:65–71; discus-sion 72–63

3. Lequesne M, Brandt K, Bellamy N et al (1995) Guidelines for testing slowacting drugs in arthritis—addendum. J Rheumatol 22:1442

4. Reginster JY, Deroisy R, Rovati LC et al (2001) Long-term effects of glucosamine sulphate on osteoarthritisprogression: a randomised, placebo-controlled clinical trial. Lancet357:251–256

5. Buckland-Wright JC, Macfarlane DG,Lynch JA, Jasani MK, Bradshaw CR(1995) Joint space width measures cartilage thickness in osteoarthritis of the knee: high resolution plain filmand double contrast macroradiographicinvestigation. Ann Rheum Dis54:263–268

6. Brandt KD, Fife RS, Braunstein EM,Katz B (1991) Radiographic grading of the severity of knee osteoarthritis:relation of the Kellgren and Lawrencegrade to a grade based on joint spacenarrowing, and correlation with arthro-scopic evidence of articular cartilagedegeneration. Arthritis Rheum34:1381–1386

7. Buckland-Wright JC, Macfarlane DG,Williams SA, Ward RJ (1995) Accura-cy and precision of joint space widthmeasurements in standard and macro-radiographs of osteoarthritic knees.Ann Rheum Dis 54:872–880

8. Buckland-Wright JC, Wolfe F, WardRJ, Flowers N, Hayne C (1999) Sub-stantial superiority of semiflexed(MTP) views in knee osteoarthritis: acomparative radiographic study, with-out fluoroscopy, of standing extended,semiflexed (MTP), and schuss views. J Rheumatol 26:2664–2674

9. Fife RS, Brandt KD, Braunstein EM etal (1991) Relationship between arthro-scopic evidence of cartilage damageand radiographic evidence of jointspace narrowing in early osteoarthritisof the knee. Arthritis Rheum34:377–382

10. Piperno M, Hellio Le Graverand MP,Conrozier T, Bochu M, Mathieu P, Vignon E (1998) Quantitative evalua-tion of joint space width in femorotibialosteoarthritis: comparison of three radiographic views. Osteoarthritis Cartilage 6:252–259

11. Spector TD, Dacre JE, Harris PA,Huskisson EC (1992) Radiologicalprogression of osteoarthritis: an 11year follow-up study of the knee. Ann Rheum Dis 51:1107–1110

12. Mazzuca SA, Brandt KD, Katz BP(1997) Is conventional radiographysuitable for evaluation of a disease-modifying drug in patients with kneeosteoarthritis? Osteoarthritis Cartilage5:217–226

13. Mazzuca SA, Brandt KD, Buckland-Wright JC et al (1999) Field test of thereproducibility of automated measure-ments of medial tibiofemoral jointspace width derived from standardizedknee radiographs. J Rheumatol26:1359–1365

14. Ravaud P, Auleley GR, Chastang C et al (1996) Knee joint space widthmeasurement: an experimental study ofthe influence of radiographic procedureand joint positioning. Br J Rheumatol35:761–766

15. Ravaud P, Giraudeau B, Auleley GR et al (1998) Variability in knee radio-graphing: implication for definition ofradiological progression in medial kneeosteoarthritis. Ann Rheum Dis57:624–629

1573

16. Dacre JE, Huskisson EC (1989) Theautomatic assessment of knee radio-graphs in osteoarthritis using digitalimage analysis. Br J Rheumatol28:506–510

17. Ravaud P, Chastang C, Auleley GR et al (1996) Assessment of joint spacewidth in patients with osteoarthritis ofthe knee: a comparison of 4 measuringinstruments. J Rheumatol23:1749–1755

18. Messieh SS, Fowler PJ, Munro T(1990) Anteroposterior radiographs ofthe osteoarthritic knee. J Bone JointSurg Br 72:639–640

19. Vignon E, Conrozier T, Piperno M,Richard S, Carrillon Y, Fantino O(1999) Radiographic assessment ofhip and knee osteoarthritis. Recom-

mendations: recommended guidelines.Osteoarthritis Cartilage 7:434–436

20. Altman RD, Hochberg M, Murphy WAJr, Wolfe F, Lequesne M (1995) Atlasof individual radiographic features inosteoarthritis. Osteoarthritis Cartilage 3[Suppl A]:3–70

21. Duryea J, Li J, Peterfy CG, Gordon C,Genant HK (2000) Trainable rule-based algorithm for the measurementof joint space width in digital radio-graphic images of the knee. Med Phys27:580–591

22. Peterfy C, Li J, Zaim S et al (2003)Comparison of fixed-flexion position-ing with fluoroscopic semi-flexed posi-tioning for quantifying radiographicjoint-space width in the knee: test-retest reproducibility. Skeletal Radiol32:128–132

23. Bartko JJ (1966) The intraclass correla-tion coefficient as a measure of reli-ability. Psychol Rep 19:3–11

24. Muller R, Buttner P (1994) A criticaldiscussion of intraclass correlation coefficients. Stat Med 13:2465–2476

25. Altman DG (1991) Practical statisticsfor medical research. Chapman & Hall,London

26. Mazzuca SA, Brandt KD, BuckwalterKA, Lane KA, Katz BP (2002) Fieldtest of the reproducibility of the semi-flexed metatarsophalangeal view in repeated radiographic examinations ofsubjects with osteoarthritis of the knee.Arthritis Rheum 46:109–113