Femoral neck anteversion measurement using linear slot scanning radiography

Medical Engineering and Physics 38 (2016) 187–191

Contents lists available at ScienceDirect

Medical Engineering and Physics journal homepage: www.elsevier.com/locate/medengphy

Technical note

Femoral neck anteversion measurement using linear slot scanning radiography Chipo Chimhundu a, Sudesh Sivarasu a, Stefan Steiner a,c, Julian Smit b, Tania S. Douglas a,∗ a

MRC/UCT Medical Imaging Research Unit and Division of Biomedical Engineering, University of Cape Town, South Africa Division of Geomatics, University of Cape Town, South Africa c Lodox Systems, Sandton, South Africa b

a r t i c l e

i n f o

Article history: Received 12 November 2014 Revised 30 August 2015 Accepted 20 November 2015

Keywords: Femoral anteversion X-ray stereophotogrammetry Orthopedic measurement

a b s t r a c t Measurements between anatomical landmarks on radiographs are useful for diagnosis and treatment planning in the orthopedic field. Direct measurement on single radiographic images, however, does not truly reflect spatial relationships, as depth information is lost. We used stereo images from a slot scanning X-ray machine to estimate coordinates of three-dimensional (3D) bony landmarks for femoral neck anteversion (FNA) measurement. A set of 7 landmarks consisting of the centre of the femoral head; the centre of the base of the femoral neck; the medial and lateral condyles; the medial and lateral posterior condyles; and finally the centre of the knee; were found to be identifiable and suitable for radiographic measurement. The reconstructed 3D coordinates were then used to define the 3D geometry of the anatomical axes required to estimate FNA. Stereophotogrammetric measurements on a sample of 30 dry right adult femurs were compared to reference values obtained using the Kingsley Olmstead method applied to photographic images. A strong positive correlation (0.998) was found and the mean ± standard deviation of the stereophotogrammetric approach (13.08 ± 6.87)° was comparable to that of the Kingsley Olmstead method (13.14 ± 6.88)°. Intra- and inter-observer reliability were high, with the lower bound of the 95% confidence interval above 0.98 for the intra-class correlation coefficient. The results merit further validation against three dimensional imaging technology such as computed tomography, to confirm stereophotogrammetry as a suitable alternative for FNA measurement. © 2015 IPEM. Published by Elsevier Ltd. All rights reserved.

1. Introduction Femoral neck anteversion (FNA) is the extent to which the femoral neck deviates forward from the transcondylar axis of the femur. FNA measurement aids hip stability assessments in orthopedics. Pre-operatively, FNA measurements enable prediction of the likely effects of surgery on patient gait and joint stability. Selection of suitable surgical implants and optimal implant positioning during surgery requires FNA information [1,2]. Clinical procedures which rely on FNA measurements include primary total hip arthroplasty, proximal femoral replacement [3], femoral derotation osteotomy [2,4] and total knee arthroplasty [2]. Abnormal FNA is commonly related to trauma in the lower extremities, effects of prior surgery, degenerative bone disease [1], developmental diseases such as cerebral palsy [5], knee joint instability [6], congenital dislocations of the hip and labral tears [7], slipped capital femoral epiphysis [1,6,7], and Legg–Calve–Perthes disease [5,7]. FNA is also believed to be a femoral neck fracture risk (FNFR) ∗

Corresponding author. Tel.: +27214066633; fax: +27214487226. E-mail address: [email protected] (T.S. Douglas).

http://dx.doi.org/10.1016/j.medengphy.2015.11.017 1350-4533/© 2015 IPEM. Published by Elsevier Ltd. All rights reserved.

predictor and aids understanding of relationships between femoral neck bone mineral density [8], FNFR and the geometry of the femoral neck axis [9]. FNA is geometrically defined in Fig. 1, which was drawn using 3D CT data. The naming convention for axes and landmarks in Fig. 1 is adapted from [4].The angle is calculated at the intersection of two planes, namely, the anteversion plane of the femur (APf ) and condylar plane of the femur (CPf ). APf is the plane containing the femoral neck axis (NAf in Fig. 1)- which passes through the centre of the femoral head (H) and the centre of the base of the femoral neck (O)- as well as the long axis of the femur (LAf ) -which passes through O and the centre of the knee (K). The second plane CPf contains LAf and is parallel to the transcondylar axis of the femur (CAf ). CAf is an axis parallel to the lateral and medial posterior condyles (Lpc and Mpc, respectively) as shown in Fig. 1. Image-based techniques that have been used for FNA measurement include the method developed in 1956 by Magiligan et al. and described in the work of Cibulka [7]; Herman bi-planar radiography [10]; computed tomography (CT) and magnetic resonance imaging (MRI) [5]; fluoroscopy [11]; and ultrasonography

188

C. Chimhundu et al. / Medical Engineering and Physics 38 (2016) 187–191

Fig. 1. Geometric definition of femoral neck anteversion (FNA). CAf = condylar axis of the femur; Lpc and Mpc = lateral and medial posterior condyles, respectively; NAf = neck axis of the femur; K = centre of the knee; H = centre of the femoral head; O = centre of the base of the femoral neck; LAf = long axis of the femur.

[12]. Generally, radiographic measurement methods, which make use of single radiographs, are incapable of adequately representing the 3D geometry of FNA and therefore often yield measurement inaccuracies. For example, ultrasonography has been found to overestimate FNA [5] with high measurement variability [13]. In order to adequately measure the FNA angle, 3D information is required, thus 3D imaging technologies such as CT and MRI have the advantage. MRI has the added advantage of non-ionizing radiation but is not suitable for postoperative assessment of patients with metal implants and is expensive. 3D volumetric femoral reconstructions from CT scans yield the most accurate FNA measurements, with average errors of 0.45° [14]. However, the high radiation exposure associated with CT is a concern. Axial CT slices instead of 3D CT have been used to estimate FNA in order to lower radiation exposure [5,6,14,15], with the most accurate axial CT method yielding accuracies of ± 1° [4]. An alternative to 3D imaging is the use of stereophotogrammetry and multi-view images to account for depth. However, literature on stereophotogrammetric measurement of FNA is limited. The first documented stereophotogrammetric FNA measurement technique [16] yielded measurement errors of ± 2.46° and was cumbersome as it involved generating 3D optical models from stereoscopic radiographs which required special stereoscopes for viewing. With the evolution of technology, low dose bi-planar radiography technologies present an opportunity for 3D measurement. As an example, the EOS radiographic system has been shown to yield FNA measurement accuracies comparable to those of CT [17]. The EOS system simultaneously captures two images and uses a model based approach to estimate a 3D patient specific model for measurement. Such systems present an alternative to imaging modalities that are inherently 3D such as MRI and CT.

We investigate the possibility of using the Lodox Statscan to acquire stereo images of the femur for FNA measurement using stereophotogrammetry. The Statscan uses linear slot-scanning radiography (LSSR) for image acquisition, which consists of a linearly moving focal spot with a narrow slot through which X-rays pass as a very narrow fan-beam. Statscan dose for an antero-posterior radiograph of the pelvis in adults has been reported as 60 μSv [18]. The difference between this modality and the bi-planar EOS system is that the Statscan acquires multi-view images in succession with a C-arm which can be rotated to acquire images from different perspectives. The Statscan does not use models for full 3D reconstruction but enables stereophotogrammetric measurement of lengths and angles through 3D reconstruction of points [19]. The aim of this investigation was to measure FNA using X-ray stereophotogrammetry applied to slot-scanning X-ray images acquired using the Lodox Statscan, taking into account the 3D geometry of the axes required for measurement. A preliminary investigation using dry bone specimens was conducted. The possibility of stereo-photogrammetric measurement of FNA would reduce radiation exposure and cost compared to CT, and would remove the need for full 3D reconstructions for radiographic measurement as is done for EOS.

2. Materials and methods A sample of 30 dry adult femurs were obtained from the Anatomy Museum at the University of Cape Town. Other studies have used a similar sample size [4,16,20]. Only normal femurs of the right side were used; none were diseased or deformed. No gender or ethnicity information was available for the samples.

C. Chimhundu et al. / Medical Engineering and Physics 38 (2016) 187–191

189

The Kingsley–Olmstead method was used as the gold standard for comparison. This method has been used for comparison with new measurement approaches since 1948 [11,12,14,21]. For a femur whose most posterior aspects of the condyles rest in contact with a flat horizontal surface, the Kingsley–Olmstead method takes the horizontal plane to represent a plane approximately parallel to the true condylar plane (CPf of Fig. 1). The angle of anteversion is then manually measured by looking straight down the femur from its proximal end and measuring the angle between the neck and horizontal axes using a goniometer (the 2D horizontal axis in this instance represents CPf of Fig. 1). Modern implementations of the Kingsley–Olmstead method use digital photographs of a femur and image analysis software as reported by Citak et al. [11]. 2.1. FNA measurement on the Lodox Statscan Radiographic FNA measurement protocols in literature vary mainly due to the different techniques used to estimate the centre of the base of the femoral neck across different modalities. In order to determine the best sets of landmarks for LSSR, two sets of trial landmarks were investigated and were defined as follows: •

•

SetA consists of the landmarks H, O, K, Lpc and Mpc as indicated in Fig. 1, where K is the mid-point of M and L (medial and lateral epicondyles) and Mpc and Lpc are the medical and lateral posterior condyles, respectively. SetB contains landmarks identical to set A, except that the posterior superior prominence of the greater trochanter (G) was used instead of O to define the femoral neck axis as reported by Citak et al. [11].

Prior to measurement, a femur specimen would be placed on the imaging trolley, with the most posterior aspects of the condyles in contact with the trolley top and the long axis of the femur along the scan direction along the centre line of the scanner to create the same radiographic environment across the samples, thus minimizing sources of error. The separation angle of the radiographic images was set to 75°. The complete set of user selected corresponding landmarks are shown in Fig. 2. The estimations of the projections of the centre of the head (H) in the antero-posterior (AP) and oblique lateral view respectively were done by finding the centre of an oval approximated using image processing software (ImageJ) as demonstrated in Fig. 2. Similarly, the projections of the centre of the base of the femoral neck (O) were calculated by the intersection of the line from the centre of the femoral head, running along the midline of the neck and an approximation of the inter-trochanteric line projected onto the two dimensional images, which approximates the base of the femoral neck. Where visualization was compromised by overlapping structures, epipolar constraints were applied as reported by Chimhundu et al. [19]. Epipolar geometry can be used to relate two images of a scene based on their relative viewpoints. It also reduces searches for correspondences from an entire image to a single line [22–24]. Stereophotogrammetry measurement accuracy relies on accuracy of point selection in images; therefore epipolar geometry was used to aid corresponding point selection when landmark visualization was poor. A visual example of the use of epipolar lines to aid selection of the lateral posterior condyle is shown in the AP view of Fig. 2, where the posterior aspects of the condyles are not visible due to overlapping structures, but are more visible in the oblique lateral view. Selection of l’pc in the lateral view enables the generation of an epipolar line from which lpc can be estimated in the AP view. The visible edges of the condyles, together with the epipolar line would be used to calculate centroid of the posterior condyle. For Lpc1 and Lpc2 representing points lying on the epipolar line as well as on the edges of the lateral posterior condyle, the midpoint

Fig. 2. Corresponding landmark selection from stereo LSSR images for FNA measurement; {g, h, o, l, m, lpc, mpc} and {g’, h’, o’, l’, m’, l’pc , m’pc } anterior projections and oblique lateral projections of {G, H, O, L, M, Lpc and Mpc }, respectively; lpc1 and lpc2 = point selections for estimation of lpc .

of lpc1 and lpc2 in was used to approximate the most posterior aspect of the lateral condyle lpc . The same was done for estimating mpc. Inconsistent estimation of landmarks required to calculate O has been found to cause high measurement variability [17], therefore the selection of O in the lateral view was also guided by selecting corresponding points on an epipolar line drawn from the anterior selection of the projection of O, which intersects with the line from h’ running down the midline of the femoral neck axis as illustrated in the oblique lateral view of Fig. 2. 2.2. Calculation of FNA angle After selection of landmarks in the AP and oblique lateral views, the FNA angle would then be calculated using the reconstructed 3D landmarks for both sets A and B of landmarks as follows: •

•

Calculate the centre of the knee K, given by the midpoint of L and M. Calculate nAPf , which is a 3 × 1 unit normal vector to a plane APf containing: ◦ H, O and K for setA. ◦ H, G and K for setB. For example, nAPf of the plane containing H, O and K for setA would be given by the cross product:

nAPf = norm((H − O ) × (K − O ))

(1)

The unit normal to APf for setB would use G in the place of O. •

Calculate a 3 × 1 unit normal vector nCPf to a plane CPf , which is a plane containing O and K and parallel to a line joining Lpc and Mpc for setA; while CPf is a plane containing G and K, being parallel to a line joining Lpc and Mpc for setB as shown in Fig. 1. For example, nCPf for set A would be given by

nCPf = norm((O − K ) × (Mpc −Lpc ))

(2)

190

C. Chimhundu et al. / Medical Engineering and Physics 38 (2016) 187–191 Table 1 Statistical measures for FNA measurements obtained using a single rater and two sets of trial landmarks A and B.

Reference Set A Set B

•

Pearson correlation coefficient

Mean ±standard deviation

Mean error ±standard deviation

– 0.998 0.748

13.14 ± 6.88 13.08 ± 6.87° 22.5 ± 7.10°

– 0.40 ± 0.22° 9.37 ± 5.19°

°

Table 2 Mean Absolute Error (MAE), Intra-class Correlation Coefficients (ICC) and Confidence interval (CI) for FNA Measurements (2 sets by each of 2 raters).

The angle of anteversion FNAθ is therefore given by:



FNAθ = cos−1

n

•n

 CPf   APf  nCPf •nAPf 



(3)

MAE Rater 1

0.40

°

2.3. Statistical analysis

Rater 2

0.49

Two sets of stereophotogrammetric FNA measurements were taken by a single rater for each femur using landmark sets A and B, and averaged. Together with the mean and standard deviation of the measurements, the Pearson correlation coefficient was used to test which of the two sets A and B showed better agreement with the photographic reference values. The result was used to determine which set would be suitable for further analysis. A second rater was introduced and tasked with taking two sets of FNA measurements for the selected set. For each rater, the mean absolute error of measurement (MAE) was calculated using Eq. 4,

Rater 1 and 2

<0.5°

MAE =

1 n |re fi − xi | i=1 n

(4)

where n is the total number of femurs, refi is the reference FNA value for femur i and xi is the corresponding stereophotogrammetric measurement . The MAE was used to assess measurement accuracy with respect to the reference values. The intra-class correlation coefficient (ICC) was then used as a measure of intra-rater reliability; it was calculated by considering repeated measurements as ratings from two independent raters as reported by Gwet [25] with a two-way model by Shrout and Fleiss [26], to indicate the reliability of measurement. MedCalc® Software was used for the ICC calculations [27]. The 95% confidence interval was also calculated. To obtain inter-rater reliability, the repeated measurement sets from each rater were averaged and the ICC was applied to the averaged measurements. Both single measures and average measures ICC were determined to assess the reliability of measurement; single measures ICC describes the reliability that can be expected when a single rater is used to measure FNA, while average measures ICC describes the reliability that can be expected if measurements from different raters are averaged and treated as the FNA measurement. The 95% confidence interval was also calculated. 3. Results and Interpretation The results summarized in Table 1 show a strong agreement between setA and the reference values as seen by the Pearson correlation coefficient (0.998). For Set B, the higher mean FNA for the sample (22.5° ) as well as mean error (9.37° ) suggests that using the landmark G, which is suggested in [11], tends to over-estimate the angle of anteversion by about 9 degrees on the Statscan. For Set A, the mean ± standard deviation values for Statscan FNA measurements on the sample of adult femurs (13.14 ± 6.88° ) are comparable to those measured on the EOS (13.4 + 9.1° ) [17], suggesting that the mean FNA of the adult population that was sampled for the study is similar to that used in other studies. However, normal values vary depending on age, ethnicity and gender [1,7,21,28,29].

°

Measure

ICC, 95% CI [lower limit, upper limit]

Single Average Single Average Single Average

ICC>0.9991,CI [0.9981, 0.9996] ICC>0.9996,CI [0.9991, 0.9998] ICC>0.9879,CI [0.9747, 0.9942] ICC>0.9939,CI [0.9872, 0.9971] ICC>0.9957,CI [0.9870, 0.9982] ICC>0.9978,CI [0.9935, 0.9991]

Table 2 is a summary of the reliability of measurements expressed by mean absolute error (MAE) and ICC for FNA measurements using set A only for two independent raters. Table 2 shows high intra-rater reliability and inter-rater reliability. Both raters had a mean absolute error in measurement of less than ± 0.5° against the reference. Herman biplanar radiography shows accuracy of 2° for FNA measurement [10]; radiographic methods involving 3 scans proposed in [4] show accuracies of 1° and fluoroscopy has been seen to show measurement accuracies of 1.4° with the use of the greater trochanter being more reliable for measurement [11]. However, the use of the greater trochanter did not prove valuable for the Statscan, possibly due to the difference in the scanning geometries of the two modalities. Our method which had average errors of approximately 0.5° considering both raters, is potentially close to that of CT, which shows accuracy of 0.45° [14], although further validation is required against CT. Compared to earlier attempts at stereophotogrammetric measurement in [16] which yielded a measurement accuracy of 2.24° and standard deviation of 1.48° , our method has better accuracy and is less cumbersome. The results obtained provide grounds for further research involving larger samples and comparison with 3D imaging technology. This study was conducted in a controlled setting where the femurs would be placed in a specific and repeatable orientation for FNA measurement, which does not necessarily correspond to radiographic protocol in vivo, but enabled comparison with the established Kingsley–Olmstead method. 4. Conclusion We have presented a method for stereo-photogrammetric FNA measurement from linear slot-scanning X-ray images. While 3D imaging technology is reliable and clinically approved for FNA measurement, the use of 2D X-ray modalities such as the Statscan would be beneficial in settings without access to 3D imaging modalities. With the stereophotogrammetric approach to FNA measurement showing promising results, there is scope for further validation of the method by means of comparison to 3D imaging systems. Conflict of interest All authors were fully involved in the study and preparation of the manuscript. The manufacturer of the imaging system used in the study provided partial funding for the study. One of the

C. Chimhundu et al. / Medical Engineering and Physics 38 (2016) 187–191

authors (SS) is jointly employed by the University of Cape Town and the manufacturer of the imaging system. Acknowledgments Funding was provided by Lodox Systems and the Technology and Human Resources for Industry Programme (THRIP) of the National Research Foundation (NRF). We thank Mr Jonathan Glenday and Dr Tinashe Mutsvangwa who assisted with the geometric representation of femoral neck anteversion. Finally, Mr. William Wasswa is acknowledged for assistance with measurement. References [1] Gulan G, Matovinovic D, Nemec B, Rubinic B, Ravlik-Gulan J. Femoral neck anteversion: values, development, measurement, common problems. Coll Antropol Feb. 2000;24(2):521–7. [2] Subburaj K, Ravi B, Agarwal M. Computer-aided methods for assessing lower limb deformities in orthopaedic surgery planning. Comput Med Imaging Graph Jun. 2010;34(4):277–88. [3] Lackman RD, Torbert JT, Finstein JL, Ogilvie CM, Fox EJ. Inaccuracies in the assessment of femoral anteversion in proximal femoral replacement prostheses. J Arthroplast Jan. 2008;23(1):97–101. [4] Murphy S, Simon S, Kijewski P, Wilkinson R, Griscom N. Femoral anteversion. J Bone Jt Surg Oct. 1987;69(8):1169–76. [5] Guenther KP, Tomczak R, Kessler S, Pfeiffer T, Puhl W. Measurement of femoral anteversion by magnetic resonance imaging — evaluation of a new technique in children and adolescents. Eur J Radiol Nov. 1995;21(1):47–52. [6] Botser IB, Ozoude GC, Martin DE, Siddiqi AJ, Kuppuswami S, Domb BG. Femoral anteversion in the hip: comparison of measurement by computed tomography, magnetic resonance imaging, and physical examination. Arthrosc J Arthrosc Relat Surg May 2012;28(5):619–27. [7] Cibulka MT. Determination and significance of femoral neck anteversion. Phys Ther Jun. 2004;84(6):550–8. [8] Cheng XG, Nicholson PHF, Boonen S, Brys P, Lowet G, Nijs J, Dequeker J. Effects of anteversion on femoral bone mineral density and geometry measured by dual energy X-ray absorptiometry: a cadaver study. Bone Jul. 1997;21(1):113– 17. [9] Bryan R, Nair PB, Taylor M. Use of a statistical model of the whole femur in a large scale, multi-model study of femoral neck fracture risk. J Biomech Sep. 2009;42(13):2171–6. [10] Hermann KL, Egund N. Measuring anteversion in the femoral neck from routine radiographs. Acta Radiol Jul. 1998;39(4):410–15. [11] Citak M, Gardner MJ, Citak M, Krettek C, Hüfner T, Kendoff D. Navigated femoral anteversion measurements: a new intraoperative technique. Injury Apr. 2008;39(4):467–71.

191

[12] Moulton A, Upadhyay S. A direct method of measuring femoral anteversion using ultrasound. J Bone Joint Surg Br Aug. 1982;64-B(4):469–72. [13] Prasad S, Bruce C, Crawford S, Higham J, Garg N. Femoral anteversion in infants: a method using ultrasound. Skeletal Radiol Aug. 2003;32(8):462–7. [14] Kim JS, Park TS, Park SB, Kim JS, Kim SI. Measurement of femoral neck anteversion in 3D. Part 2:3D modelling method. Med Biol Eng Comput Nov. 2000;38(6):610–16. [15] Suh KT, Kang JH, Roh HL, Moon KP, Kim HJ. True femoral anteversion during primary total hip arthroplasty: use of postoperative computed tomography– based sections. J Arthroplast Jun. 2006;21(4):599–605. [16] Wientroub S, Boyde A, Chrispin A, Lloyd-Roberts G. The use of stereophotogrammetry to measure acetabular and femoral anteversion. J Bone Joint Surg Br May 1981;63-B(2):209–13. [17] Folinais D, Thelen P, Delin C, Radier C, Catonne Y, Lazennec JY. Measuring femoral and rotational alignment: EOS system versus computed tomography. Orthop Traumatol Surg Res Sep. 2013;99(5):509–16. [18] Maree G, Irving B, Hering E. Paediatric dose measurement in a full-body digital radiography unit. Pediatr Radiol Oct. 2007;37(10):990–7. [19] Chimhundu C, Smit J, Sivarasu S, Douglas T. Inter-landmark measurements from lodox statscan images. J Med Devices 2013;11. [20] Chung CY, Lee KM, Park MS, Lee SH, Choi IH, Cho T-J. Validity and reliability of measuring femoral anteversion and neck-shaft angle in patients with cerebral palsy. J Bone Jt Surg May 2010;92(5):1195–205. [21] Zalawadia DA. Study of femoral neck anteversion of adult dry femora in Gujarat region. Natl J Integr Res Med 2010;1(3):7–11. [22] Kovacs L, Zimmermann A, Brockmann G, Baurecht H, Schwenzer-Zimmerer K, Papadopulos NA, et al. Accuracy and precision of the three-dimensional assessment of the facial surface using a 3-D laser scanner. IEEE Trans Med Imaging Jun. 2006;25(6):742–54. [23] Yoo SK, Ge W, Collison F, Rubinstein JT, Vannier MW, Kim HJ, et al. Three-dimensional localization of cochlear implant electrodes using epipolar stereophotogrammetry. IEEE Trans Biomed Eng May 2004;51(5):838–46. [24] Zhang JH, Shi XL, Wang YY, Lv L, Wu J, Zhang YF. Biplanar radiographic 3D reconstruction of vertebrae using non-stereo corresponding points based on epipolare geometry. In: Proceedings of the 6th World Congress of Biomechanics (WCB 2010). August 1-6, 2010 Singapore, 31. Berlin Heidelberg: Springer; 2010. p. 1266–9. [25] Gwet KL. Intrarater reliability. In: Wiley StatsRef: Statistics Reference Online. John Wiley & Sons, Ltd; 2014. [26] Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 1979;86(2):420–8. [27] MedCalc MS. Software. Ostend, Belgium: MedCalc for Windows; 1993. [28] Koerner JD, Patel NM, Yoon RS, Sirkin MS, Reilly MC, Liporace FA. Femoral version of the general population: does ‘normal’ vary by gender or ethnicity? J Orthop Trauma Jun. 2013;27(6):308–11. [29] Manaster A, Crim G, Miller P, Roberts R, Sanders. Diagnostic Imaging and Surgical Anatomy Musculoskeletal. 1st ed. AMYRSIS; 2007.