Gait & Posture 24 (2006) 475481 www.elsevier.

com/locate/gaitpost

Three-dimensional kinematics and kinetics of total knee arthroplasty during level walking using single plane video-uoroscopy and force plates: A pilot study
Monika Silvia Zihlmann a,*, Hans Gerber a, Alex Stacoff a, bor Sze kely b, Edgar Stu ssi a Kathrin Burckhardt b, Ga
a b

rich, Switzerland Laboratory for Biomechanics, Swiss Federal Institute of Technology, Zu rich, Switzerland Computer Vision Laboratory, Swiss Federal Institute of Technology, Zu

Received 14 June 2005; received in revised form 10 October 2005; accepted 9 December 2005

Abstract The goal of the study was to simultaneously obtain accurate kinematic and kinetic data from a total knee arthroplasty (TKA) during level walking, by coupling force plate data with the kinematics of TKA measured by a movable video-uoroscopic system. Kinematic and kinetic information of a TKA is crucial for the improvement of implant designs and for the increased longevity of the implant components. Instrumented gait analysis, with skin mounted marker tracking and force plates, is a well-established method for the acquisition of kinematic and kinetic data of TKA in vivo and for non-invasive estimation of joint function. However, resultant moments at the knee joint are inaccurate with this method, due to skin movement artifacts. Video-uoroscopy reduces these inaccuracies by means of the direct tracking of the implant components with X-ray. However this measuring technique carries disadvantages: it provides only kinematic data, and the image intensier covers a limited eld of view. This paper presents a newly developed measuring technique, which enables a more accurate resultant moments calculation for level walking than could be achieved by conventional instrumented gait analysis. # 2006 Elsevier B.V. All rights reserved.
Keywords: Movable video-uoroscopy; Level walking; Kinematics and kinetics; Total knee arthroplasty; Force plate data

1. Introduction Accurate in vivo kinematic and kinetic analysis of total knee arthroplasty is important to understand the complexity of knee joint mechanics after surgery. This knowledge is crucial for the prevention of higher stress and strain on the ligament structure as well as on the implant components. Better understanding of knee joint mechanics could lead to better surgical strategies, improved implant design, and increased longevity of the implant components. These factors improve patient satisfaction and eventually reduce health care costs. As the functionality determines the load and motion of the joint, there is a relationship between the longevity of the
nggerberg, 8092 Zu rich, * Corresponding author at: HCI E451, ETH Ho Switzerland. Tel.: +41 1 633 61 32; fax: +41 1 633 11 24. E-mail address: mzihlmann@ethz.ch (M.S. Zihlmann). 0966-6362/$ see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.gaitpost.2005.12.012

implant components and the functionality of the TKA [1]. Thus reliable subject kinematic and kinetic data are required. Instrumented gait analysis is a well-established method for gathering kinematic and kinetic data of the knee joint in vivo and for non-invasively estimating the functionality of joints [2,3]. Optical markers are placed on the subjects extremities in order to gain kinematic information from the segments of interest, and force plates xed on the oor simultaneously measure ground reaction forces. The main problem with this approach is the sizeable error caused by skin movement artefacts [4,5]. Point cluster methods were used to improve movement analysis [6]. However, Stagni et al. [7] found a standard deviation of skin marker trajectory of thigh and shank marker cluster of 20 markers each of 31, and 21 mm, respectively. They found root mean square errors in internal and external rotations and

476

M.S. Zihlmann et al. / Gait & Posture 24 (2006) 475481

ab/adduction of 192% and 117% relative to the corresponding range. Single plane video-uoroscopy is a method, which enables the reconstruction of the three-dimensional position and orientation (pose) of the implant components more accurately by avoiding errors due to skin and muscle movements [4]. A registration algorithm estimates the pose of the implant components from the single plane projection view of the uoroscopic image series [810] within an accuracy of 1.5 mm. Standard video-uoroscopy has the disadvantage of the small eld of view of the image intensier, which is usually in a xed position with a eld of view of 320 mm, making it impossible to obtain kinematic data from the knee during level walking. Zihlmann et al. [11] proposed a movable uoroscopic system enabling the tracking of the knee joint during level walking, therefore collecting kinematic data of the implant components with video-uoroscopy during several gait cycles. Video-uoroscopy is limited to kinematic data only, so that the load at the knee joint cannot be calculated. Some studies on knee joint kinematics and kinetics were performed using force plates and video-uoroscopy [12 14]. However, this data was not obtained simultaneously. Simultaneous kinematic and kinetic data acquisition is important in order to have reliable information about the functionality of a knee joint. To our knowledge, no previous studies have presented simultaneous kinematic and kinetic data during level walking as accurate as shown in this work. The goal of this pilot study was to attain kinematic and kinetic data during level walking more accurately than could be achieved by conventional instrumented gait analysis. This was implemented by coupling movable video-uoroscopy with force plate data.

KISTLERTM (Type 9281B, KISTLER AG, Switzerland) force plates, which were xed on the oor. The global coordinate systems origin was set in the centre of the third force plate, the y-axis was set in gait direction, the z-axis vertical to the oor, and the x-axis medio/lateral. 2.1.1. KISTLER force plate The force plates used for this study were ve 400 mm 600 mm multicomponent force plates based on piezo-electric sensors. All force plates were entirely separated mechanically from the surrounding oor thus avoiding an interaction with the uoroscopic system (Fig. 1). The sampling frequency of the force plates was 1000 Hz. The global coordinate systems origin was set at the centre of the third force plate. 2.1.2. Optical tracking system VICON The eight cameras of the optical tracking system VICON (V612 MX40) were calibrated to a space of 2000 mm 4000 mm 2000 mm with the coordinate systems origin at the centre of the third force plate. All series were measured with a sampling frequency of 100 Hz. 2.1.3. Movable video-uoroscopy The C-arm of the uoroscopic unit, together with the Xray source and the image intensier, is mounted on the unit mover. The system accelerates and decelerates thereby maintaining the knee joint within the eld of view of the uoroscope [15]. 2.1.4. Distortion correction of the uoroscopic images The distortion in the uoroscopic images was corrected by a calibration grid, which was xed on the image intensier and lmed before each measurement cycle. The distortion of the images was corrected by determining the

2. Materials and methods 2.1. Experimental setup The subject had to perform several consecutive strides on the laboratory walkway, a distance of approximately 10 m. During this task the subjects kinematic data were gathered with a movable video-uoroscopic system (BV Pulsera, Philips Medical Systems, Switzerland), by a pulsed mode of 25 Hz and 8 ms shutter time. The C-arm of the uoroscope was xed on a motor driven trolley, (thereafter the unit mover) enabling the tracking of the knee joint movement by keeping the implant components within the eld of view. Three reective markers (VICONTM, Oxford Metrics Inc., Oxford, UK) were placed on the unit mover and the C-arm in order to record the trolleys movement with optical motion tracking in the laboratorys coordinate system (subsequently called global coordinate system). The kinetics were measured by ve

Fig. 1. The force plate is completely decoupled from the surrounding oor. The gure illustrate the force plate and its sensors mounted on a separate support of the building. The gait direction is perpendicular to the image plane.

M.S. Zihlmann et al. / Gait & Posture 24 (2006) 475481

477

bilinear transformation between the known and measured coordinates of control points on the calibration grid. Due to the rigid connection of the X-ray source with the image intensier, the calibration was performed statically and was used over all uoroscopic images. The principal point lies at the image centre, the optical focus length is 1700.3 mm given by the manufacturer. 2.1.5. Three-dimensional reconstruction of the implants pose A three-dimensional analysis of each uoroscopic image was achieved by tting a synthetic X-ray image of the implant component (CAD model) to the original X-ray image [16]. After a rough initial estimation of its pose, the implants exact six degrees of freedom were found by an iterative minimisation of the difference between the synthetic and the original image. The estimated parameters were the position vector t of the implant, parallel (tx and ty) and perpendicular (tz) to the image plane, and the orientation, dened by the angles of out-of-plane rotation xrot and yrot and the angle zrot of the rotation in the image plane. The rotation angles xrot, yrot, and zrot are inserted in the standard denition of the rotation matrices, Rxrot , Ryrot , and Rzrot . The orientation of the implant components is calculated by multiplying the rotation matrices in the order Rzrot , Rxrot , Ryrot . The matrix Rfem is the orientation of the femoral component, Rtib the orientation of the tibial component. A third order Butterworth low pass, zero-phase forward and reverse lter was applied in order to slightly smooth the pose reconstructions. A normalised cut-off frequency of 0.1 was used. 2.2. Subject The subject (sex: male; height: 1.8 m; weight: 88 kg) for this pilot study had a total knee replacement to treat knee osteoarthritis (balanSysTM xed bearing, Mathys Ltd. Bettlach, Switzerland), 1 year after surgery, without any complications. 2.3. Coupling the measurement systems 2.3.1. Transformation of the uoroscopic image focus into the laboratorys global coordinate system The reconstruction of the implants pose was performed relative to the focus of the uoroscopic image. With the use of force plates, a transformation of the implants six degrees of freedom into the global coordinate system was required. In order to perform this transformation, the calibration grid was xed in a frame, which was screwed onto one force plate (Fig. 2) in a well-dened position relative to the force plates origin. A program written in MATLAB (Matlab 6.5.0, The MathWorks Inc., Natick, Massachusetts) calculated the focus location in the global coordinate system by comparing

Fig. 2. The calibration grid was xed in a frame on a dened position on the force plate. Three reective markers are xed on image intensier, C-arm, and X-ray source.

the location of the grids centre and the distance of the grid lines of the uoroscopic image with the measured true location and size. VICON tracked the markers during all the events thus enabling the estimation of the focus location over all image frames. The focus moves in the xy-plane only (xglobal, yglobal) because the vertical distance of the C-arm was adjusted to the length of the subjects lower leg and xed at this height. The focus in the global coordinate system was calculated by a linear transformation (Eq. (1)) into the xy-plane, where M (Eq. (2)) is the transformation matrix containing two unit vectors, e1 (marker 1 to marker 2) and e2 (marker 1 to marker 3), in its columns, c is the vector from focus to marker 1, and fglobal is the time-dependent vector of the focus location over all measured frames in the global coordinate system f global M1 c M 2 3 e1x e2x 0 M 4 e1y e2y 0 5 0 0 1 (1) (2)

Eq. (3) is the transformation from the focus coordinate system into the global coordinate system, where R (Rfem or Rtib) is the rotation matrix built by the transformations shown in Eq. (1). tfocus = (tx, ty, tz) is the implants position vector in the focus coordinate system, and tglobal = (tx, ty, tz)

478

M.S. Zihlmann et al. / Gait & Posture 24 (2006) 475481

the position vector in the global coordinate system tglobal R tfocus f global 2 3 0 1 0 R 40 0 15 1 0 0 (3) (4)

2.3.2. Time synchronisation A reective marker was placed at the end of a stick with a length of 500 mm. This stick was introduced into the uoroscopic image with a short downwards movement at the beginning of each movement task after starting the X-ray beam and the VICON system. This vertical movement was visible both in the VICONs vertical coordinates and in the corresponding uoroscopic image, enabling a time synchronisation of the measuring systems. 2.4. Finite axis of rotation The nite axis of rotation and the corresponding rotation angle of each time step, Dt = 40 ms, between the femoral and tibial implant component were calculated by extracting vector, ux,y,z and angle, b of the relative rotation matrix Rrel (Eq. (5)). The relationship between the rotation matrix and the screw vector and angle is given in Eq. (6) (Equation of Rodrigues). There u  u is the diadic product, I the identity matrix, and is the skew symmetrical vector product of u. The components of R as function of xrot, yrot, and zrot are compared with the entries of R as function of u and b. Rrel Rfem R0tib (5)

Fig. 3. Knee joint model in the sagittal plane to calculate resultant moments in the joint by reducing the ground reaction force into the nite centre of rotation COR.

considering the condition listed above can be estimated by using differential calculus (Eq. (8)) 3 2 3 3 2 0 0 Mres 4 0 5 4 CORy POFAy 5 4 Fy 5 0 CORz Fz 2

(7)

(6) DMres Fz DPOFAy 2 Fz DCORy 2 2.5. Error of resultant moments estimation An error analysis was performed to estimate the errors affecting resultant forces and moments calculation in the knee joint. Resultant forces were calculated by using the measured ground reaction forces of the force plate. Resultant moments were calculated by reducing the moments into the nite centre of rotation, COR, of the knee joint (see Fig. 3). Thus, the resultant moments estimation is a function of the location vector of COR, the point of force application, POFA, and the ground reaction force components, F x, F y and F z. Assuming quasi static behaviour and reducing the movement into the sagittal plane, an uncertainty estimation of the resultant moments of the knee joint was performed, i.e. the error in resultant moments remains a function of F z, F y, CORy, CORz and POFAy (Eq. (7)), and the error listed in Table 1. The uncertainty of the resultant moments of the knee joint Fy DCORz 2 CORy POFAy DFz 2 CORz DFy 2 1=2 (8)

Table 1 RMSE of the components inuencing resultant moments estimation of the knee joint Inuence factor Three-dimensional reconstruction Focus location, global Axis of rotation estimation global Fz Fy POFA calculation KISTLER force plate KISTLER force plate Method of estimation In-plane translation, CORy, CORz Spatial rotation angle, b VICON marker tracking RMSE 0.17 mm 1.578 1 mm 1.45 mm 4 N [17] 5 N [17] 3 mm [17]

M.S. Zihlmann et al. / Gait & Posture 24 (2006) 475481

479

The errors affecting the resultant moments calculation are listed below. 2.5.1. Error of the force vector The error in ground reaction forces components while using KISTLER force plates was found to be about 4 N for F z and F y [17]. The point of force application was found to have an error of 3 mm in a diameter of about 100 mm around the centre of the force plate. 2.5.2. Error of the uoroscopic focus location in the global coordinate system Four factors affect the uncertainty of the location estimation of the uoroscopes focus. These are: (i) The calibration grids well-dened xation to the force plate. (ii) The spatial coordinate estimation of the three markers of the uoroscopic system. (iii) The uncertainty of the estimation of the focus in the uoroscopic image. (iv) The stiffness of the C-arm. The calibration grids exact position could be measured with an uncertainty of 0.1 mm. The spatial coordinate estimation of the three markers of the uoroscopic system was measured to have an absolute uncertainty of 1 mm. The uncertainty of the focus location in the uoroscopic image was estimated to be 0.1 mm. To evaluate the stiffness of the C-arm while the system is moving, metal discs with a diameter of about 5 mm were bonded on the bottom of a rigid plastic box, which was xed on the image intensier, so that the discs were held in a distance of 300 mm parallel to the image intensier. Fluoroscopic images were taken while the unit mover was accelerating and decelerating. The position of the discs were compared in every taken image-by-image subtraction (gray scale). No difference in the images could be found. Hence, we assume that the C-arm is stiff enough for our application. In conclusion, the uncertainty in uoroscopic focus location in the global coordinate system is 1 mm. 2.5.3. Error of implants pose reconstruction A static and dynamic in vitro validation study was performed to estimate the uncertainty of the implants pose reconstruction. The three-implant components were xed, allowing no relative movements between the implant components. The exact position and orientation of the tibial component relative to the femoral component was measured by a three-dimensional optical digitiser (Breuckmann OPTO Top HE-200, Germany), which measured surface points within an accuracy of 0.01 mm. The CAD geometries of the implant components were then tted in the measured surface point cloud (Raindrop Geomagic Studio 6.0), while the relative position and orientation of the components coordinate systems origin were calculated. Four images of

the implant components were taken from different point of view. One series was taken by moving the implant components with a velocity of 0.3 m/s relative to the image focus. The matching algorithm calculated the implants pose in all frames relative to the image focus. The RMSE of the pose (distance t, and spatial rotation angle b) of the femoral part relative to the tibial part were calculated. The dynamic validation study found an RMSE of the distance of 3.25 mm and a RMSE of the rotation angle of about 1.578. A repeatability study was performed while using the matching algorithm 15 times on the same images. The ratio of the standard deviations of the in-plane to out-of-plane translation was found to be 16. Thus, a rough estimate of the accuracy of the in-plane translation can be 3.25 mm/ 16 = 0.20 mm.

3. Results 3.1. Error of resultant moments estimation The preceding results show that the error of the direction of the axis of rotation can be neglected. Thus, the error of the inplane translation relative to the focus is a factor, which has to be taken account of when estimating the nite axis of rotation. The in-plane translation of the uoroscopic system equals the yz-plane of the global coordinate system, where the y-axis was the subjects gait direction. The uncertainty in estimating the focus in the global coordinate system has to be taken into account when calculating the centre of rotation. These two errors are independent and therefore their variance can be p p added, equals 0:2 mm 1 mm 1:45 mm. The axis of rotation was mainly parallel to the x-axis of the global coordinate system, since the largest rotation between the implant components during level walking occurs in the yz-plane (knee joint exion). Therefore, the worst case error of the nite axis of rotation location in the global coordinate system is the sum of the error of the inplane reconstruction and the error of the focus location in the global coordinate system. The sources of errors affecting resultant moments estimation are summarised in Table 1. Assuming F y = 50 N, F z = 800 N, CORy POFAy = 50 mm, and CORz = 450 mm, taking the values of the uncertainties as listed in Table 1 and using the formula given in Eq. (8), the error of the resultant moment DMres would be 3.3 N m. 3.2. Resultant moments at the knee joint The results of coupling force plate data with the reconstructed video-uoroscopic sequences are illustrated in Fig. 4. This gure shows the nite axis of rotation between the femoral and the tibial implant component during level ssi et al. [18]). walking in a four dimensional plot (Stu Ground reaction force vectors, where the TKA leg is in

480

M.S. Zihlmann et al. / Gait & Posture 24 (2006) 475481

4. Discussion Conventional instrumented gait analysis with skin mounted markers has the disadvantage of measuring artefacts due to skin movement relative to the underlying bones. Video-uoroscopy is a well-established method of more accurately measuring knee joint kinematics by avoiding the use of skin markers. The small eld of view of the image intensier and the ability to only gain kinematic data are the two main disadvantages of this system. The goal of the present study was to acquire kinematic and kinetic data during level walking more accurately than could be achieved by conventional instrumented gait analysis. The presented measuring technique enables kinematic and kinetic data acquisition of TKA during the stance phase of level walking more accurately than could be achieved by a tracking system with the use of skin mounted markers. A sensitivity analysis of each estimation and calculation step showed that the uncertainty in calculating the axis of rotation of the knee joint was a translational error in space of 3.25 mm, and an error of the rotation angle b of 1.578 which is comparable with the accuracy achieved by other authors [9,10]. Kinematic and kinetic information, with the accuracy mentioned above, enable a more reliable estimation of the resultant forces and moments, by reducing ground reaction forces into the nite axis of rotation of each time step, than conventional instrumented gait analysis. Stagni et al. [7] found a RMS difference of 17.2% and 23.4%, respectively, of the motion range during a sit down tasks. The standard deviations of internal and external rotations and ab/ adduction were much larger. These large errors compromise the use of these variables in the clinical interpretation of gait analysis. In particular, when calculating resultant moments at the knee joint, there are errors due to skin movement. Assuming an uncertainty of about 24 N m due to the error in COR estimation of about 30 mm, this would lead to an uncertainty of about 50% of the resultant moment. This magnitude of the uncertainty in the resultant moment due to skin movement artefacts was also found by Barth [17]. Barth made his error calculation by comparing computed resultant moments and measured ones via a strain gage instrumented lower leg prosthesis. This study shows that the error of the resultant forces at the knee joint is 3.3 N using this measuring system. This is about seven times less than conventional instrumented gait analysis with skin mounted markers would achieve. The movable video-uoroscopy system coupled with force plates allows the subject to walk freely without any restrictions, thus obtaining the subjects usual gait pattern while obtaining kinematic as well as kinetic data.

Fig. 4. Finite axis of rotation between the implant components and force vectors of one force plate. Dt = 40 ms.

contact with the force plate, are plotted in the same gure. The resultant moments at the knee joint during the stance phase of gait were calculated by reducing the moments in the nite axis of rotation in the sagittal plane (Fig. 5) by cutting the nite axes of rotation with the yz-plane of the global coordinate system through the systems origin (x = 0). The resultant moment at the knee joint reaches 50 N m. Taking the results of DMres of 3.3 N m (see Section 3.1) related to the uncertainty in pose reconstruction and the uncertainty in focus location in the global coordinate system, the ratio of DMres/Mres is 6.6%. Assuming an uncertainty in COR estimation of about 30 mm, which is comparable with the error due to skin movement artefact, DMres would be about 24 N m.

5. Conclusion
Fig. 5. Resultant moments at the knee joint during the stance phase of level walking. Unltered.

This pilot study showed that it is feasible to measure ground reaction forces and kinematic data simultaneously

M.S. Zihlmann et al. / Gait & Posture 24 (2006) 475481

481

with a moving uoroscopic system. With the avoidance of skin movement artefacts more accurate in vivo kinematic information of TKA during the stance phase of level walking can be obtained compared to instrumented gait analysis. Therefore, the resultant moments in the knee joint can be estimated up to seven times more accurately than conventional instrumented gait analysis. This work presents an improvement in measurement accuracy for the analysis of the stance phase of gait, thereby providing a basis for better inverse dynamic calculation of the load.

Acknowledgements Dr. H.C. Robert Mathys Foundation, Bettlach, Switzerland, Mathys Ltd. Bettlach, Switzerland, Philips Medical Systems, Switzerland, Spital Bern Ziegler, Berne Switzerland.

References
[1] Andriacchi TP. Biomechanics and gait analysis in total knee replacement. Orthop Rev 1988;17(5):4703. [2] Ishii Y, Terajima K, Koga Y, Takahashi HE, Bechtold JE, Gustilo RB. Gait analysis after total knee arthroplasty. Comparison of posterior cruciate retention and substitution. J Orthop Sci 1998;3(6):3107. [3] Su FC, Lai KA, Hong WH. Rising from chair after total knee arthroplasty. Clin Biomech (Bristol Avon) 1998;13(3):17681. [4] Sati M, de Guise JA, Larouche S, Drouin G. Quantitative assessment of skin-bone movement at the knee. Knee 1996;3:12138. [5] Holden JP, Orsini JA, Siegel KL, Kepple TM, Gerber LH, Stanhope SJ. Surface movement errors in shank kinematics and knee kinetics during gait. Gait Posture 1997;5(3):21727. [6] Andriacchi TP, Dyrby CO, Johnson TS. The use of functional analysis in evaluating knee kinematics. Clin Orthop Relat Res 2003;410:4453.

[7] Stagni R, Fantozzi S, Cappello A, Leardini A. Quantication of soft tissue artefact in motion analysis by combining 3D uoroscopy and stereophotogram-metry: a study on two subjects. Clin Biomech (Bristol Avon) 2005;20(3):3209. [8] Banks SA, Hodge WA. Accurate measurement of three-dimensional knee replacement kinematics using single-plane uoroscopy. IEEE Trans Biomed Eng 1996;43(6):63849. [9] Hoff WA, Komistek RD, Dennis DA, Gabriel SM, Walker SA. Threedimensional determination of femoral-tibial contact positions under in vivo conditions using uoroscopy. Clin Biomech (Bristol Avon) 1998;13(7):45572. [10] Zuf S, Leardini A, Catani F, Fantozzi S, Cappello A. A model-based method for the reconstruction of total knee replacement kinematics. IEEE Trans Med Imaging 1999;18(10):98191. ssi E. Three[11] Zihlmann MS, Gerber H, Burckhardt KV, Stacoff A, Stu dimensional in vivo kinematics of articial knee joints during level walking using a moving video-uoroscope. In: Proceedings of the 13th conference of the Canadian society for biomechanics; 2004. [12] Fregly BJ, Sawyer WG, Harman MK, Banks SA. Computational wear prediction of a total knee replacement from in vivo kinematics. J Biomech 2005;38:30514. [13] Fantozzi S, Benedetti M, Leardini A, Banks S, Cappello A, Assirelli D, et al. Fluoroscopic and gait analysis of the functional performance in stair ascent of two total knee replacement designs. Gait Posture 2002;15:110. [14] Deluzio KJ, Banks S, Costigan PA, Ladouceur D, OFlynn H, Wyss UP, et al. Kinematics and kinetics of total knee replacement patients; uoroscopy, motion analysis, and force plate data. In: Proceedings of the 11th congress of the Canadian society for biomechanics; 2000. ssi E. [15] Zihlmann MS, Stacoff A, Romero J, Kramers-de Quervain I, Stu Biomechanical background and clinical observations of rotational malaignment in TKAliterature review and consequences. Clin Biomech 2005;20(7):6618. [16] Burckhardt KV. Locating implants with respect to the bone in diagnostic X-ray images of the pelvis, PhD Thesis, Swiss Federal Institute of Technology, Zurich, 2002. [17] Barth W. Signicance of knee joint loads calculated using inverse dynamics, PhD Thesis, Swiss Federal Institute of Technology, Zurich, 1999. ssi E, Aebersold P, Debrunner HU. Darstellung des Ganges durch [18] Stu vierdi-mensionale Vektoren. Biomed Tech (Berl) 1978;23.