The use of Roentgen stereophotogrammetry to study micromotion of ...
-
Upload
nguyendien -
Category
Documents
-
view
220 -
download
2
Transcript of The use of Roentgen stereophotogrammetry to study micromotion of ...
The use of Roentgen stereophotogrammetry to study
micromotion of orthopaedic implants
Edward R. Valstar a,*, Rob G.H.H. Nelissen a, Johan H.C. Reiber b, Piet M. Rozing a
aDepartment of Orthopaedics, Leiden University Medical Center, P.O. Box 9600, 2300 RC Leiden, The NetherlandsbDivision of Image Processing, Department of Radiology, Leiden University Medical Center, P.O. Box 9600, 2300 RC Leiden, The Netherlands
Received 31 October 2001; accepted 3 May 2002
Abstract
Roentgen stereophotogrammetry is the most accurate Roentgen technique for three-dimensional assessment of micromotion
of orthopaedic implants. The reported accuracy of Roentgen Stereophotogrammetric Analysis (RSA) ranges between 0.05 and
0.5 mm for translations and between 0.15j and 1.15j for rotations. Because of the high accuracy of RSA, small patient groups
are in general sufficient to study the effect on prosthetic fixation due to changes in implant design, addition of coatings, or new
bone cements. By assessing micromotion of a prosthesis in a short-term (i.e. 2 years) clinical RSA study, a prediction can be
made on the chance of long-term (i.e. 10 years) loosening of the prosthesis. Therefore, RSA is an important measurement tool to
screen new developments in prosthetic design, and to prevent large groups of patients from being exposed to potentially inferior
designs. In this article, the basics of the RSA technique are explained, and the importance of clinical RSA studies is illustrated
with two examples of clinical RSA studies which RSA delivered very valuable information. Thereafter, two recent
developments in RSA that have been implemented at Leiden University Medical Center are presented: digital automated
measurements in RSA radiographs and model-based RSA.
D 2002 Elsevier Science B.V. All rights reserved.
Keywords: Roentgen stereophotogrammetry; model-based RSA; motion; orthopaedic implants; X-ray; medical imaging
1. Introduction
Artificial joint replacement is a common treatment
for joints that have been affected by trauma, artrosis,
or rheumatoid arthritis. Worldwide, 1 million total hip
replacements and 500,000 total knee replacements are
performed every year. The maximum lifespan of a
prosthesis will be about 15 to 20 years.
Unfortunately, some prostheses have to be revised
before the expected maximum lifespan. This may be
necessary when the prosthesis has worn out, or when
the prosthesis has loosened with respect to the sur-
rounding bone.
In general, loosening starts with progressive micro-
motion, in the range of 0.2–1 mm, of the prosthesis
relative to the surrounding bone. Once it has started, it
is a continuous process that will destroy bone, and as
a result, the prosthesis will start to migrate over larger
distances. At present, this process of prosthetic loos-
ening and bone destruction can only be stopped by
revision of the prosthesis.
0924-2716/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.
PII: S0924-2716 (02 )00064 -3
* Corresponding author. Tel.: +31-71-526-2975; fax: +31-71-
526-6743.
E-mail address: [email protected] (E.R. Valstar).
www.elsevier.com/locate/isprsjprs
ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389
A revision operation is much more demanding for
the patient than the implantation of a primary pros-
thesis, and the results are inferior to those of primary
prostheses: many patients suffer from pain, a
decreased range of motion, and the loosening rate of
revision prostheses is much higher than for primary
prostheses.
Therefore, it is of utmost importance to develop
prostheses that have a long lifespan. Since loosening
of a prosthesis starts with micromotion, knowledge
about micromotion is important, as it may predict
future loosening (Karrholm et al., 1994; Ryd et al.,
1995). By studying micromotion, one can obtain
insight in the loosening process of prostheses, which
can be used to improve prostheses.
In clinical practice, loosening of prostheses is
assessed indirectly at successive radiographs by meas-
uring radiolucent—dark—lines around the prosthesis
and assessing positional differences of the prosthesis
relative to the bone. Radiolucent lines indicate the
presence of a fibrous layer around the prosthesis that
is always present when a prosthesis is loose. In Fig. 1
some basic measurements in conventional radiographs
for assessment of prosthesis position and orientation
are indicated.
These measurements are not very accurate: radio-
lucency may occur in areas that are overprojected by
the metal of the implant and thus might not be
observed: as a result, the amount of radiolucency may
be underestimated (Nelissen, 1995; Reading et al.,
1999). Migration of the prosthesis is assessed by
measuring changes in the relative positions of its
prosthetic landmarks and the bony landmarks over
time. However, bony landmarks are not sufficiently
distinctive and are therefore difficult to measure in a
reproducible manner. For these reasons, measurements
on plain radiographs are not accurate. In total hip
arthroplasty, for example, migration measurements
may have an accuracy between 5 and 12 mm (95%
Fig. 1. (a) A conventional radiograph of a pelvis with two hip
prostheses. These hip prostheses consist of a femoral stem and a cup.
The circles indicate the head of the femoral stem. The cup on the right
hand is still in place. The cup on the left hand side, however, has
penetrated the pelvis over a distance d, a distance of approximately 20
mm. The angle of inclination of the cup is indicated by a. This angleshould ideally be 45j; in this patient, the angle has decreased to 30j.This cup will have to be revised. (b) A conventional radiograph of the
femur of the same patient as in (a). The femoral stem of the prosthesis
has subsided over distance d, approximately 18 mm. The two
approximate vertical lines in the image indicate the central lines of the
femur and the prosthesis. Ideally, the angle between these two lines, a,should be 0j. In this case, the angle is 20j. This prosthesis component
will also have to be revised.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 377
confidence interval), depending upon the choice of
bony landmarks (Malchau et al., 1995).
Several attempts have been made to increase the
accuracy of measurements in plain radiographs.
Improvements have been made by the standardisation
of the position of the patient, by the use of additional
landmarks, and by the use of software that performs
the measurements in a reproducible and objective
manner (Hardinge et al., 1991). A measurement
technique that combines these three improvements is
the Einzel Bild Roentgen Analyse that has an accuracy
of 1 mm (Krismer et al., 1995). However, to measure
sub-millimetre micromotion of implants during the
first half-year after implantation or to detect differ-
ences between migration patterns of small groups of
implants, this accuracy will not be sufficient.
Therefore, in 1974, Selvik (1989) developed a
highly accurate technique for the assessment of
three-dimensional migration of prostheses, Roentgen
Stereophotogrammetric Analysis (RSA). The reported
accuracy of RSA ranges between 0.05 and 0.5 mm for
translation and between 0.15j and 1.15j for rotations
(95% confidence interval; Karrholm, 1989).
In this article, the basics of the RSA technique will
be explained, and the importance of clinical RSA
studies is illustrated with two examples of clinical
RSA studies in which RSA delivered very valuable
information. Thereafter, two recent developments in
RSA that have been implemented at Leiden Univer-
sity Medical Center will be introduced: one concern-
ing digital automated measurements in RSA
radiographs and one concerning model-based RSA.
Finally, conclusions are drawn and some recommen-
dations for future research are given.
2. Basics
RSA is the most accurate Roentgen technique for
the assessment of micromotion of orthopaedic
implants. To each that accuracy, however, several
steps have to be taken that make the technique rather
complicated in a clinical setting.
2.1. Bone and prosthesis markers
To accurately measure migration in RSA radio-
graphs, bony landmarks are not sufficiently distinc-
tive. In order to obtain well-defined measurement
points, tantalum beads are inserted into the bone with
a special insertion instrument. These beads have a
diameter of 0.5, 0.8, or 1 mm. Due to their small size
and spherical shape, their projection will not be
influenced by changes in patient position or Roentgen
focus position. Therefore, the position of these
markers can be measured with great accuracy. Since
most prostheses do not have landmarks that can be
measured in a reproducible manner either, they have
to be marked with at least three non-collinear markers
(Fig. 2).
2.2. Roentgen set-up
In RSA, two synchronised Roentgen tubes are used
to obtain two projections of an area of interest of a
patient. Using the information in these two projec-
tions, it is possible to reconstruct the three-dimen-
sional position of markers in that area. The two
Roentgen tubes are positioned at approximately 1.60
m above the Roentgen cassette at a 20j-angle to the
Fig. 2. The Interax total knee prosthesis consists of two components:
one component that is inserted in the femur and one that is inserted in
the tibia. Three 2-mm markers have been attached to the tibial
component, and seven markers have been inserted in the tibia.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389378
vertical (Fig. 3a and b). A calibration box is used to
calibrate the Roentgen set-up. The calibration box that
we mainly use in our studies has two planes that hold
tantalum markers of which the three-dimensional
position is accurately known. The markers in the
plane close to the radiographic film are denoted
fiducial markers, the markers in the plane distant from
the radiographic film are denoted control markers.
The fiducial markers define the three-dimensional
fiducial coordinate system and the control markers
are used to assess the position of the Roentgen foci.
2.3. Measurement of two dimensional marker projec-
tions
After the RSA radiograph has been taken (Fig. 4),
the coordinates of the bone and prosthesis markers
have to be measured accurately. In conventional RSA
research, this was done using a manually operated
measuring table with an accuracy of 0.02 mm. Manual
digitisation of marker coordinates is a rather tedious
task that might take up to 45 min per film. Therefore,
software has been developed that automates this task,
and drastically reduces analysis time. This digitally
automated approach to RSAwill be described in more
detail in Section 4.
2.4. Calibration
2.4.1. Transformation from radiographic coordinates
to fiducial coordinates
In order to be able to calculate the three-dimen-
sional positions of markers and landmarks on the
implant, the measured coordinates have to be trans-
formed to the lower plane of the calibration box
(fiducial plane). This is done with these two equa-
tions:
xfid;i ¼l1xrad;i þ l2yrad;i þ l3
l7xrad;i þ l8yrad;i þ 1i ¼ 1; . . . ; n; ð1Þ
yfid;i ¼l4xrad;i þ l5yrad;i þ l6
l7xrad;i þ l8yrad;i þ 1i ¼ 1; . . . ; n; ð2Þ
where, (xfid,i,yfid,i) are two-dimensional fiducial coor-
dinates of a point, (xrad,i,yrad,i) are two-dimensional
Fig. 3. (a) The RSA set-up. Two synchronised Roentgen tubes are positioned above a calibration box. (b) The joint of interest is positioned at the
intersection of both Roentgen bundles.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 379
radiographic coordinates of a point, and n is the
number of points. The l-parameters are assessed by
using the known positions of the fiducial markers on
the calibration box and their measured projections on
the radiograph. At least four non-linear fiducial
markers and their projection are needed to calculate
the l-parameters.
The parameters are first estimated in a linear
manner. Therefore, Eqs. (1) and (2) are rewritten, so
that they are linear for the l-parameters:
xfid ¼ l1xrad þ l2yrad þ l3 � l7xradxfid � l8yradxfid; ð3Þ
yfid ¼ l4xrad þ l5yrad þ l6 � l7xrad yfid � l8yrad yfid: ð4Þ
The solution of this problem may be found by
performing a QR-decomposition (Golub and Van
Loan, 1989).
This linear estimate is used as a start point for
a non-linear Newton–Gauss algorithm (Dennis
and Schnabel, 1983). The cost function that is used
is:
J ¼ ete; ð5Þ
where, e is an error vector that may be written as:
e ¼
l1xrad;1þl2yrad;1þl3l7xrad;1þl8yrad;1þ1
� xfid;1
l1xrad;2þl2yrad;2þl3l7xrad;2þl8yrad;2þ1
� xfid;2
]
l1xrad;nþl2yrad;nþl3l7xrad;nþl8yrad;nþ1
� xfid;n
l4xrad;1þl5yrad;1þl6l7xrad;1þl8yrad;1þ1
� yfid;1
l4xrad;2þl5yrad;2þl6l7xrad;2þl8yrad;2þ1
� yfid;2
]
l4xrad;nþl5yrad;nþl6l7xrad;nþl8yrad;nþ1
� yfid;n
266666666666666666666666664
377777777777777777777777775
: ð6Þ
Fig. 4. An RSA radiograph of a total knee prosthesis from two synchronised Roentgen tubes. Note the markers in the bone and the markers
attached to the prosthesis (see Fig. 2). Markers that are located outside the bone of the patient are calibration box markers that are used to define
a three-dimensional laboratory coordinate system, and to assess the Roentgen foci positions.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389380
The error vector e and the cost function J are a
function of the l-parameters. In order to determine the
search direction for the optimisation, the Jacobean has
to be determined:
Jac ¼ BJ
Be
Be
Bl: ð7Þ
The expression for the Newton–Gauss optimisation
is:
lnew ¼ lold � ðJactJacÞ�1ðJacteÞ; ð8Þ
where, lnew holds the l-parameters that are the result
of the current optimisation step and lold is the result
of the previous optimisation step. After the l-
parameters have been assessed, points that have
been measured in the radiograph are transformed
to the fiducial coordinate system by means of Eqs.
(1) and (2).
The quality of this transformation is expressed as
the distance between the fiducial markers and their
transformed projections.
2.4.2. Calculation of foci positions
The next calibration step is the assessment of
position of both Roentgen foci. For this assessment,
the markers in the upper plane of the calibration
box, the control markers, are used. Through these
markers (ci) and their transformed projections (ciV),projection lines are determined:
riðaiÞ ¼ ci þ aiðci � ciVÞ�l < ai < l i ¼ 1; . . . ; n; ð9Þ
where n indicates the number of markers that are
used. In the ideal situation, these lines will intersect
in one point. However, measurement errors occur
and the position of the focus f has to be determined
by solving the least squares problem (Soderkvist,
1990):
minf ;ai
N
ðc1V� c1Þ 0 . . . 0 I3
0 ðc2V� c2Þ ] I3
] O 0 ]
0 . . . 0 ðcnV� cnÞ I3
2666666664
3777777775
�
a1
a2
]
an
f
26666666666664
37777777777775
�
c1
c2
]
cn
2666666664
3777777775N; ð10Þ
that is solved by a QR-decomposition (Golub and
Van Loan, 1989).
2.4.3. Calculation of three-dimensional marker posi-
tions
The determination of the three-dimensional posi-
tion of the tantalum markers is similar to the deter-
mination of the focus position. The projections of the
marker in the two images are denoted by t1 and t2. The
position of the two Roentgen foci is denoted by f1 and
f2. The equations of the projection lines that connect
the transformed projections and the corresponding
foci are:
l1ðaÞ ¼ f1þ að f
1� t1Þ �l < a < l;
l2ðbÞ ¼ f1þ bð f
2� t2Þ �l < b < l: ð11Þ
The three-dimensional position of the marker is
the position where these two lines intersect. How-
ever, due to measurement errors, the lines will not
intersect but they will cross each other at a short
distance. The three-dimensional position of the
marker p is assumed to be at the middle of the
shortest line connecting the two projection lines.
The length of this shortest line is denoted crossing
linear error. The three-dimensional position of p is
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 381
the solution of the least squares problem (Soderkvist,
1990):
minp;a;b
N
ðt1 � f1Þ 0 I3
0 ðt2 � f2Þ I3
264
375
a
b
p
26666664
37777775�
f1
f2
264
375N;
ð12Þ
that is dissolved by a QR-decomposition (Golub
and Van Loan, 1989).
2.5. Motion of rigid bodies
After the three dimensional position of the markers
has been assessed, the relative motion of the prosthe-
sis with respect to the bone can be calculated. The
bone markers function as a reference rigid body
relative to which the motion of the second rigid body,
the prosthesis, is calculated. Since patients cannot be
positioned in exactly the same position and orientation
between follow-ups, the position and orientation of
this reference rigid body change between follow-ups.
In order to obtain the same position and orientation of
the bone, the bone markers in the first radiograph and
in the follow-up radiographs are matched onto each
other; thereafter, the relative motion of the prosthesis
with respect to the bone can be calculated. The results
from the motion calculations are a rotation matrix and
a translation vector.
2.5.1. Calculation of rotation matrix and translation
vector
Assume that we have n points in a rigid body and
let a1,. . .,an be the positions of these points on
instance one and let b1,. . .,bn be the positions on
instance two. In order to assess a rotation matrix M
and a translation vector d, the following equation has
to be solved:
minM ;d
Xni¼1
NMai þ d � biN2; ð13Þ
so that M is an orthogonal matrix.
This problem may be solved in several ways
(Spoor and Veldpaus, 1980; Veldpaus et al., 1988),
but the most elegant way to solve this problem has
been described by Soderkvist (1990). This method
that is based on singular value decomposition is
presented in this section. M is the orthogonal rotation
matrix and d is the translation vector. An expression
for d is Soderkvist, 1990:
d ¼ 1
n
Xni¼1
ðbi �MaiÞ ¼ b�Ma: ð14Þ
When this expression is substituted in expression (13),
the only unknown remains M:
minM
Xni¼1
NMðai � aÞ � ðbi � bÞN2: ð15Þ
When we define A=[a1�a,. . .,an�a] and B=[b1�b,. . .,bn�b], the problem may be written as:
minM
NMA� BN; ð16Þ
so that M is an orthogonal matrix.
The solution of the rotation matrix is:
M ¼ UV t; ð17Þ
in which
BAt ¼ UAV t ð18Þ
is the singular value decomposition. The solution of d
is found when M is substituted in Eq. (14).
2.5.2. Relative motion
In clinical RSA studies, one is often interested in
the motion of an implant with respect to the surround-
ing bone. This means that we are interested in the
relative motion between two rigid bodies. When these
rigid bodies are denoted A and B and the relative
motion is calculated between time t0 and t1, the
relative motion may be calculated as (Soderkvist,
1990):
At1cMAAt0 þ dAut and Bt1cMBBt0 þ dBu
t; ð19Þ
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389382
where, ut=[1,. . .,1]. The relative rotation may be
expressed as:
Mrel ¼ MtAMB: ð20Þ
After the origin o is positioned in the geometrical
centre of At0:
o ¼ 1
nA
XnAi¼1
At0;i; ð21Þ
the relative translation may be expressed as:
drel ¼ MtAðdb � dAÞ þ ðMrel � I3Þo; ð22Þ
where, I3 is a unity matrix. The translation vector drelthat has been calculated with Eq. (22) might be
difficult to understand for the clinician. Therefore,
the difference in position of the geometrical centre of
a rigid body at two points in time is often presented
instead of drel.
3. Clinical studies
Because of the high accuracy of RSA, small patient
groups are in general sufficient to study the effect on
prosthetic fixation due to changes in implant design,
addition of coatings to the prosthesis, or new bone
cements. RSA has been applied in many studies that
were primarily conducted in Sweden and in the other
Scandinavian countries. More than 3000 patients have
been included in several studies and more than 150
scientific papers have been published (overview in:
Karrholm, 1989; Ryd, 1992). Topics that have been
studied by RSA are: prosthetic fixation, joint stability
and joint kinematics, fracture stability, skeletal
growth, vertebral motions, and spinal fusion.
3.1. The Boneloc cement disaster
The importance of evaluating new developments in
small patient groups before a mass introduction on the
market can be illustrated by the introduction of Bone-
loc cement (Biomet, Warsaw, IN, USA) in 1991.
Cement is used to fixate a prosthesis in the bone.
Bone cement is a polymer that is made by mixing a
liquid monomer and a powder. During the polymer-
isation process that follows mixing, the polymer is
formed and heat is created: the advantage of Boneloc
cement was the lower polymerisation temperature, 43
jC instead of 80 jC in traditional cements. This lower
temperature was expected to reduce local cell death
and a better bone-cement interlock would be obtained,
thus improving the fixation of prostheses.
However, in clinical practice, a deterioration in
prosthetic fixation was observed: several clinics
reported loosening of implants after using Boneloc
cement. As a consequence, two clinical RSA studies
were started: a total knee study with 19 patients
(Nilsson and Dalen, 1998) and a total hip study with
11 patients (Thanner et al., 1995). These RSA studies
supported the clinical observations: implants fixated
with Boneloc migrated significantly more than
implants fixated with conventional cement. Therefore,
Boneloc cement leads to an increased risk for revision
caused by aseptic loosening.
Unfortunately, at that point in time, Boneloc had
been used in more than 1000 cases in Norway alone.
After a period of 4.5 years, the revision rate of the
prostheses was 14 times higher than for prostheses
fixed with conventional cement (Furnes et al., 1997).
This might have been prevented by a pre-marketing
clinical test of the fixation with an adequate RSA
study.
3.2. The effect of hydroxyapatite on fixation of knee
prostheses
There are two approaches for fixing prostheses in
bone. In the first approach, cement is used to form a
strong interdigitation between the bone and the pros-
thesis. In the second approach, the prosthesis is intro-
duced into a preformed cavity in the bone that exactly
matches the surface of the prosthesis; bone ingrowth
into the prosthetic surface will provide fixation. Bone
ingrowth can be stimulated with special surface coat-
ings that are sprayed onto the prosthetic surface. One of
these coatings is hydroxyapatite.
At Leiden University Medical Center, a prospec-
tive, randomised, double-blind study was executed to
evaluate three different means of fixing tibial (lower
leg bone) components to the bone in knee replacement
(Nelissen et al., 1998). The aim was to study the effect
of the addition of hydroxyapatite on the fixation of
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 383
uncemented knee prostheses. Eleven prostheses fixed
with cement, 10 hydroxyapatite-coated prostheses
fixed without cement, and 10 non-coated prostheses
fixed without cement were studied. RSA was used to
assess micromotion of the components during a 2-year
follow-up period.
With this small cohort of patients and within the
short follow-up period of 2 years, cemented tibial
components and hydroxyapatite-coated tibial com-
ponents fixed without cement were found to have far
less micromotion along the three orthogonal axes
as compared to non-coated tibial components fixed
without cement: At the 2-year follow-up evaluation,
the subsidence of the non-coated components was
�0.73F0.924 mm, the subsidence of the cemented
components was�0.05F0.109mm, and of the hydrox-
yapatite-coated components �0.06F0.169 mm. So,
after 2-year follow-up, the subsidence of the non-
coated components was significantly larger than the
subsidence of the other two groups. Because of its
small size, this difference could not have been detected
with conventional Roentgen measurement.
In conclusion, micromotion of hydroxyapatite-
coated tibial components fixed without cement was
similar to that of tibial components fixed with cement.
Therefore, hydroxyapatite, a biological mediator, may
be necessary for the adequate fixation of tibial com-
ponents when cement is not used. The orthopaedic
company selling the prosthesis took the outcome of
the study very seriously; the promotion of the non-
coated cementless prosthesis was terminated.
4. Recent developments in RSA
4.1. Digital RSA
A disadvantage of conventional film-based RSA is
that it requires a lot of user interactions. In each
radiograph measurement, points have to be labelled.
Subsequently, the coordinates of all of these points
have to be measured manually using a highly accurate
measuring table.
In order to reduce the total analysis time of RSA
radiographs, a software package has been developed
that is able to perform the measurements of the
coordinates automatically in digital RSA images
(RSA Clinical Measurement System (RSA-CMS),
firm Medical Imaging Systems (MEDIS), Leiden,
The Netherlands). RSA-CMS (Vrooman et al., 1998;
MEDIS, 2000) can handle scanned conventional
radiographs or direct digital radiographs Digital Imag-
ing and Communications in Medicine in (DICOM)
format: a worldwide standard for digital images in
medicine. This software package runs on a PC with
the Windows NT operating system (Fig. 5).
RSA-CMS utilises image processing algorithms
developed specifically for the automatic detection
and identification of RSA markers, i.e. the calibration
markers, the bone markers, and the prosthesis markers
in RSA radiogrphs. The positions of the marker
centres are automatically determined by using an
extension and improvement of the circle finding
algorithm described by Duda and Hart (1972). After-
wards, the marker positions are enhanced to sub-pixel
accuracy by estimating a paraboloid through the grey
scale profile of the projected markers (Vrooman et al.,
1998). So, not only information of the contour pixels
of the marker is used, but the information of pixels in
the projected area is used for an optimal result.
By means of a fitting algorithm, the calibration
markers—fiducial markers and control markers—are
extracted from the total group of markers and auto-
matically labelled. Furthermore, the software mat-
ches markers in the two radiographs, which make a
stereo pair, reconstructs the spatial coordinates of the
markers, and finally calculates the micromotion of
the endoprosthesis. With RSA-CMS, the RSA pro-
cedure can be performed fully automatically. If
necessary, the user can interactively correct inter-
mediate results for misdetections such as artefacts in
the film surface, numbers in the patient tag, wires,
and screws.
The accuracy of the digital RSA system was
compared to the accuracy of a manually operated
RSA system from Sweden (Win-RSA, Tilly, Lund,
Sweden). For this purpose, we used radiographs of a
phantom and radiographs of patients that were pro-
vided by the Lund University Hospital, Lund, Sweden
(Valstar et al., 2000). In a phantom experiment, the
manually operated system produced significantly bet-
ter results than the digital system, although the max-
imum difference between the median values of the
manually operated system and the digital system was
as small as 0.013 mm for translation and 0.033j for
rotations. These slightly less accurate results were
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389384
probably caused by the film digitiser that was used. In
radiographs of patients, a better scanner was used and
the manually operated system and the digital system
produced equally accurate results: no significant dif-
ferences were found. Again, it was demonstrated that
digital RSA provides a high accuracy.
During the course of our project, digital RSA
systems were also under development at several
other institutions. The first results of digital RSA
were published by Østgaard et al. (1997). In this
semi-automatic system, the positions of markers have
to be indicated manually and the software refines
those positions. Only one validation study was
carried out with this system, but it was never used
in a clinical setting. In Oxford, a digital RSA-system
was developed (Gill et al., 1998) that has been used
in a clinical setting (Alfaro-Adrian et al., 1999). The
Oxford system is also able to assess the position and
orientation of hip implants by using prosthetic land-
marks (Turner-Smith and Bulstrode, 1993). Another
development is the digital measurement module that
was created for the commercially available UmRSA
system (RSA Biomedical Innovations, Umea, Swe-
den). An increase of the accuracy of this digital
module was reported in a comparative study with the
manually operated implementation (Borlin, 1997).
For another commercially available RSA system,
the WinRSA-system (Tilly Medical Products, Lund,
Sweden), a digital measurement module was recently
developed.
Fig. 5. Screen layout of the RSA-CMS software with an analysed RSA radiograph of a hip prosthesis. Note the different labels for fiducial,
control, bone, and prosthesis markers and the lines that connect the corresponding markers in the two radiographs.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 385
No results of this system have yet been published.
Both systems—UmRSA and WinRSA—operate in a
semi-automatic manner, whereas RSA-CMS has the
advantage that it is fully automatic, i.e. the software
finds the markers without any user interaction. The
development of all these digital RSA-systems dem-
onstrates that there is a definite need for RSA-systems
that are faster and easier to use than conventional RSA
systems.
4.2. Model-based RSA
Attaching tantalum markers to prostheses is a
prerequisite for RSA but it may be difficult and is
sometimes even impossible. Furthermore, marking of
implants is an expensive procedure and in some
countries only allowed by the regulatory bodies after
extensive testing and comprehensive documentation.
Therefore, we developed a model-based RSA
method that uses a triangulated surface model of
the prosthesis (Fig. 6). A projected contour of this
model is calculated and this calculated model con-
tour is matched onto the detected contour of the
actual implant (Fig. 7a) in the RSA radiograph (Fig.
Fig. 6. The solid model of the Interax femoral component together
with its meshed representation.
Fig. 7. The optimisation process for model-based RSA illustrated with the interax tibial component: (a) After the region of interest has been
determined by the observer, the contour of the prosthesis is automatically detected by means of the Canny operator. (b) The first estimation of the
position and orientation of the prosthesis model is projected onto the radiograph. (c) An intermediate result of the optimisation procedure: the
overlap of both contours is increasing. (d) The final result of the optimisation procedure: an optimum overlap of both contours has been obtained.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389386
7b). The difference between the two contours is
minimized by variation of the position and orienta-
tion of the model (Fig. 7c). When a minimal differ-
ence between the contours is found, an optimal
position and orientation of the model has been
obtained (Fig. 7d).
The method was validated by means of a phantom
experiment (Valstar et al., 2001). The phantom con-
sisted of a Plexiglas cylinder with 12 1-min spherical
markers embedded in its surface. Three prosthesis
components were used in this experiment: the fem-
oral and tibial component of an Interax total knee
prosthesis (Stryker Howmedica Osteonics, Ruther-
fort, USA) and the femoral component of a Profix
total knee prosthesis (Smith and Nephew, Memphis,
USA). For each experiment, one of the components
was rigidly attached to the base plane of this
cylinder. The calculated model position and orienta-
tion were compared to the position and orientation of
the cylinder. Since the actual motion between the
prosthesis and cylinder was zero, any change in
relative position indicates an error in the assessment
of the micromotion parameters by the model-based
RSA method.
For the prosthetic components used in this study,
the accuracy of the model-based method was found to
be lower than the accuracy of traditional RSA. For the
Interax femoral and tibial components, significant
dimensional tolerances were found that were probably
caused by the casting process and manual polishing of
the components surfaces. For these components, sys-
tematic errors were found for the translations and the
rotations. The largest standard deviation for any trans-
lation was 0.19 mm and for any rotation it was 0.52j.For the Profix femoral component that had no large
dimensional tolerances, the largest standard deviation
for any translation was 0.22 mm and for any rotation it
was 0.22j.From this pilot study, we may conclude that the
accuracy of the current model-based RSA method is
sensitive to dimensional tolerances of the implant. We
aim at improvement of this model-based RSA method
so that it will be insensitive to large dimensional
tolerance and that it will provide an accuracy that is
comparable to the accuracy of traditional RSA. Cur-
rently, these improvements to the model-based RSA
method are subject of research that is carried out at
our department.
5. Conclusion
RSA is a highly accurate, but rather complicated
measurement technique for the assessment of micro-
motion of prostheses. Since progressive micromotion
is an important indicator for inadequate fixation (i.e.
loosening) of prostheses, and extensive short-term
micromotion might indicate a future revision opera-
tion of the prosthesis, clinical RSA studies are impor-
tant. This is illustrated by the outcomes of clinical
RSA studies after the introduction of Boneloc cement,
and in the clinical RSA study in which we studied the
effect of hydroxyapatite on the fixation of knee
prostheses. With conventional Roentgen techniques,
the negative results found in these studies could not
have been obtained in such a short evaluation period,
and with such a small number of patients involved. By
performing clinical RSA studies, a lot of unnecessary
suffering may be prevented.
At Leiden University Medical Centre (LUMC),
simplification of the RSA technique by automation
of the measurements and the introduction of model-
based RSA are two important research topics. Vali-
dation test proved that the novel automated RSA
system has a high accuracy, and subsequently, it is
now used in several clinical RSA studies. However,
when comparing model-based RSA to conventional
RSA, this new technique was less accurate when
implants with large dimensional tolerances were
used. The model-based RSA method needs to be
modified in order to obtain a high accuracy for these
implants. These modifications are currently subject
of research.
6. Future research
Although the current RSA-software functions
properly, its functionality should be extended. With
the current RSA-system, markers are detected with
sub-pixel accuracy. This is done by estimating of a
paraboloid through the grey-scale distribution of a
projected marker. This sub-pixel accuracy can only be
obtained when markers have a uniform background.
Since these requirements are not always fulfilled in
clinical practice, the development of a model for
markers with a non-uniform background needs to be
pursued (Borlin, 1997).
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 387
A better integration of RSA-CMS with Picture
Archiving and Communication Systems (PACS)
should be considered in order to take full-advantage
of the ongoing digitisation and integration of digital
radiology. Currently, there are several DICOM stand-
ards (i.e. for angiography, computed tomography,
magnetic resonance imaging), but a DICOM standard
for RSA still has to be defined. Such a DICOM
standard will make the exchange of RSA-radiographs
between hospitals feasible.
As said before, the model-based RSA-method
needs to be improved to become an alternative for
conventional RSA. One could consider a technique
that omits badly scaled parts of the implant from the
optimisation procedure or a technique that uses addi-
tional three-dimensional measurements on the surface
of each implant in order to scale the model to match
the individual implant. Currently, these modifications
are subject of a study that is being carried out at our
department.
Although in two clinical RSA studies, a correlation
between the short-term micromotion results and the
long-term revision rate caused by aseptic loosening
was demonstrated (Karrholm et al., 1994; Ryd et al.,
1995), more research is needed to fully understand
this relation. Therefore, it is important to extend the
follow-up of patients that have been included in short-
term RSA-studies. Only then, the importance of
performing small clinical RSA-studies can be vali-
dated and can an extrapolation of short-term (i.e. 2
years) RSA micromotion results to long-term (i.e. > 10
years) clinical results be fully justifies. In this per-
spective, we would recommend that a database is set-
up with results of all clinical RSA studies. This
database could be used as register for objective results
on prosthesis fixation.
References
Alfaro-Adrian, J., Gill, H.S., Murray, D.W., 1999. Cement migra-
tion after total hip replacement. A comparison of Charnley Elite
and Exeter femoral stems using RSA. Journal of Bone and Joint
Surgery 81 (1), 130–134.
Borlin, N., 1997. High precision measurements in digital radio-
graphs. Licentiate thesis, Dept. of Computing Science, Umea
University, Report UMINF 97.10.
Dennis, D.E., Schnabel, R.B., 1983. Numerical Methods for Uncon-
strained Optimization and Nonlinear Equation. Prentice Hall,
Englewood Cliffs, NJ.
Duda, R.O., Hart, P.E., 1972. Use of the hough transformation to
detect lines and curves in pictures. Communications of the
ACM 15 (1), 11–15.
Furnes, O., Lie, S.A., Havelin, L.I., Vollset, S.E., Engesaeter, L.B.,
1997. Exeter and Charnley arthroplasties with Boneloc or high
viscosity cement. Comparison of 1,127 arthroplasties followed
for 5 years in the Norwegian Arthroplasty Register. Acta Ortho-
paedica Scandinavica 68 (6), 515–520.
Gill, H.S., Alfaro-Adrian, J., Murray, D.W., Markers, B., 1998. A
new hybrid video/digitiser based Roentgen stereometric meas-
urement system. Journal of Biomechanics 31 (Suppl. 1), 88.
Golub, G.H., Van Loan, C.F., 1989. Matrix Computations, 2nd edn.
John Hopkins Univ. Press.
Hardinge, K., Porter, M.L., Jones, P.R., Hukins, D.W., Taylor, C.J.,
1991. Measurement of hip prostheses using image analysis. The
MAXIMA hip technique. Journal of Bone and Joint Surgery 73
(5), 724–728.
Karrholm, J., 1989. Roentgen stereophotogrammetry. Review of
orthopedic applications. Acta Orthopaedica Scandinavica 60
(4), 491–503.
Karrholm, J., Borssen, B., Lowenhielm, G., Snorrason, F., 1994.
Does early micromotion of femoral stem prostheses matter?
4–7-year stereoradiographic follow-up of 84 cemented pros-
theses. Journal of Bone and Joint Surgery 76 (6), 912–917.
Krismer, M., Bauer, R., Tschupik, J., Mayrhofer, P., 1995. EBRA a
method to measure migration of acetabular components. Journal
of Biomechanics 28 (10), 1225–1236.
Malchau, H., Karrholm, J., Wang, Y.X., Herberts, P., 1995.
Accuracy of migration analysis in hip arthroplasty. Digitized
and conventional radiography, compared to radiostereometry
in 51 patients. Acta Orthopaedica Scandinavica 66 (5), 418–
424.
MEDIS, 2000. MEDIS website: RSA-CMS, digitally auto-
mated Roentgen stereophotogrammetry. http://www.medis.nl/
MainFrameProducts/Products/RsaCms/MainFrameRSA.htm
(accessed 23 April 2002).
Nelissen, R.G.H.H., 1995. Fundamental aspects of the clinical eval-
uation of total joint prostheses. The total condylar knee prosthe-
sis. Thesis, Dept. of Orthopaedics, Leiden University, The
Netherlands.
Nelissen, R.G.H.H., Valstar, E.R., Rozing, P.M., 1998. The effect of
hydroxyapatite on the micromotion of total knee prostheses. A
prospective, randomized, double-blind study. Journal of Bone
and Joint Surgery 80 (11), 1665–1672.
Nilsson, K.G., Dalen, T., 1998. Inferior performance of Boneloc
bone cement in total knee arthroplasty: a prospective random-
ized study comparing Boneloc with Palacos using radiostereom-
etry (RSA) in 19 patients. Acta Orthopaedica Scandinavica 69
(5), 479–483.
Østgaard, S.E., Gottlieb, L., Toksvig Larsen, S., Lebech, A., Talbot,
A., Lund, B., 1997. Roentgen stereophotogrammetric analysis
using computer-based image-analysis. Journal of Biomechanics
30 (9), 993–995.
Reading, A.D., McCaskie, A.W., Gregg, P.J., 1999. The inadequacy
of standard radiographs in detecting flaws in the cement mantle.
Journal of Bone and Joint Surgery 81 (1), 167–170.
Ryd, L., 1992. Roentgen stereophotogrammetric analysis of pros-
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389388
thetic fixation in the hip and knee joint. Clinical Orthopaedics
and Related Research (276), 56–65.
Ryd, L., Albrektsson, B.E., Carlsson, L., Dansgard, F., Herberts, P.,
Lindstrand, A., Regner, L., Toksvig Larsen, S., 1995. Roentgen
stereophotogrammetric analysis as predictor of mechanical
loosening of knee prostheses. Journal of Bone and Joint Surgery
77 (3), 377–383.
Selvik, G., 1989. Roentgen stereophotogrammetry. A method for
the study of the kinematics of the skeletal system. Acta Ortho-
paedica Scandinavica 60 (4), 1–51, Suppl. 232.
Soderkvist, I., 1990. Some numerical methods for kinematical anal-
ysis. Licentiate thesis, Dept. of Computing Science, University
of Umea, Report UMINF-186-90.
Spoor, C.W., Veldpaus, F.E., 1980. Rigid body motion calculated
from spatial coordinates of markers. Journal of Biomechanics 13
(4), 391–393.
Thanner, J., Freij Larsson, C., Karrholm, J., Malchau, H., Wesslen,
B., 1995. Evaluation of Boneloc. Chemical and mechanical
properties, and randomized clinical study of 30 total hip arthro-
plasties. Acta Orthopaedica Scandinavica 66 (3), 207–214.
Turner-Smith, A., Bulstrode, C.J., 1993. Stereoradiogrammetry for
the prediction of hip replacement survival. In: Turner-Smith, A.
(Ed.), Micromovement in Orthopaedics, vol. 10. Oxford Medi-
cal Engineering Series, pp. 113–123.
Valstar, E.R., Vrooman, H.A., Toksvig-Larsen, S., Ryd, L., Nelis-
sen, R.G.H.H., 2000. Digital automated RSA compared to man-
ually operated RSA. Journal of Biomechanics 33 (12), 1593–
1599.
Valstar, E.R., de Jong, F.W., Vrooman, H.A., Rozing, P.M., Reiber,
J.H.C., 2001. Model-based Roentgen stereophotogrammetry of
orthopaedic implants. Journal of Biomechanics 34 (6), 715–
722.
Veldpaus, F.E., Woltring, H.J., Dortmans, L.J., 1988. A least-
squares algorithm for the equiform transformation from spatial
marker co-ordinates. Journal of Biomechanics 21 (1), 45–54.
Vrooman, H.A., Valstar, E.R., Brand, G.J., Admiraal, D.R., Rozing,
P.M., Reiber, J.H.C., 1998. Fast and accurate automated meas-
urements in digitized stereophotogrammetric radiographs. Jour-
nal of Biomechanics 31 (5), 491–498.
E.R. Valstar et al. / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 376–389 389