CGA FAQ: Marker Sets

As you know, I recently took delivery of a Vicon 370 system. I'm of
course very pleased to have one, but I thought I'd share one concern
which I have so far.

The standard clinical gait analysis (Vicon Clinical Manager) protocol
uses the so-called Davis marker set, which consists of Sacrum, right
and left: ASIS, a lateral mid-thigh marker on a wand, a knee marker on
the "flexion axis" of the knee, a lateral mid-shank marker on a wand,
lateral malleolus and foot (dorsum). The knee flexion axis is determined
during a static trial by applying what is called a knee alignment
device, which is a 3-marker pod which one aligns by with where the axis
of the knee is thought to be (by asking the patient to alternately flex
and extend their knee). This static trial also calculates the angles
between the thigh and shank wands for use in anthropometric calculations
to derive the knee joint centre. Once the knee axis has been defined,
the alignment device is removed, the knee marker is attached in its
place, and the gait is then ready to be analysed.

I wonder what people think of this procedure? It worries me, and I
suspect I'm not the only one, that such an important part of the
analysis process is highly dependant on a subjective judgement (defining
the knee axis of flexion). As you know, the knee axis is known to change
quite markedly depending on the knee angle, travelling a few centimetres
up the femur in an evolute as the knee flexes.

So I wonder if you could answer the following questions for me...
 

   1.1. If you use a Vicon yourself, do you get good results from the knee alignment device?
   2.2. If you use another marker set, can you please describe it, and tell  us whether you think it is superior to the Davis set?
   3.3. Is anybody routinely using a method for calculation of the instantaneous axis of the knee, rather than simply predicting
     it from anthropometry and a static trial?

Please don't think I'm criticising the Davis set at this stage - I'd
just like to know what people think of it, since it seems to be such a
crucial part of the analysis. I'd also be willing to conduct some
experiments to assess the realtive merits of any other marker sets
people are using. Seems to me we should know the strengths and
weaknesses of each before we can interpret results.

Look forward to your replies.

Chris
--
Dr. Chris Kirtley MD PhD
Dept. of Rehabilitation Sciences
The Hong Kong Polytechnic University
 

The Davis marker set I know of is different from the one you described.
(See Davis et al. (1991) A gait analysis data collection and reduction
technique, Human Movement Science, 10/5, pp 575-587).

The ELITE system (BTS, Milan, Italy) uses the marker set in the above
paper. I have serious reservations with the method, but I've learned to
live with it.

The knee varus/valgus and rotation angles are said to be inaccurate by
the authors, but I also have problems with the hip rotation angles too.

I am looking forward to the discussion on this topic.

Gabor
--
Dr Gabor Barton MD                   CGA@gaitlab.demon.co.uk
Manager, Gait Analysis Laboratory    UIN (ICQ): 2625928
Alder Hey Children's Hospital        tel: +44 (0)151 252 5949
Eaton Road, Liverpool, L12 2AP, UK   fax: +44 (0)151 252 5846
 

Chris:
        You have asked some very interesting questions
about the marker set used by Vicon.  It seems that everyone
who gets one of their systems goes through some wrestling
with the model and how to cope.  Thanks for bring this to
the list for discussion.
        We have a Vicon 370 system here in Portland.  I
have used it for about 4 years and love it.  It does have
its limitations, but to me they are well balanced with it
accuracy and speed.
        Vicon user group meetings always include a session
on marker placement and discussion of the limitations of
the model.  I was just leading a marker placement group at
the Gait and Clinical Movement Analysis Society meeting in
San Diego, California last month so these topics are fresh
in my mind.
        Here in Portland we capture a static trial without
the knee alignment devise (KAD) prior to collecting walking
trials and another with the KAD following walking.
        We place the makers by aligning the thigh wand to
the knee axis.  We ask the patient stand and to lift their
foot off the floor by flexing their knee to 90 degrees while
keeping the thigh vertical.  The patient holds a railing for
added stability. We align the thigh wand perpendicular to
the shank with the knee flexed to this 90 degree position.
        We then place the knee marker on what appears to be
the center of rotation.  We ask the patient to squat and
rise, flexing the knee.  It is important to have the
patient face straight ahead during marker placement.
Looking down at the action makes wand placement impossible
due to rotations. We have an assistant who's job it is to
engage the patient and keep then facing forward.
        We use a eye liner pencil (rubs off easily) to mark
the proposed center of rotation, and watch the knee flex
and extend.  This first mark is placed by feeling the joint
line and determining the center of the condyle in
extension. The manual says about 1 cm above the joint line
but it is of course different for smaller of larger
patients.  If the dot appears to move excessively, we try
another mark above or below the first to see if one rotates
about the other. We shoot for greatest accuracy in
extension.  After about 300 or so patients all the variants
will be seen and even the impossible knees will just get
hard. :-) I have moved a marker 1 cm in each direction and
found very minimal changes in the hip and knee motions in
all planes.  Knee extension changed less than 2 degrees
for example.
        After walking tails are captured, we take an
additional static trial with the knee alignment device.
This allows us to determine if all wands and markers remain
correctly aligned and allows calculation of an offset which
can be entered into the Vicon Clinical Manager (VCM)
calculations. There are some risks associated with this
approach (Micah Forstein of Los Angeles Childrens Hospital
has suggested that it is possible to enter an "impossible"
offset, one which alters the thigh rotation without
changing the knee axis to a more posterior position for
example.) but our offsets are usually less than 10 degrees,
making the risks acceptable.
        Or you can choose to use the KAD trial as the
static and process all trials by this method in VCM.  We
don't like this because it forces the tibial wand
parallel to the the knee axis.  We don't feel that the
transmalloelar axis is parallel to the knee axis, but
rather feel that it is more external.  Or you must enter
a measure offset to correct this, which almakes me wonder
if I can measure the tibial rotation why can't I place a
wand there just as accurately?  And which one do I accept
if they are different?  Which one is wrong?
        We are one of the few labs to include a transverse
plane knee rotation graph (tibial rotation) in our reports.
Many others consider these too error prone to include but
we have compared our to CT scans of transmalloelar axes and
found high R squared values.  We feel that the tibial
rotation on the gait report is at least as accurate as
clinical evaluation or CT scans. (see Aktas et al. (1998)
Gait and Posture 7:2 (March)p. 182)
        Including this graph allows us to visually confirm
the interaction of the proposed hip rotation, knee
flexion/extension axis, knee varus/valgus axis and tibial
rotation.  If the coronal knee motion begins to have the
appearance of the knee motion in the sagittal plane, then
wand placement error is suspected.  The closer to neutral
(zero) it is the better we feel about the accuracy of the
hip rotation and knee axis.  Provided we feel that there is
no "true" coronal motion of the knee.
        If we enter an offset into VCM to alter the thigh
wand placement and therefore the calculation of the knee
axis, we can reprocess the trial and watch the changes to
the transverse plane thigh rotation, the sagittal knee
motion, the coronal knee motion and the transverse plane
tibial rotation.  It is the interaction of these which
makes one feel confident about the offset or uncomfortable
with the offset.
        Now that I have gone on forever, what was your
question?  Oh yeah.  Yes the knee alignment device assists
in placing the knee joint center in many patients.
        I need to point out that VCM uses the thigh wand to
perform a cord function which moves the knee marker from
the lateral position on the skin into the virtual space
within the knee.  After the hip joint center is established
by the leg length calculation (now if you want to talk
accuracy and error, just open this can of worms) VCM uses
the hip joint center, the knee marker and the thigh wand to
establish the plane which defines the thigh.  The knee
marker is moved medially along this plane (defined by the
thigh wand) into the center of the knee.
        You are correct to point out the skin movement at
the knee and its effect on knee motion, mainly in the
sagittal plane.  I have routinely calculated the distance
between the hip joint center and knee joint center and the
distance between the knee joint center and the ankle joint
center on a series of about 70 patients.  This distance
between the knee and ankle looks suspiciously like the knee
flexion/extension axis and varies about 3 cm (shorter in
flexion).
        I asked Roy Davis about instantaneous knee joint
centers from non-connected thigh and shank segments and he
said they had tried it but that it was so error prone that
it was useless.  Perhaps someone else has done more work on
this issue recently.  Did you get BodyBuilder with your
package?  You could always develop your own maker set in
your spare time. :-)
        I had better stop and get something done today.
Thanks for keeping the list going and confirming the web
site existence.  We true believers never doubted.

        Michael

Michael Orendurff, MS
Clinical Biomechanist
Gait Analysis Laboratory
Portland Shriners Hospital
MSO@shcc.org
 

First and Foremost, I appreciate this type of discussion.  My objective is to assist the Clinical Gait Community in
further making systems better so that the children are the main beneficiary. I have the most respect to all
researchers.
 
I have only used a Vicon system a few times in recent months but use a Motion Analysis system daily.  I head up the
Biomechanics Division for Motion Analysis Corporation.   Marker sets are among the most controversial topic in gait
assessment.  As a former researcher with an advanced degree in Biomechanics, I am amazed what restrictions clinicians will
accept based upon the limitations of their system.
 
To our knowledge, there are two markers sets used in Clinical Gait Analysis.
 

         Full Body marker set, and
         Lower body (Sacrum R/L Asis and below) with and/or without heel markers

 
Now the Full Body marker set is then subdivided into two options:
 

       Modified Helen Hayes Marker set utilizing wands at the l/r thigh and lower leg and surface markers at other
     locations
       Cleveland Clinic Marker set utilizing triads at the l/r Thigh and lower leg (Used exclusively by Motion Analysis
     Corporation) and based upon cadavour studies at Cleveland Clinic Foundation.

 
The Cleveland Clinic Marker set has always been the more difficult marker set since it requires a static left and right leg
video record in order to calculate and determine joint centers for the ankle, knee and hip.  The Modified Hayes marker set,
relied upon Knee Alignment Deviced, Ankle devices, Clinicians eyeballing the locations and or utilizing some device specifically
designed by the clinician.  Again, Motion Analysis takes a static video recording of the legs even with the Hayes marker set to
calculate and determine joint centers.
 
One problem that I personally see with the alignment devices is that if the alignment device is not properly aligned,  IE:
Aligned 3 degree upward or downward,  Is the patient's gait assessment then measured in this 3 degree clinician error.  It
would seem practical and simple that the video system should be able to calculate automatically a static measurement for Joint
Centers calculations and the use such calculations for the dynamic assessment.
 
The Second problem that I see is with a "standard wand" located at the l/r thigh and lower leg.  I know with the Triad
(Cleveland Clinic Markers) the patient has three markers at the thigh instead a single wand and the same with the lower leg.
With the three points specifically located on the segment, you would get a more accurate assessment of thigh and lower leg
rotation especially in children with severe internal or external rotation of the thigh and lower leg.  I think Stanhope and Holden
studied a similar marker set as reported in Gait and Posture.  Of course the wand vs the Triad tracks faster but with today's
systems of auto identification, the clinician needs to emphasize accuracy since time has been negated.
 
The third issue that I see is that there is "NO Standard" marker set  even with in the Vicon  users.  I know that I can name
several Vicon users that with the same system their markers sets are different.  Now, I am not attacking Vicon nor Promoting
Motion Analysis.  The fact remains that years ago these systems created by Oxford Metrics couldn't track a cluster of markers
and Motion Analysis systems could not track markers without editing.   But today's systems are different, and are on the verge
of Passive Marker Real-time display.
 
Now is the time where the Clinical Gait Community needs to come to an agreement on:
 

        1) A full body marker set, ( I question the validity of measuring only the lower body only to determine surgical
     outcomes).
        2) Data reporting format.

 
Once determined, tell the manufacturers and let them further design the hardware and software to accept  your
agreed upon standards.
 
Motion Analysis is listening to your needs (and sure others are too) and are making changes in its software to accommodate
research and collaboration in Clinical Gait by all users of systems.  Whether you produce a C3D or a P3D (which we produce
both), we hear that the researcher wants the binary data to collaborate and share data and it all starts with the "acceptable"
marker sets (nomencalture included) and we (manufacturers) make changes.
 
Again, I respect this forum, my competitors, clinicians and not advertising but providing input from a manufacturer's
perspective.
 
Motion Analysis Corporation
Daniel India, Vice President
3617 Westwind Blvd
Santa Rosa, CA  95403   USA
HQ Tel:  707-579-6500  Direct 847-945-1411
HQ Fax 707-526-0629  Direct  847-945-1442
dan.india@motionanalysis.com
 

Hi Chris

Glad to hear that you are happy with the Vicon.

As you point out Vicon Clinical Manager VCM utilises an exact
implementation of the Davis model as published in:

Kadaba, Ramakrishnan and Wootten of Helen Hayes Hospital
(J Orthop. Res., Vol. 8, No., 3, 1990)
 Davis, Ounpuu, Tyburski and Gage of Newington Children's Hospital
(Human Movement Sciences #5, Volume 10, 1991)

If you don't have these papers to hand I will be happy to fax you
copies.

As for the 'Davis Marker set'  the beauty of this if how it reduces
the 7 segments to a mere 13 markers during the dynamic trial.  As you
have rapidly noticed one of the limitations of this is the need for
very careful marker placement.

Roy Davis recently spoke at our Vicon Users meeting at NASCGMA and
presented some work on the errors created by misplaced markers.  It is
quite remarkable how robust this technique is!  Also many studies have
been done to compare marker placement repeatability between staff in
various labs and between labs. The findings again show how comparable
data from this model is.  Also I have seen work comparing skin markers
versus transcutaneous  (patients with Ilizarhov frames ready to be
removed) that show pleasingly consistent results.

So what is clear is that this is a good, easy to use and repeatable
tool and therefore universally applicable.  However this raises the
question: is this sufficient for the new millennium?  To which the
answer is: we don't really know!  Until recently there have been few
other options for comparison and certainly no models using enough
markers to give a rigorous (3 or more markers per segment)
measurement.

Now for the ADVERT.  You have a 370 and BodyBuilder.  (I am aware of
some other labs who have been putting multiple markers on segments. Up
to 43 in total for their full body model. They are comparing this to
VCM type data.)  With your kit it will be easy to produce alternative
models.

Geoff Shaw <geoff.shaw@metrics.co.uk>
 

Chris,
   We have a 5 camera, Vicon 370 system that we use for bilateral lower
extremity gait analysis.  We do use the Vicon Clinical Manager (VCM) with the
marker set you described in your posting.  We also use the KAD and enter a value
for tibial torsion.  Regarding your first question:

1. If you use a Vicon yourself, do you get good results from the knee
alignment device?

I am not very confident in the hip and foot rotation measures as well as knee
varus and valgus, that we get using VCM and the KAD.  Transverse plane rotations
of the thigh and shank are always interpreted with caution.  We attempt to
define the knee flexion axis accurately by marking the axis during a physical
exam.  But, as you pointed out, it is still subjective.  You can increase your
confidence by watching the output knee varus and valgus curve.  If there is an
increase in knee varus and an internal hip rotation offset during swing, when
there is an increase in knee flexion, you may enter an external thigh rotation
offset in your session file and reprocess your data.  This may help correct
errors in placement of the KAD.  I have additional documentation that may not be
included in the manual regarding this adjustment process.  Hope this helps.

Greg Lange

     Gregory W. Lange, M.S.                         (210) 705-6597 voice
     Engineer, Biomechanics Laboratory              (210) 705-6567   fax
     Santa Rosa Outpatient Rehabilitation Center    greg_lange@srhcc.org
     for Children and Adults
     2701 Babcock Road, San Antonio, Texas,  78229,  USA
     http://www.santarosahealth.org/BioMechanics.html
 

Although we do not use a passive marker system, we do use a protocol for
locating the knee and hip center similar to the one Chris has described. We
presently employ a Selspot II optoelectric system where arrays of LEDs, mounted
on rigid plastic disks, are strapped to the mid section of the feet, shanks,
thighs, pelvis, trunk, arms and head. The knee and hip centers are located by
first using a hand held pointer, upon which is mounted an array of LEDs, to
specify the sagittal plane upon which the joint center is assumed to lie. The
subject then performs a large range of motion task - chair rise. An axis of
rotation method is then used to compute directional vectors of proximal and
distal segment endpoints; the intersection of the directional vectors with the
sagittal pointing plane is assumed to be the joint center of rotation (see Riley
et al., J Biomech 23, 503-506, 1990).

We recently had an opportunity to examine the accuracy of this technique by
comparing repeated measures of hip center location in a single subject (several
times over three years), using the above protocol, to accurate photogrammetric
measures of the same subject's hip centers from his explanted pelvis following
death (see McGibbon et al., Clin Biomech, 12, 491-495, 1997). The pointing
technique was, on average, within 1.2 cm of the "true" hip center. Anatomical
scaling techniques for locating the hip center were also applied; errors ranged
from 1.5 to 5 cm using various published scaling factors. However, we have not
had the opportunity to repeat these experiments on other humans, so
generalization is limited. Nevertheless, the pointing method does appear
satisfactory.

Chris A. McGibbon, PhD <ronny@gait52.mgh.harvard.edu>
MGH Biomotion Laboratory
Boston MA
 

Thanks for all your comments.

It seems to me that there are two separate problems (right?):
 

   1.1. To define an axis for each segment: e.g. hip joint centre to knee joint centre
   2.2. To define an axis of rotation for one segment wrt another

Could someone enlighten me why we actually need (2). In reality there is
no axis - there may be several or none - so why don't we simply measure
the motion of, say, shank with respect to thigh? An axis is really
superfluous to the problem, isn't it? Or is it simply to express the
result in clinically meaningful terms (sagittal plane flexion/extension,
frontal plane ab/adduction)?

I can see that (1) implicitly requires the joint centres to be defined.
Could we not, therefore, find some method to get away from the joint
centres as definitions for the local axes?

Please correct me if I'm not understanding the problem - 3D kinematics
was never my strong point!

Chris
--
Dr. Chris Kirtley MD PhD
Dept. of Rehabilitation Sciences
The Hong Kong Polytechnic University
 

The question you ask is a fairly fundamental one about kinematic
analysis and the linked segment model concept. If you use unlinked
segments (with no common axes) it becomes harder to describe the motion
in terms of three rotations as we do now. Not impossible, but difficult
to do such that the answers can be visualised. It is well known that the
concept of the knee flexion axis is an approximation, but it's a very
useful one, and over the range of motion used in gait, quite a good one.
There are much more significant sources of error.

The concept of a centre of rotation in the knee is less obviously
valid than at the hip. At the hip, the mechanical ball-and-socket
structure means the problem reduces to estimating the centre of the
socket. At the knee, the choice of the midpoint of the flexion axis is
rather an arbitrary one and less obviously related to anatomical
structure. As usual, the response to such questions is - can anyone
suggest a better method which is as practical for routine use? We hope
that people will experiment with the alternatives, and that the better
alternatives will emerge over time.

Martin Lyster <martin.lyster@metrics.co.uk>
 

If you have two linear segments floating freely in space, you can always
define a common plane and measure the angle between them (this is what
370 does if you define an angle using four points and make a pop-up
graph). Alternatively you could do some kind of projection into fixed
planes. However it is less easy to interpret than the linked segment
model in which there is assumed to be a common axis about which the two
segments rotate wrt each other. Also, if the segments are not linked in
your model, I don't know how you can do kinetics.

I've been thinking some more about the free body problem. Its not a
simple question. There is a related problem in mechanics which goes
something like, if a car drives into a lamp post and the bottom of the
lamp post snaps off the ground, what is the subsequent motion of the
lamp post. It's not an easy problem.

Martin <martin.lyster@metrics.co.uk>
 

Chris, you pose some interesting questions. Consider your two problems:

   1.1. To define an axis for each segment: e.g. hip joint centre to knee joint centre
   2.2. To define an axis of rotation for one segment wrt another

For determining joint angles or segmental kinematics in general, the joint
center (2) is not required. The center of rotation, however, does have utility.
For instance, knowing where the center of rotation of the knee is in relation to
the the femoral and tibial surfaces allows prediction of the contact kinematics
(the Oxford school of thought that the knee's instant center in the sagittal
plane is located at the intersection the ACL and PCL "isometric" fibres is one
application in the design of prosthetic components where contact kinematics have
been inferred from knowing the joint center - see for example Goodfellow and
O'Connor, JBJS, 60B, 355-369, 1978). The instant center is also a convenient
place to sum moments because it is generally assumed that articular forces in
normal synovial joints (under low coefficient of friction - ice on ice) do not
have shear components and hence the line of action of the articular force acts
through the center of rotation: this allows elimination of the articular force
in the moment equation for computing tendon or ligament forces (which provide
the shear restraint).

I can see that (1) implicitly requires the joint centres to be defined.

Not really. It was common to see 2D motion analysis labs place markers on
external landmarks approximating joint centers (such as lateral epicondyle or
lateral malleolus), but these are really just segment reference markers for
determining angles (and in my opinion a mistake to assume they are also joint
centers). In modern 3D systems, any three (or more) markers forming a reference
plane can be used to determine the 6 DOF kinematics of the segment, provided one
knows the transformation between the reference plane and the assumed location of
the segment coordinate axis. The origin of the coordinate system does not have
to be located at a joint center (although it usually is).

Could we not, therefore, find some method to get away from the joint centres as definitions for the local axes?

I believe this to be one issue at the center of the debate for standardization
of gait lab measures (along with other issues such as sequence of rotation
matrix decomposition into angles, etc), and I'm sure there is someone out there
who can shed some light on the current status of implementing these standards.

Regards,
Chris A. McGibbon, PhD
MGH Biomotion Lab
Boston, MA <ronny@gait52.mgh.harvard.edu>
 

This has turned into an essay, but I think the gait analysis community is
getting really lazy in its widespread and largely uncritical acceptance of
the Kadaba/Davis model.  I'm off for two weeks leave beginning today so
I'll not be able to give any feedback, but this may be no bad thing.

I haven't really got time to contribute to this discussion but it is one
that concerns/interests me. Chris McGibbon's comment that the present drive
for standardisation might be addressing the use of marker sets is
particularly concerning because if this standardisation takes place in the
near future we will undoubtedly be stuck with some variant of the
Kadaba/Davis model as the standard. There is no doubt that these models
have achieved an immense amount in informing the way we analyse, present
and interpret gait analysis data but I feel strongly that the time is
coming when the limitations of the models are becoming more fully understood
and when work should be being done on better alternatives.

Most of the limitations are acceptable as long as they are understood but
as gait analysis becomes a more and more standard procedure I suspect a
smaller proportion of staff involved in its application are actually fully
aware of these limitations. There are particular problems when the output
of standard gait analysis packages are being used to drive other
biomechanics models such as muscle length calculations.

Chris Kirtley is correct in that there are essentially two requirements,
which  I would state slightly differently:

   1.1) to determine the position and orientation of each segment (mathematically, to define an axis system)
   2.2) to determine the relative orientation (and for some applications position) of one segment with respect to another (He is
     also correct in that this in itself does not actually require the definition of a joint centre).

One of the problems with the Kadaba/Davis model is that it seeks to use the
minimum number of markers possible and in doing so merges the two steps
above into one. This was important when the model was developed as the
technology available then could not cope with very many markers. Modern
technology however does not impose such a limitation.

In order to determine the position and orientation of a segment it is
necessary to define three points on the segment. For the pelvis these
points are three markers. To determine the thigh you also need three points
but Kadaba/Davis spotted that the hip joint is fixed in both the pelvis and
thigh axis systems. You can therefore get away with one less marker by
using the hip joint centre as one of the points for the thigh. A similar
argument uses the knee joint centre as one of the points for the shin and
the ankle joint centre for the foot. Thus we save ourselves three markers
on each side. This is technically irrelevant given the capacities of modern
data capture systems.

However, in using this approach we have introduced several problems.
Firstly, the system is hierarchical so any inaccuracy in determining one
segment's axis system will carry on to an inaccuracy in determining the
axis systems for all distal segments. Secondly, by definition the system
cannot cope with polycentric joints where the joint centre is not fixed in
the axis system of both the proximal and distal segment. Thirdly, the only
sensible way of determining joint angles is as rotations about the fixed
joint centres. Fourthly, because all the information captured by the system
is used in deriving the model there is no redundant information which can
be used to check its validity.

These are theoretical limitations. The most important practical limitation
has already been alluded to by Chris Kirtley, the difficulty in placing the
thigh wands accurately/repeatably. This is an art not a science and those
who do place markers repeatably do it after many months (if not years) of
practice, thinking about the process and critically reviewing their
results. Many centres, I am sure, simply do not place them repeatably. It
doesn't need an alignment device, there are other methods. It is quite easy
to place the markers repeatably on adults who can stand still and
co-operate it is quite a different matter on kids who can't (or won't)
stand still. Having said this it is comparatively easy to spot. If the knee
has medio-lateral stability and the varus-valgus trace has the same shape
as the knee flexion-extension curve then the most likely explanation is
that this is an artefact of poor thigh marker placement leading to
 incorrect definition of the thigh axis system. For able bodied subjects
knee varus-valgus is essentially zero throughout the gait cycle. If you get
a varus-valgus signal for such subjects which looks anything at all like
the flexion-extension graph (or a reflection of it in the x-axis) then you
are not putting the thigh markers on correctly. Because of the fact that
the thigh marker is used to determine the knee joint centre this may also
have a significant effect on hip flexion-extension, knee flexion-extension,
ankle dorsi/plantar flexion. Much more concerning when dealing with CP is
that if you see this artefact then the hip rotation profile will also be
wrong. Beware the surgeon who performs proximal derotation osteotomies of
the  femur based on gait analysis data without an understanding of this
limitation of the Kadaba/Davis model!

I am looking at a technique which uses this varus/valgus artefact to
determine a correction factor for thigh wand alignment, similar to that
calculated when using the knee alignment device. It works well for those
with a good range of knee flexion during gait but still requires work for
those with limited knee flexion (ie real patients!). Unless you suspect
medio-lateral instability of the knee or are supremely confident in your
placement of thigh markers of knee alignment devices I suspect that the
most reliable definition of the sagittal plane of the thigh is that in
which the knee flexes.

Of course this is being a little unfair on the model. The main problem is
that the thigh does not have sufficient well defined anatomical landmarks
to define internal/external rotation. However the model's requirement of
the markers chosen leading to a fairly reliable definition of the knee
joint centre restrict choice here. Freeing ourselves from the hierarchical
definition of segments to an independent definition of segments could prove
a distinct advantage.

Getting away from the requirement to use as few markers as possible also
allows the possibility of giving us extra information. Cappozzo's group in
Italy have pioneered the use of more markers not less. This gives redundant
information which can be used to as the basis for optimisation techniques
which lead to more reliable determination of segment axis systems. This is
a particularly good way of working with skin movement artefact. If the
effect of skin movement is different at different points then "averaging"
the readings from several markers can reduce those effects.

It has been pointed out that independent determination of segment axis
systems leads to more noisy data for joint angles. An alternative and
equally valid expression of this is to say that, by forcing artificial
constraints on the data, the Kadaba/Davis suppresses noise at the expense
of the reliability of the signal. Determining the instantaneous centre of
rotation is notoriously noisy (it is not, however, actually required to
obtain joint angles). However the noise is essentially a measure of the
validity of the model. By seeking to minimise the number of markers
required the Kadaba/Davis model is very poor at giving such measures of
data validity.

Does anyone else fancy discussing the limitations of the model for
determining hip joint centre from pelvic markers?

This has turned into a bit of an essay rather than a response to the
original question. I do hope it puts the questions regarding the widespread
adoption of this marker set on the agenda and makes people think more
critically about its application.

I would end by saying that until someone comes up with a better model which
is easy to use clinically, we will, like the vast majority of gait labs
throughout the world, continue to use the Kadaba/Davis model. Better the
devil you know ...

Richard Baker  <baker@unite.co.uk>
Gait Analysis Service Manager
Musgrave Park Hospital
Stockman's Lane
BELFAST    BT9 7JB

Tel:   01232 669501 ext 2155
Fax:  01223 683816
 

I haven't really got time to contribute to this discussion but it is one that concerns/interests me. Chris McGibbon's comment that
the present drive for standardisation might be addressing the use of marker sets is
particularly concerning because if this standardisation takes place in the near future we will undoubtedly be stuck with some
variant of the Kadaba/Davis model as the standard. There is no doubt that these models
have achieved an immense amount in informing the way we analyse, present and interpret gait analysis data but I feel strongly
that the time is coming when the limitations of the models are becoming more fully understood and when work should be being
done on better alternatives.

Most of the limitations are acceptable as long as they are understood
but as gait analysis becomes a more and more standard procedure I suspect a
smaller proportion of staff involved in its application are actually fully
aware of these limitations. There are particular problems when the
output of standard gait analysis packages are being used to drive other
biomechanics models such as muscle length calculations.

Chris Kirtley is correct in that there are essentially two requirements, which  I would state slightly differently:
 

   1.1) to determine the position and orientation of each segment(mathematically, to define an axis system)
   2.2) to determine the relative orientation (and for some applicationsposition) of one segment with respect to another (he is
     also correct in that this in itself does not actually require the definition of a joint centre).

One of the problems with the Kadaba/Davis model is that it seeks to use
the minimum number of markers possible and in doing so merges the two steps
above into one. This was important when the model was developed as the
technology available then could not cope with very many markers. Modern
technology however does not impose such a limitation.

In order to determine the position and orientation of a segment it is
necessary to define three points on the segment. For the pelvis these
points are three markers. To determine the thigh you also need three
points but Kadaba/Davis spotted that the hip joint is fixed in both the pelvis
and thigh axis systems. You can therefore get away with one less marker by
using the hip joint centre as one of the points for the thigh. A similar

argument uses the knee joint centre as one of the points for the shin
and the ankle joint centre for the foot. Thus we save ourselves three
markers on each side. This is technically irrelevant given the capacities of
modern data capture systems.

However in using this approach we have introduced several problems.
Firstly, the system is hierarchical so any inaccuracy in determining one
segment's axis system will carry on to an inaccuracy in determining the
axis systems for all distal segments. Secondly, by definition the system
cannot cope with polycentric joints where the joint centre is not fixed
in the axis system of both the proximal and distal segment. Thirdly, the
only sensible way of determining joint angles is as rotations about the fixed joint centres.
Fourthly, because all the information captured by the system is used in deriving the model
there is no redundant information which can be used to check its validity.

These are theoretical limitations. The most important practical
limitation has already been alluded to by Chris Kirtley, the difficulty in placing
the thigh wands accurately/repeatably. This is an art not a science and
those who do place markers repeatably do it after many months (if not years)
of practice, thinking about the process and critically reviewing their
results. Many centres, I am sure, simply do not place them repeatably. It
doesn't need an alignment device, there are other methods. It is quite
easy to place the markers repeatably on adults who can stand still and
co-operate it is quite a different matter on kids who can't (or won't)
stand still. Having said this it is comparatively easy to spot. If the
knee has medio-lateral stability and the varus-valgus trace has the same
shape as the knee flexion-extension curve then the most likely explanation is
that this is an artefact of poor thigh marker placement leading to incorrect definition
of the thigh axis system. For able bodied subjects knee varus-valgus is essentially zero throughout the gait cycle. If you
get a varus-valgus signal for such subjects which looks anything at all like the flexion-extension graph (or a reflection of it in the
x-axis) then you are not putting the thigh markers on correctly. Because of the fact that the thigh marker is used to determine
the knee joint centre this may also have a significant effect on hip flexion-extension, knee
flexion-extension, ankle dorsi/plantar flexion. Much more concerning when dealing with CP
is that if you see this artefact then the hip rotation profile will also be wrong. Beware the surgeon who performs proximal
derotation osteotomies of the  femur based on gait analysis data without an understanding of this
limitation of the Kadaba/Davis model!

I am looking at a technique which uses this varus/valgus artefact to
determine a correction factor for thigh wand alignment, similar to that
calculated when using the knee alignment device. It works well for those with a good range of knee flexion during gait but still
requires work for those with limited knee flexion (ie real patients!). Unless you suspect
medio-lateral instability of the knee or are supremely confident in your placement of thigh markers of knee alignment devices I
suspect that the most reliable definition of the sagittal plane of the thigh is that in
which the knee flexes.

Of course this is being a little unfair on the model. The main problem
is that the thigh does not have sufficient well defined anatomical
landmarks to define internal/external rotation. However the model's requirement of
the markers chosen leading to a fairly reliable definition of the knee
joint centre restrict choice here. Freeing ourselves from the
hierarchical definition of segments to an independent definition of segments could
prove a distinct advantage.

Getting away from the requirement to use as few markers as possible also
allows the possibility of giving us extra information. Cappozzo's group
in Italy have pioneered the use of more markers not less. This gives
redundant information which can be used to as the basis for optimisation
techniques which lead to more reliable determination of segment axis systems. This
is a particularly good way of working with skin movement artefact. If the
effect of skin movement is different at different points then
"averaging" the readings from several markers can reduce those effects.

It has been pointed out that independent determination of segment axis
systems leads to more noisy data for joint angles. An alternative and
equally valid expression of this is to say that, by forcing artificial
constraints on the data, the Kadaba/Davis suppresses noise at the
expense of the reliability of the signal. Determining the instantaneous centre
of rotation is notoriously noisy (it is not, however, actually required to
obtain joint angles). However the noise is essentially a measure of the
validity of the model. By seeking to minimise the number of markers
required the Kadaba/Davis model is very poor at giving such measures of
data validity.

Does anyone else fancy discussing the limitations of the model for
determining hip joint centre from pelvic markers?

This has turned into a bit of an essay rather than a response to the
original question. I do hope it puts the questions regarding the
widespread adoption of this marker set on the agenda and makes people think more
critically about its application.

I would end by saying that until someone comes up with a better model
which is easy to use clinically, we will, like the vast majority of gait labs
throughout the world, continue to use the Kadaba/Davis model. Better the
devil you know ...

Richard Baker <baker@unite.co.uk>
Gait Analysis Service Manager
Musgrave Park Hospital
Stockman's Lane
BELFAST    BT9 7JB

Tel:   01232 669501 ext 2155
Fax:  01223 683816
 

ronny@gait52.mgh.harvard.edu wrote:

I believe this to be one issue at the center of the debate for
standardization of gait lab measures (along with other issues such as sequence of rotation
matrix decomposition into angles, etc), and I'm sure there is someone out
there who can shed some light on the current status of implementing these standards.

In the last 9 years I have been involved in several projects for Gait
Analysis standardization, within the CAMARC II European project and now
within the North American Society of Gait and Clinical Movement Analysis -
Subcommette for Gait Analysis Modelling (NASGCMA). Because I haven't seen
reply from any member of the commettee, I would just like to say that
several efforts have already been made to try to propose to the Gait
Analysis comunity a common reference model for marker placement and joint
kinematics and kinetics calculation. The members have already agreed many
points and are still thinking about many others. I think we will be able to
submit a preliminary proposal of the model at the next NASGCMA meeting in
Dallas, March 1999. Most of the agreed issues are present in the paper:
Position and orientation in space of bones during movement: anatomical
frame definition and determination      A.Cappozzo, F.Catani, U.DellaCroce,
A.Leardini      Clinical Biomechanics, Vol.10(4) 1995.

I would suggest the reading of such paper also because it may help at least
in the definition of the most critical points in clinical gait analysis.
The proposed use of the anatomical landmark calibration seems also to be
now well accepted, because applied (althought in different contexts and
applications) in many gait analysis models.

I would be glad to receive comments, suggestions, criticism on that paper,
which could be considered as a first draft for future models. It has
already been accepted in the CAMARC II project as the official model of the
European comunity.

I would like just to give my 2 cent contribution to this point

   1.1. To define an axis for each segment: e.g. hip joint centre to kneejoint centre
   2.2. To define an axis of rotation for one segment wrt another

Could someone enlighten me why we actually need (2). In reality there is
no axis - there may be several or none - so why don't we simply measure
the motion of, say, shank with respect to thigh? An axis is really
superfluous to the problem, isn't it? Or is it simply to express the
result in clinically meaningful terms (sagittal plane flexion/extension,
frontal plane ab/adduction)?

In order to calculate clinical meaningful variables we first need to
determine position and orientation of reference coordinate systems rigidly
associate to the bones, and oriented according to the functional anatomy.
Potitions of joint centres and other anatomical landmarks are necessary for
this purpose. We then need to define axes about which we can consider that
the joint rotations occur. These axes ought to be selected according to
traditional way of thinking of clinicians, providing an objective
instrument to describe flexion/extension, ab/adduction and
internal/external rotation. The mechanical convention from Grood and Suntay
1983 seems to be a good way for this representation. To calculate joint
rotations we can select whatever axis. However, to make the final results
with some clinical meaning we do need a precise definition of the axis of
rotation. A standardisation of these definition would allow for result
comparison and data exchange.

More problems are instead involved in the accurate and repeatable
identification of the anatomical landmarks, and their artefact-free
tracking in space by external markers. Many investigations are still in
progress to address this problems.

Richard Baker wrote:

Does anyone else fancy discussing the limitations of the model for
determining hip joint centre from pelvic markers?

We have in press on J Biomechanics the following paper:
Validation of a functional method for the estimation of hip joint centre
location        A.Leardini, A. Cappozzo, F. Catani, S.Larsen, A. Petitto, V.
Sforza, G. Cassanelli, S. Giannini

Here is the abstract:
The present study deals with the accuracy with which the hip joint center
(HJC) is estimated using different methods. The functional approach has been applyed by
calculating the center of the best sphere described by markers placed on the thigh in a pelvic
anatomical frame during several trials of hip rotations. Two different prediction approaches, for
which HJC is estimated by regression equations using antrophometric measurements (Bell 1990, Davis
1991), were also tested. The accuracy of the three techniques was investigated by comparing their predictions with
measurements obtained using the Roentgen Stereophotogrammetric Analysis (RSA), who was
assumed to provide the true HJC location. In contrast with some previous studies, the functional approach seems to be
more reliable than the prediction approaches proposed so far. If care is taken in
maximizing the amplitude of rotations and minimizing the skin movement artefact, then this approach may
provide local HJC coordinates with a minimum error of few millimeters. The method is
therefore suggested in gait tests in which is applicable, as it allows a reliable evaluation of
the subject-specific HJC location even in cases of pelvis and hip deformities.

Alberto Leardini
Movement Analysis Laboratory
Department of Orthopaedic Surgery - Istituto Ortopedico Rizzoli

Address:
Via Di Barbiano, 1/10
40136 Bologna
ITALY

Tel & Fax: ++39 51 6366561 (secretary)

E-mail: <VI6BOQ71@ICINECA.CINECA.IT>
 

Can I also throw in a question related to a commen Richard Baker threw in
to the marker set debate:  why do people think that the medial rotation on
a motion graph is related to bony torsion??  Surely the model is trying to
measure relative motion of the thigh in relation to the pelvis?  In the
Vicon static trial with KAD's, you try to get a handle on the amount of
torsion, so that the motion graph will indeed reflect movement.  In many
cases of increased medial femoral torsion plus femoral head anteversion the
hip is actually, anatomically laterally rotated (although it still looks
medially rotated from the outside, due to the bony configuration).  You can
do proximal derotation to correct torsional alignment in the long bones,
but it will not necessarily alter hip rotation curves, even though things
like foot progression angle may improve towards normal.  Portland Shrine
presented a paper on that showed this at AACPDM 2 years ago in Minneapolis.
 I think the assessment for torsional correction via bony surgery is mainly
via clinical exam +/- 3D or 2D CT or other imaging study, not via hip
rotation motion graphs!

Jenni

Jenni Dabelstein, Consultant Physiotherapist
Technology Unit (Equipment Technology Services)
Cerebral Palsy League of Queensland
354 Bilsen Road, Geebung 4034
AUSTRALIA
ph:  61 7 3874 2050
fax: 61 7 3874 2051
e-mail:  techunit@iname.com
 

I'm worried too about the knee flexion axis definition with the
Vicon. I'm using it with children, and other problems apart, the KAD
won't stay on a child's knee in any where near the right place.  I
use the single knee marker instead, but I'm not convinced that I can
place it within 5 maybe 8 mm of the 'correct'? place.  Add a similar
error to the thigh marker, just 50 / 70mm off the mechanical axis of
the femur (100 if you use a wand, but they are too easy for kids to
catch and move) and the knee flexion axis is 30 degrees off.
Having said that, I usually have no more than 10 degrees of knee
valgus showing in the kinematic report, but its always going to mask any real varus / valgus.

We are examining this problem and others related to Vicon /
children, having just established our 'team' at Sheffield Children's
lab.

Rob

J.R.Strachan <J.R.Strachan@sheffield.ac.uk>
Royal Hallamshire Hospital
Sheffield
UK  


Back to FAQ