Tweet
Briefly.
Kinematic fitting (sometimes also called optimization) assumes that there is
an error/cost function that reaches a minimum when joint parameters and
joint angles attain their correct values. Explaining the complete rationale
behind this assumption is left as an exercise for others!
Kinematic fitting modifies some/all joint parameters and joint angles, often
in a nested manner, to minimize a cost function (e.g. cost = sum of squared
distances between measured and predicted marker trajectories). Further
explanation can be found in Lu T-W and O'Connor JJ, J Biomech, 32:129-134,
Charleton et al, Gait Posture, 20:213-221, or Rienbolt et al, J Biomech,
38:621-626.
As a personal note -- I have had **no end of grief** trying to use
optimization approaches. I believe that this is due to their sensitivity wrt
"false" minima, and other forms of indeterminacy. For example, when applying
this approach to the talo-crural/sub-talar comples, one can often find very
good (low cost) solutions that are totally meaningless (i.e. "optimal" axes
at approximately right angles to known anatomical orientation).
The moderator has some experience with this kind of approach at the
foot/ankle and may want to chime in (as may the honorable Dr. Baker
regarding his LE experiences).
Functional methods directly translate models of joints, which are generally
significant simplifications (i.e. hip as 3 dof, or knee as having fixed
axis) into equation form - then solve these equations.
Functional methods use observed motions to predict joint parameters.
Example, hip joint as center of sphere (Cappozzo A, Hum Movement Sci,
3:27-50) or as pivot of intersecting finite axes of rotation between pelvis
and thigh (Schwartz and Rozumalski, J Biomech, 38:107-116), or effective
(stationary) knee axis as mode of observed finite axes of rotation between
thigh and shank (Schwartz and Rozumalski, J Biomech, 38:107-116). These are
just some examples - as stated previously there are many very good
functional algorithms.
Regarding ROM - both methods require this type of data. Sensitivity to small
range has not been compared between the two; though there is some nice work
examining sensitivity to ROM *within* several functional algorithms.
As implemented at Gillette, small ROM is not an issue since the ROM trials
are assisted by the tester. For the method to work, only the markers on
segments adjacent to the joint need to be unmolested (e.g. pelvis and right
thigh segment must be untouched for a right hip ROM trial). We usually
provide balance support (hand-holding) and assist with the ROM by moving the
limb for/with the patient.
We use four ROM trials (R hip, L hip, R knee, L knee). The hip trials are in
a "star arc" pattern -- shown to be quasi-optimal by Camomilla V et al, J
Biomech, 39:1096-1106. The knee trials are a passive/aided ROM (approx 10-60
degrees flexion). The entire process takes 5-7 minutes, even with subjects
who have significantly compromised neuromuscular and cognitive abilities. As
noted in previous posting, we use this method (along with the PIG model) on
**every** patient we see, regardless of age, size, involvement, etc...
In terms of effect on moments -- we have lots of data comparing functional
to PIG moments since we collect both sets of data on every subject. **On
Average** the two models are quite similar with the exception of Hip
Ab/Adduction -- where the errantly narrow regression-based hips (Leardini et
al, J Biomech, 32:99-103) result in a bias/shift. I have presented this data
at several forums (abstracts and talks) but regret that I have been too
busy/lazy/etc...to write it up as a peer-reviewed article (OK, now I'm more
motivated). Despite the average findings -- individual differences (that is
differences between the two models on individuals) can be substantial.
Regarding reliability -- we have *limited data* (small N, non-pathological,
acquired as part of our lab's routine QA efforts) showing significantly
better reliability with functional data (kinematics and moments). Again,
this has been presented at conferences and the like, but is not in press
(mea culpa).
I hope this helps clarify things.
-Mike-
_______________________
Michael H. Schwartz, Ph.D.
Director of Bioengineering Research
Gillette Children's Specialty Healthcare
Associate Professor - Orthopaedic Surgery
University of Minnesota
Graduate Faculty - Biomedical Engineering
University of Minnesota
Kinematic fitting (sometimes also called optimization) assumes that there is
an error/cost function that reaches a minimum when joint parameters and
joint angles attain their correct values. Explaining the complete rationale
behind this assumption is left as an exercise for others!
Kinematic fitting modifies some/all joint parameters and joint angles, often
in a nested manner, to minimize a cost function (e.g. cost = sum of squared
distances between measured and predicted marker trajectories). Further
explanation can be found in Lu T-W and O'Connor JJ, J Biomech, 32:129-134,
Charleton et al, Gait Posture, 20:213-221, or Rienbolt et al, J Biomech,
38:621-626.
As a personal note -- I have had **no end of grief** trying to use
optimization approaches. I believe that this is due to their sensitivity wrt
"false" minima, and other forms of indeterminacy. For example, when applying
this approach to the talo-crural/sub-talar comples, one can often find very
good (low cost) solutions that are totally meaningless (i.e. "optimal" axes
at approximately right angles to known anatomical orientation).
The moderator has some experience with this kind of approach at the
foot/ankle and may want to chime in (as may the honorable Dr. Baker
regarding his LE experiences).
Functional methods directly translate models of joints, which are generally
significant simplifications (i.e. hip as 3 dof, or knee as having fixed
axis) into equation form - then solve these equations.
Functional methods use observed motions to predict joint parameters.
Example, hip joint as center of sphere (Cappozzo A, Hum Movement Sci,
3:27-50) or as pivot of intersecting finite axes of rotation between pelvis
and thigh (Schwartz and Rozumalski, J Biomech, 38:107-116), or effective
(stationary) knee axis as mode of observed finite axes of rotation between
thigh and shank (Schwartz and Rozumalski, J Biomech, 38:107-116). These are
just some examples - as stated previously there are many very good
functional algorithms.
Regarding ROM - both methods require this type of data. Sensitivity to small
range has not been compared between the two; though there is some nice work
examining sensitivity to ROM *within* several functional algorithms.
As implemented at Gillette, small ROM is not an issue since the ROM trials
are assisted by the tester. For the method to work, only the markers on
segments adjacent to the joint need to be unmolested (e.g. pelvis and right
thigh segment must be untouched for a right hip ROM trial). We usually
provide balance support (hand-holding) and assist with the ROM by moving the
limb for/with the patient.
We use four ROM trials (R hip, L hip, R knee, L knee). The hip trials are in
a "star arc" pattern -- shown to be quasi-optimal by Camomilla V et al, J
Biomech, 39:1096-1106. The knee trials are a passive/aided ROM (approx 10-60
degrees flexion). The entire process takes 5-7 minutes, even with subjects
who have significantly compromised neuromuscular and cognitive abilities. As
noted in previous posting, we use this method (along with the PIG model) on
**every** patient we see, regardless of age, size, involvement, etc...
In terms of effect on moments -- we have lots of data comparing functional
to PIG moments since we collect both sets of data on every subject. **On
Average** the two models are quite similar with the exception of Hip
Ab/Adduction -- where the errantly narrow regression-based hips (Leardini et
al, J Biomech, 32:99-103) result in a bias/shift. I have presented this data
at several forums (abstracts and talks) but regret that I have been too
busy/lazy/etc...to write it up as a peer-reviewed article (OK, now I'm more
motivated). Despite the average findings -- individual differences (that is
differences between the two models on individuals) can be substantial.
Regarding reliability -- we have *limited data* (small N, non-pathological,
acquired as part of our lab's routine QA efforts) showing significantly
better reliability with functional data (kinematics and moments). Again,
this has been presented at conferences and the like, but is not in press
(mea culpa).
I hope this helps clarify things.
-Mike-
_______________________
Michael H. Schwartz, Ph.D.
Director of Bioengineering Research
Gillette Children's Specialty Healthcare
Associate Professor - Orthopaedic Surgery
University of Minnesota
Graduate Faculty - Biomedical Engineering
University of Minnesota