Announcement

Collapse
No announcement yet.

Knee Axis estimate - loaded vs unloaded? dependence on ROM?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Knee Axis estimate - loaded vs unloaded? dependence on ROM?

    Hi all, my apologies if this has been discussed previously;

    I work in a clinical motion capture lab and we analyze gait and more recently, sports motions. We always (when appropriateness criteria are met) use a posthoc knee-axis correction (KAC) algorithm to 'correct' the thigh coordinate system, with the optimization goal of minimizing cross-talk between the sagittal and coronal planes (adapted from Baker et al. 1999).

    Our thinking was that if we are doing KAC for gait because manual identification of "the correct knee axis" is difficult (in quotes because the knee axis is not static, so any chosen static knee axis is not 'correct', but simply a result of our selected optimization), the same issue may result for sports motions. However, we also though that if we are seeing a different range of motion (RoM) in sports motions (up to 90deg), whereas in gait we are seeing less (60deg), should we also be performing the KAC on an activity representative of the effective RoM for sports motions?

    To test what effect this would have, we collected deep squat trials from participants, and ran our KAC on both their gait, and their deep squat, and they are indeed different.

    Rather than reasoning the difference, I'd like to explore why/if a KAC should be based on different data (gait vs squat), and if an estimated singular knee axis should always be determined based on the activity being investigated - i.e. large knee RoM activity, use a large RoM for KAC

    And should it matter much of the joint is loaded during the trials on which a KAC is performed? Does the knee axis also change in loaded vs unloaded conditions?

    I know some labs use the Knee Alignment Device (KAD), and I wonder if there is an analagous situation for that method - is it crucial to identify the KAD attachment points based on the full knee RoM that is expected in the recorded activity?

    A 6DOF knee joint might make this a moot concern, but we've unfortunately not been able to implement a marker model for a 6DOF knee that doesn't introduce evident non-physiologic translation between the proximal and distal segments..

    Lastly - please feel free to comment just to say that you have a knee axis identification method that you never question! It doesn't seem like there is consensus on this in the field, but please tell me if I'm wrong.

    Thanks all!
    Tim

  • #2
    Tim,
    It is recognized that aligning modelled (estimated) segment axes with the functional (real) axes of rotation is essential to calculating meaningful joint rotations as well as resultant joint forces and moments. As you may be aware, methods for aligning segment axes with function axes of rotations have appeared in the literature for well over two decades. Despite this, the attention to understanding the nature of the errors and methods for their correction has been limited.
    The following represent my experience and views on axes alignment.

    “We always (when appropriateness criteria are met) use a posthoc knee-axis correction (KAC) algorithm to 'correct' the thigh coordinate system, with the optimization goal of minimizing cross-talk between the sagittal and coronal planes (adapted from Baker et al. 1999).”
    In the literature axes alignment is predominantly limited to the knee, minimising co-variance (cross-talk) between knee flex/extension and knee abd/adduction rotations via an offset about the thigh longitudinal axes. What is not recognised is assessing modelled axes alignment and minimizing differences relative to functional axes of rotation is a fundamental requirement of motion analysis, critical to validity and reliability, and should be a routine part of every subject calibration procedure. This includes all modelled joints, axes of rotation and required for all activities.
    To understand non-linear errors in calculated joint rotations, correction of axes alignment and the inter dependency of ordered rotations consider the modelled (estimated) thigh axes. With a rotational error about the thigh longitudinal axis and misaligned med/lateral axis relative to the functional (real) knee flex/ext axes of rotation. The further the knee flexes in the sagittal plane the further the modelled shank longitudinal axis moves away or out of alignment with the true plane of knee flexion. This requires additional characteristic but erroneous calculated knee abd/adduction and int/external rotations to bring the modelled shank axes back into alignment with the true shank position. These additional modelled rotations are not real but a product of calculating an ordered rotational sequence that starts from misaligned modelled medio-lateral axis. The greater the knee real flexion the larger the introduced non-linear errors in all three calculated joint rotations. Although thigh axes misalignment results from multiple sources of systematic and random errors when reconstructing segment axes (marker placement, 3D data reconstruction and tracking, post processing, rigid fixation devices, joint degrees of freedom). The misalignment of thigh med/lateral axes leads to characteristic, non-random, errors in calculated knee abd/add, knee int/ext, and knee flex/ext rotations as well as an offset in hip int/ext rotation. A small thigh med/lateral axis misalignment in the frontal plane can also be introduced by a medially or laterally positioned hip joint centre. Also producing characteristic errors in calculated knee abd/add and Int/Ext rotations but with an offset in calculated hip abd/adduction. Therefore, correcting thigh med/lateral axis alignment and errors in calculated hip and knee joint rotations requires correction in two planes, the thigh longitudinal axes offset and medio-lateral correction of the hip joint centre.
    Methods for aligning modelled axes with functional axes of rotation in the literature commonly minimise co-variance (crosstalk) between calculated knee flex/ext and abd/adduction angles during gait. Deriving a rotational offset about the thigh longitudinal axes to minimise this co-variance. This will improve knee joint angle validity and reliability over no axis alignment. However, minimization of covariance has shortfalls. It assumes the knee functional axes of rotation coincides with minimum co-variance between flex/ext and abd/add, ignores natural relationships and co-variance that may exists between joint rotations, ignores non-linear errors and co-variance that are also present in knee int/external rotation, it does not account for the interdependence of all three cardan rotations or the dependency of segment proximal and distal joint rotations. Reworking normal gait data (n = 40) gave a correlation between knee flex/ext and knee abd/add of 0.81 and with knee Ext/Int of -0.19. Introducing a thigh longitudinal axis offsets found the minimum cross correlation between knee flex/ext and abd/add required an internal 4.4 deg rotation of the thigh. While the minimum cross correlation between knee flex/ext and ext/Int required an external 4.1 deg rotation of the thigh about the longitudinal axis. With knee flex/ext displaying an almost universal double positive peak (stance, swing) and small knee abd/add movements close to zero (1.8 degrees during stance), minimising co-variance may be predetermining the nature of the knee abd/add curve.

    “… if we are doing KAC < knee-axis correction > for gait because manual identification of "the correct knee axis" is difficult …, the same issue may result for sports motions.”
    Reconstructing all segment axes from external markers is inexact, based on approximations and multiple sources of error that propagate through the analysis process. As mentioned, carrying out a subject calibration that assesses errors and aligns segment axes with function axes of rotations is critical to obtaining meaningful joint rotations. This applies to all modelled joints (legs, trunk and arms), axes of rotation, and all movement to be analysed.

    “… we also though that if we are seeing a different range of motion (RoM) in sports motions (up to 90deg), whereas in gait we are seeing less (60deg), should we also be performing the KAC on an activity representative of the effective RoM for sports motions?”, and similarly; “… if an estimated singular knee axis should always be determined based on the activity being investigated - i.e. large knee RoM activity, use a large RoM for KAC”
    I use a controlled slow squat in a dynamic calibration procedure to assess and align modelled axes with functional axes of rotation as it is weight bearing with predominantly sagittal plane movements at the hip, knee and ankle joints. In this movement it is reasonable to expect that each joint abd/add and int/ext joint rotations will be minimal. The squat is also of sufficient ROM, about 90 degrees at the knees, to recognise and minimise the characteristic non-linear errors in calculated joint angles across the hips, knees and ankles. The dynamic calibration routine is importantly independent of the movement that will be analysed (walk, run, jump, landing, side-step, etc.).
    It is important to firstly address the multiple sources of error that are part of the modelling process. If left unchecked these may adversely affect reconstructed segment axes and confound attempts to align segment axes. Methods include marker placement, 3D data reconstruction and tracking, post processing, avoiding rigid fixation devices, and use of virtual joint centres and 6 degrees of freedom joints.
    It is also worth noting that I use an initial static calibration to obtain approximate segment axes alignments as a starting point for the dynamic calibration. In the static pose I measure pelvic med/lat tilt, hip abduction, knee flexion, rear foot angle, and foot progression angle. These measurements along with estimated knee abd/adduction and ext/Internal rotations and foot abduction angle, are used to give a first estimate of segment alignment.

    “ … we collected deep squat trials from participants, and ran our KAC on both their gait, and their deep squat, and they are indeed different.”, and; “ Rather than reasoning the difference, I'd like to explore why/if a KAC should be based on different data (gait vs squat) …”
    I would be hesitant to use a deep knee squat as you could be introducing real knee int/external rotation and abd/adduction into the movement that you are attempting to minimise. It is for this reason you do not use the whole gait cycle to minimize non-linear errors in knee abd/adduction and ext/internal rotations. Gait single foot support produces on average 1.8 deg (+- 0.9) adduction at the knee. While the swing phase sees 7.4 deg (+- 2.4) knee abd/adduction ROM. Some of this deviation in stance and swing may be skin movement artefact. A flexed knee during swing will see either internally, neutral or externally rotated lower leg and accompanying abd/adduction at the knee. Potentially forcing axes misalignment to reduce a real rotation during swing at the expense of minimal abduction during stance. As mentioned, use the squat as the dynamic calibration movement to align axes. Similarly, I would not use the jump, run, side step, land… etc. movement itself to minimise abd/adduction rotations in order to align axes with axes of rotation. As you are potentially changing a real rotation that you are trying to measure.

    “… should it matter much of the joint is loaded during the trials on which a KAC is performed? Does the knee axis also change in loaded vs unloaded conditions?”

    As mentioned, for the legs it should be a controlled, weight bearing and predominantly a sagittal plane movement. It is a requirement of the approach (minimising co-variance or nonlinear errors) that joint abd/adduction and int/external rotations can be expected to be minimal. An unweighted flexed knee has the potential for internal or external rotation with abduction or adduction if not controlled.

    “ A 6DOF knee joint might make this a moot concern …”

    A subject calibration procedure that assesses and aligns modelled axes with functional axes of rotation is essential regardless of the degrees of freedom modelled at the joints. I only use 6 degrees of freedom joints when reconstructing segment axes, calculating joint angles and calculating resultant joint forces and moment in an inverse dynamics approach. I would not use 3 df joints as it imposes unrealistic constraints when reconstructing segment axes in an inexact, error prone, and loosely defined system. Maintaining these joint constraints during movement will force the rigid linked model into unrealistic joint rotations. Which, in my view, makes 3 df joints incompatible with minimize non-linear errors and aligning axes across multiple body segments.

    “… you have a knee axis identification method that you never question!”
    Unfortunately, motion analysis is full of poor methods based on assumptions that are never questioned.

    “It doesn't seem like there is consensus on this in the field, but please tell me if I'm wrong.”
    I would agree that there is not a lot of consensuses on methods. Despite a thread of research in this area over the past 20+ years it has not received the critical though and investigation needed. Re-iterating, my responses are based on my experience in motion analysis and may raise more questions than it answers.

    Cheers
    Allan Carman

    Comment


    • #3
      Baker's KAC algorithm finds a knee axis that minimizes knee ab-adduction. As I understand it, the goal is to create a more reproducible and more reliable joint coordinate system (JCS), so that 3D knee rotations are less sensitive to errors in marker placement.

      If you want to apply this to movements other than gait, you would have to investigate that reproducibility, like Baker in the 1999 paper. It may not work as well as in gait. Loads are larger, possibly go outside of the sagittal plane, and you may well have large ab-adduction rotations that are real and can't be explained as crosstalk. In our work on healthy athletes [1], we defined a JCS based on a carefully defined standing posture and foot alignment, rather than on markers. Similar to the KAD you mentioned. That was quite reproducible, but results are possibly contaminated if the participant was not able to stand with straight knees.

      A possible alternative would be to include some gait trials in your test protocol for sports movements. Then the sports movements can be analyzed with a JCS that has been corrected using a KAC based on the gait data. It will be just as reliable as it was in gait analysis (because it was done in the same way), and you're not assuming that the sports movement has no ab-adduction.

      Ton van den Bogert

      [1] McLean et al. (2005) Association between lower extremity posture at contact and peak knee valgus moment during sidestepping: Implications for ACL injury​. Clin Biomech 20: 863–870.

      Comment

      Working...
      X