Identification of the contribution of the ankle and hip joints to multi-segmental balance control
© Boonstra et al.; licensee BioMed Central Ltd. 2013
Received: 23 January 2012
Accepted: 24 January 2013
Published: 22 February 2013
Human stance involves multiple segments, including the legs and trunk, and requires coordinated actions of both. A novel method was developed that reliably estimates the contribution of the left and right leg (i.e., the ankle and hip joints) to the balance control of individual subjects.
The method was evaluated using simulations of a double-inverted pendulum model and the applicability was demonstrated with an experiment with seven healthy and one Parkinsonian participant. Model simulations indicated that two perturbations are required to reliably estimate the dynamics of a double-inverted pendulum balance control system. In the experiment, two multisine perturbation signals were applied simultaneously. The balance control system dynamic behaviour of the participants was estimated by Frequency Response Functions (FRFs), which relate ankle and hip joint angles to joint torques, using a multivariate closed-loop system identification technique.
In the model simulations, the FRFs were reliably estimated, also in the presence of realistic levels of noise. In the experiment, the participants responded consistently to the perturbations, indicated by low noise-to-signal ratios of the ankle angle (0.24), hip angle (0.28), ankle torque (0.07), and hip torque (0.33). The developed method could detect that the Parkinson patient controlled his balance asymmetrically, that is, the right ankle and hip joints produced more corrective torque.
The method allows for a reliable estimate of the multisegmental feedback mechanism that stabilizes stance, of individual participants and of separate legs.
KeywordsBalance control Closed-loop system identification Multivariate systems Asymmetry
Maintaining an upright posture is a relatively easy task for healthy humans [1, 2]. However, bipedal upright stance is inherently unstable, as small deviations from the upright posture result in disturbing torques due to gravity, which drives the system further away from upright posture . To stay upright, the body generates corrective torques to counteract the effects of internal (e.g., motor and sensory noise) and external (e.g., uneven surfaces) perturbations.
When postural deviations are small, the body is often simplified as an inverted pendulum pivoting at the ankles, which describes the so-called ankle strategy [3–5]. However, several studies demonstrated that human movement during stance is multi-segmental [6–8] and for example, the hips substantially contribute to upright stance (i.e., the hip strategy ). Human balance control is a closed-loop multi-segmental process, i.e., sensory signals about the movement of the body are fed back to the central nervous system (CNS), and the CNS controls the muscles to generate adequate responses [10, 11]. In a noisy closed-loop system, like human balance control, causality is difficult to determine and the dynamics of the different components (i.e., the body and the stabilizing mechanisms located in the CNS) affect both the input (joint angles) and output signals (joint torques). To “open” the loop and to separate the dynamics of the different components, the balance system need to be perturbed [11, 12]. Furthermore, when estimating the dynamics in a multivariate system, multiple perturbations need to be applied [13, 14].
Most studies investigating the multivariate nature of balance control do not take the multivariate noisy closed-loop nature into account, by either not using perturbations [15–17], or by using only one perturbation [18–20]. Only two studies investigated the multivariate nature of balance control by applying two perturbations [2, 21].
Fujisawa and colleagues  investigated the role of the hip joint to upright stance by applying pseudorandom perturbations (bandwidth 0 - 0.83 Hz) while manipulating the support surface width. Subsequently, an ARMAX model (with joint angles as input and joint torques as outputs) was fitted to the data to obtain the Frequency Response Functions (FRFs) of a two-segment model of balance control. Results showed an increase of balance contribution of the hip joint, when the support surface became narrower.
Jeka and colleagues  identified neural feedback during upright stance in 18 subjects, while applying two mechanical perturbations (springs attached to a linear motor) and one sensory perturbation (visual scene rotations). By comparing the identified neural feedback (from joint angles to weighted electromyograms (EMGs) of the leg and trunk segments) with a large range of cost functions, it was concluded that the CNS stabilizes stance with near minimum muscle activation.
Ageing and many neurological diseases are associated with balance impairments and falls . Understanding the (patho)physiology of upright stance could aid to detect individuals with an increased risk of falls, help to design and evaluate intervention programs or monitor disease progression. Therefore, for clinical applications, it is very important to obtain a reliable individual estimate of balance control.
Of all neurological diseases, PD patients are at the highest risk of falling [22–24], but the pathophysiology of balance impairments in PD remains unclear [25, 26]. Recently, it was suggested that one of the factors contributing to decreased balance control in PD patients, is impaired trunk control [27, 28] or a decreased intersegmental coordination [29, 30]. Another factor could be asymmetrical balance control, that is when one leg produces more force than the other leg to maintain an upright posture. Asymmetries in balance control have been rarely studied in PD, although it is an asymmetrical disease . One study  found balance control asymmetries in 24% of the PD participants, indicating that balance asymmetries are important in PD.
Currently, there is no method available that can identify a multisegmental stabilizing mechanism of balance control on an individual level, separating the contribution of the joints of the left and right body side. We developed and evaluated a non-parametric MIMO (Multiple-Input-Multiple-Output) identification method based on the previously used non-parametric system SISO (Single-Input-Single-Output) identification method . To obtain a reliable individual response, periodic perturbations were applied, which have the advantage of having power at specific frequencies, decreasing the measurement time, and increasing the participants response. In addition, the stabilizing mechanisms were estimated based on left and right joint torques (contrary to weighted EMGs; ), to be able to investigate balance control asymmetries.
In sum, our goal was to develop a method that can reliably estimate the stabilizing mechanisms of the closed-loop multivariate balance control system of individual participants, making a distinction between the contribution of the left and right leg. The (clinical) applicability is demonstrated in an experiment that perturbed the balance of seven healthy participants and a PD patient with a novel device that can apply two independent mechanical perturbations.
A two degree-of-freedom (DoF) balance control model (described extensively in Appendix A) was implemented in Matlab (The Mathworks, Natick, USA) and simulated with Simulink (equations were solved with a fifth order Dormand-Prince algorithm). The human balance model consisted of a two-segment human body with two actuators (ankle and hip), which were controlled using feedback of the joint angles (ankle and hip). In the model, no distinction between the left and right leg was made.
two perturbation forces with noise (2F-N),
two perturbation forces without noise (2F-Nn),
platform perturbation and ankle torque with noise (PLT-N),
platform perturbation and ankle torque without noise (PLT-Nn),
one perturbation and one perturbation round with noise (PL-N),
one perturbation and one perturbation round without noise.
Perturbation amplitudes, scaling of the perturbation signal, Goodness of Fit and Noise-to-Signal Ratios of the model simulations and the balance control experiment
PL: 0.06; T: 20Nm
PL: 0.06m; T: 16Nm
PL: 0.07; T: 18.4Nm
PL: 0.06m; T: 16Nm
PL: 0.06; T: 20Nm
GOF (1/Hz 2 )
Seven healthy participants (two female, mean age 65 yrs. std 5.7) and a PD patient (male, 57 yrs) participated in the study. The participants gave written informed consent prior to participation. The protocol was approved by the local medical ethics committee and was in accordance with the Declaration of Helsinki.
Apparatus and recording
Two independent perturbations were administered with a computer-controlled six DoF motion platform (Caren, Motek, Amsterdam, The Netherlands) and a custom-built actuated device able to apply perturbing forces in the anterior and posterior direction at the sacrum, called the pusher (see Figure 2). The pusher was attached to the platform and actuated using a series elastic actuator  controlled with an electro motor (Maxon motor ag, Sachseln, Switzerland). The pusher was force controlled using a custom-built controller in xPC (The Mathworks, Natick, USA) and had a bandwidth of 10 Hz and a maximum torque of 50 Nm. The gravitational pull due to the weight of the pusher was compensated for, such that the participants did not experience additional forces other than the perturbation force and a small force due to the pusher’s inertia. The interaction force in between the subject and the pusher was measured with a six DoF force transducer (ATI-Mini45-SI-580-20).
Body kinematics and the platform movements were measured using motion capture (Vicon Oxford Metrics, Oxford, UK) at a sample frequency of 120 Hz. Reflective spherical markers were attached to the following anatomical landmarks: the first metatarsal, calcaneus, medial malleolus, the sacrum, the manubrium and the last vertebrae of the cervical spine (C7). In addition, a cluster of three markers was attached to both anterior superior iliac spines on the pelvis. Furthermore, one additional marker was attached to the foot and two markers were attached to the lower leg (one on the tibia) to improve the estimation of the ankle joint rotational axis. Also, markers were attached to the knee (just below the lateral epicondyle) and shoulder joints (just in front of the acromion). Three markers were attached to the platform. Reactive forces from both feet were measured with a dual forceplate (AMTI, Watertown, USA), embedded in the motion platform. The signals from the dual forceplate, the 6 DoF force transducer and the perturbation of the pusher were sampled at 600 Hz and stored for further processing.
During the experiment, participants stood with their arms folded in front of their chest on the dual forceplate and strapped to the pusher, with a strap band that opened with a click buckle, with their eyes open. They were instructed to maintain their balance without moving their feet, while multisine platform movements and multisine force perturbations were applied simultaneously in the forward-backward direction; see ‘Disturbance signals’. Participants wore a safety harness to prevent falling, but it did not constrain movements, provide support or orientation information in any way.
Before any data was recorded, the participants got acquainted to the perturbations. The experimenter determined the maximal amplitude the participant could withstand while keeping the feet flat on the floor and assessed whether the participant could withstand this amplitude for the total of four trials. Four double perturbation trials of 180s were recorded: in the first two trials, the perturbations had the same sign. In the other two trials, the perturbations had opposite signs. If needed, the participants were allowed rest in between trials.
For both the model simulations and the experiment we used the same perturbation signal. During the model simulations the perturbation signal was used to either produce two perturbing forces, or a combination of a platform translation and a torque around the ankle (see ‘Model simulations’). In the experiment the perturbations were applied with a movement platform and an actuated backboard (see ‘Apparatus and recording’).
The perturbation signal was a multisine with a period of 34.13 s (equal to 212 = 4096 samples at a sample rate of 120 Hz) [4, 35]. This signal contained power at 112 frequencies in the range of 0.06–4.25 Hz. To increase the power at the excited frequencies the signal was divided into five frequency bands: 0.06-2.37 Hz (80 frequencies), 2.63-2.84 Hz (8 frequencies), 3.11-3.31 Hz (8 frequencies), 3.57-3.78 Hz (8 frequencies), 4.04-4.25 Hz (8 frequencies). The frequency points outside these frequency bands were not excited. The signal is unpredictable for participants, because the signal consists of many sinusoids. The power of the signal was optimized by crest optimization .
The human body, i.e. the plant, is considered as a double-inverted pendulum, consisting of a leg and a Head-Arms-Trunk (HAT) segment. Stabilizing mechanisms generate ankle and hip torques based on sensory information of the joint angles, (see Figure 1).
The stabilizing mechanisms have passive components such as muscle stiffness, generated by passive muscle properties and tonic activation. The active part incorporates the controller within the CNS (e.g. reflexive muscle activation), muscle activation dynamics, and time-delays, representing the neural signal conduction times.
Movements from the upper body segment will influence the movements of the lower body segment and vice versa due to mechanical coupling [36, 37]. The stabilizing mechanisms have to deal with this mechanical coupling, which is especially expressed in coupling terms between ankles and hips (i.e., ). The direct terms represent the corrective actions of the ankle and hip joint, based on the ankle and hip joint angle. In other words, this system is a multiple input (two joint angles) multiple output (two joint torques) system. When considering the corrective actions of both legs separately, two stabilizing mechanisms are defined; one for each leg [4, 38].
For the model simulations, the perturbations, inputs (joint angles) and outputs (joint torques) of the model were determined for further processing. For the experiment these signals were calculated as described below.
From the recorded movement trajectories of the markers, the position of the center-of-mass (CoM) of the predefined segments and of the whole body and the position of the joints were estimated by custom written software [39, 40]. In short, in each segment a local coordinate frame was determined on the basis of the position of anatomical landmarks, according to the method described by Koopman et al., 1995 . The mass, CoM position and the inertia tensor moment of the predefined segments (i.e., feet, legs and HAT) and the joint positions were determined with regression equations [41, 42]. Subsequently, the CoMs were determined as the weighted sum of the separate segment CoM positions . From the static trial, the average distance in the sagittal plane from the ankle to the total body CoM (i.e., the length of the pendulum; lCoM) was determined. The sway angle was calculated from lCoM and the horizontal distance from the CoM to the mean position of the ankles. Forces and torques of the force plate and force sensor were filtered with a fourth-order low-pass Butterworth filter with a cut-off frequency of 8 Hz and subsequently resampled to 120 Hz. Forces and torques of the force plate were corrected for the inertia and mass of the top cover . On the basis of the corrected forces and torques and recorded body kinematics, ankle and hip joint torques were calculated with inverse dynamics . In addition, the applied platform perturbation was reconstructed from the platform markers.
Multiple-Input-Multiple-Output closed-loop system identification
With Y(f), a two-by-two matrix with on the first column the responses of the first perturbation round (i.e., ankle and hip joint angles or torques) and on the second column the responses of the second perturbation round. denotes the estimate of the cross spectral density (CSD) of the disturbance and the outputs (joint angles and joint torques).
With and the estimated CSD from the perturbations to the corrective torques and from the perturbations to the joint angles. Note that C is a two-by-two matrix, see also Figure 1. p is a vector with the two disturbances, θ(f) is a vector with ankle and hip joint angles, and Tc(f) is a vector with ankle and hip joint torques for each frequency f; all expressed as Fourier coefficients. The method assumes that the system does not change between the two separate perturbation rounds, with the two different combinations of the periodic excitation signal (Eq. 1).
For both the model simulations and the experiment, data was obtained for eight response cycles of the perturbation signal for each perturbation round. Subsequently, the data were Fourier transformed and only the Fourier coefficients at the excited frequencies were used for further processing. These were averaged over the eight cycles, and the average Fourier coefficients were used to calculate the power- and cross spectral density (PSD and CSD, respectively). The PSDs and the CSDs were smoothed by averaging over four adjacent frequency points . The FRFs were calculated according to Eq. 2 and 3 to obtain a non-parametric spectral estimate of the total stabilizing mechanisms.
As the corrective torque which has to be delivered by the participants is dependent on gravity, all FRFs were normalized for the participants’ mass and length, i.e. the gravitational stiffness (mgl), with m the total body mass, l the length of the pendulum (from the ankles to the Center of Mass; CoM), and g the gravitational constant. The average FRF over all participants was obtained by taking the mean over the individual normalized FRFs. Note that, as we used a dual forceplate in the experiment, the experimentally obtained Fourier coefficients of the left and right FRFs were added to obtain the total FRFs.
Reliability of the estimated MIMO frequency response functions
To determine whether the above described MIMO closed-loop system identification technique gives reliable estimates of the stabilizing mechanisms, several indicators were calculated (described below).
Goodness of estimate
Where a perfect estimation of the transfer function will result in a GOF of zero, i.e., the lower the GOF the better the estimation.
Where U p (f) represents the periodic response and σ u 2(f) the variance of the remnant. The NSR was calculated in the frequency domain with f frequency in Hz. When multiple realizations are simulated or recorded, the periodic response is obtained by calculating the average over the realizations; the remnant can be estimated by calculating the variance over the realizations.
A small NSR indicates low variability of the system’s response to the perturbation signal over the multiple periods. This indicates time invariant behavior and a low presence of noise . As such, it gives insight into whether linear, time-invariant, system-identification methods can be used and it gives an estimate of how well the system is perturbed. More importantly, it quantifies how reliable the estimate of the stabilizing mechanism is. For example, a NSR of 1 indicates that the recorded data contains 50 percent response and 50 percent remnant. This means that describing the system as a deterministic linear time invariant (LTI) system (expressed by the estimated FRFs in this study) explains 50 percent of the recorded data.
Single-Input-Single-Output frequency response functions
To assess whether the two perturbation rounds in the experiment lead to a change in strategy of the participants (i.e., time-variant behavior), we estimated Single-Input-Single-Output (SISO) FRFs from sway angle to ankle torque and from sway angle to hip torque with the joint-input-output-method (Eq. 3; ).
In which p represents the platform disturbance and y an output signal. φp,y, φp,p and φy,y are the CSDs and PSDs of the perturbation and output signals. By definition, coherence varies between 0 and 1, where coherence close to one indicates a low noise level and time-invariant behaviour.
Balance contribution of the left and right body side
With FRFl,r the left or right FRF and FRFt the total FRF. The ● indicates the dot product of the FRFs. In this way the contribution of the left or right leg to the total balance control was expressed as a proportion. For example, a proportion of 0.8 for the left leg means that the left leg contributed for 80% to the total body stabilization. This was done for each separate MIMO FRF (see Eq. 2 and 3).
Table 1 shows the GOF values and NSRs of the different simulations. In case of no sensor and measurement noise, a platform acceleration in combination with a perturbation torque around the ankle (PLT-Nn) gave the same results as two perturbation forces (2F-Nn). In these conditions, the small GOF values indicated that the stabilizing mechanisms were well estimated.
Figure 3 shows the model transfer function and the estimated frequency response functions of the PLT-N and PL-N conditions. Clearly, applying two perturbations resulted in very good estimations (see also Table 1). However, applying only one perturbation to estimate the stabilizing mechanisms of a MIMO system resulted in incorrect estimates, although the responses were time-invariant as shown by the NSRs values. In other words, applying one perturbation resulted in biased estimates of the stabilizing mechanisms. Note that the GOF values of the PL-Nn and PL-N simulations were of the same magnitude.
Single-Input-Single-Ouput frequency response functions
Multiple-Input-Multiple-Output frequency response functions
In general, for the healthy controls, the relationship between the ankle joint angle and the ankle joint torque remained roughly constant over the whole frequency range, whereas the hip joint gain was high at the low frequencies (<0.7 Hz), low at the mid frequencies (0.7-2 Hz) and increased at higher frequencies (>2 Hz). The coupling between the ankle joint angle to hip torque increased with frequency, whereas the FRF remained roughly constant over the frequency range. The PD patient showed similar gain patterns for the and FRFs, but for the and FRFs the gains at the lower frequencies were much higher than those of the healthy controls, indicating an increased postural stiffness in the PD patient.
The phase of the and FRFs of both the HCs and the PD patient decreased with increasing frequency (i.e. a phase lag), indicating a neural time delay. The phase of the showed a phase shift of about +360° around 1.5 Hz. For the phase the largest difference between the HCs and PD patient is found in the FRF. Both groups showed a negative phase shift around 2 Hz, but this shift was larger in the PD patient.
Balance contribution of the left and right body side
Bipedal upright stance is multisegmental and requires the coordinated activity of multiple joints, including the ankles and hips. We developed a system identification method to investigate the balance control contribution of the ankles and hips of the left and right leg separately of individual subjects. To investigate both DoF and their interactions in a feedback loop, two perturbations are required. Therefore, in our lab we developed an actuated pusher placed on a motion platform to provide these independent perturbations. Using model simulations, it was demonstrated that these perturbations can identify the system reliably. The (clinical) applicability of the method was demonstrated in seven healthy controls and a PD patient.
Evaluation of the Multiple-Input-Multiple-Output identification method
The model simulations indicated that two independent perturbations are necessary to identify the stabilizing mechanisms of a two DoF MIMO system. Applying only one perturbation to the model resulted in biased estimates of the stabilizing mechanisms. This bias was not influenced by noise level. Two configurations with two independent perturbations were evaluated, that is, a combination of a platform acceleration and a disturbance torque around the ankle and two perturbation forces in the forward-backward direction. The stabilizing mechanism were estimated very well with both configurations, and the differences between the two approaches were small. After adding sensor and measurement noise to the simulation, the stabilizing mechanisms were still estimated well with both configurations, with a slightly worse result when perturbing with two forces. In short, the model simulations showed that the implemented MIMO identification method correctly identified the stabilizing mechanisms indicated by small differences between the theoretical transfer function of the model and the estimated frequency response functions, even in the presence of realistic levels of noise.
In the experiment with human participants, the perturbations were applied with a motion platform and a custom-made actuated backboard (i.e., the first configuration of the model simulations, which gave slightly better results). The amplitudes of the applied perturbations were easy to withstand, both for the healthy participant as for the PD patient, making the method suitable for use in a large range of participants.
To be clinically relevant, the MIMO identification method should be able to reliably estimate the different segmental contributions to the total balance control of a single participant. First, the quality of our estimation was expressed in the NSR. A low NSR indicates that a large percentage of the data is captured by the estimated stabilizing mechanisms. The NSR gives the ratio between the (periodic) response to the perturbations and the remnant. As the method assumes a LTI system, remnants can be due to a) nonlinearities of the perturbed system, b) time-variant system behavior, c) unmeasured system noise, and d) measurement noise . Note that a low NSR does not necessarily mean that the system is linear.
The average NSRs of the healthy controls of the ankle and hip joint angle and of the ankle and hip torque were 0.24 and 0.28, 0.56 and 0.35, respectively. This results in an average NSR of 0.36, meaning that about 74% (i.e., 1/(1+0.36)) of the response is captured by the estimated stabilizing feedback mechanisms. The PD patient had lower NSRs than the average healthy subject, indicating less variability over the repetitions of the perturbation signal. Unfortunately, it is impossible to compare our results with respect to reliability of the estimates with other studies using two perturbations as these studies did not report any parameters quantifying reliability and only report averages over participants.
Secondly, the used MIMO identification scheme consists of two perturbation rounds: in one round, the perturbations have the same sign, while in the other round, the perturbations have opposite signs. To determine whether participants did not change their balance control behaviour in both perturbation rounds, SISO FRFs for the ankle and hip joints were determined. Gain, phase, and joint torque coherence were comparable for both perturbation rounds; hence the participants did not change their balance control strategies. Note that the sway angle and ankle and hip coherence was lower in the second perturbation round at the lower frequencies (<0.7 Hz), indicating less time invariant behavior (possibly due to fatigue) and/or a higher noise level. This could be due to not randomizing the perturbation rounds. However, this did not result into quantitatively different behavior, as the SISO FRFs were similar for both perturbation rounds.
The last requirement of the method is that it is able to distinguish between the balance contributions of the separate legs to investigate balance asymmetry. By using a dual forceplate and calculating the FRFs for each leg separately, balance control asymmetries were detected in a PD patient.
Perturbing with a different set-up (for example two push-pull rods), at different locations or by using different signals could have elicited different responses. This is not an artifact of the method, but reflects the adaptability of the nervous system .
In sum, two unique features of the presented method are the applicability on the individual level and separation of balance contribution of each body side. The first was accomplished by applying multisine perturbation signals, which have the advantage of improving the estimation of FRFs because they concentrate signal power in a limited set of frequencies and are periodic. This results in reliable individual results and shorter measurement times [35, 46]. Secondly, by measuring the reaction forces of each foot with a dual forceplate balance control, asymmetries can be detected.
Comparison with other multivariate methods
Most published studies investigating the multivariate nature of balance control do not use perturbations, or use only one perturbation [15, 16, 18–20]. However, the model simulations indicated that two independent perturbations are required to estimate the stabilizing mechanisms of a multivariate balance control system. Using only one perturbation in the model simulations gave biased and erroneous results.
The other two available methods [2, 21] differ from the presented method in this paper. Our method is non-parametric, where Fujisawa  used a parametric method (ARMAX model structure). A parametric method has the advantage that it, theoretically, can better separate the measurement noise from the actual signals. However, knowledge about the structure of the system is required. A non-parametric method has the advantage that no prior knowledge of the system is needed . Despite the differences between the applied methods in  and in this manuscript the obtained FRFs are in the same range for the low frequencies. That is the FRF starts at unity gain which slightly increases with frequency, whereas the , and have lower gains at the lower frequencies (~0.05) but also increase with frequency. Note that we can compare the FRFs only in the low frequency range, as Fujisawa and colleagues  used a perturbation signal with frequencies up to 0.83 Hz.
Recently, Jeka and colleagues [2, 47] used an approach to investigate task goals for upright stance, similar to the one presented here. There are, however, a few fundamental differences. First of all, Kiemel et al.  used filtered white noise as a perturbation signal, whereas we used multisines, which have the advantage of improving the individual estimate of the stabilizing mechanisms. Also, they define the input of the stabilizing mechanisms as weighted EMG signals of the anterior and posterior body sides, whereas we have used joint torques. Therefore, the method presented by Jeka and colleagues [2, 47] focused on separating the reflexive from the total contribution of the stabilizing mechanisms by measuring EMG signals. This measurement and analysis of EMG signals can easily be added to our method. In order to detect balance control asymmetries, joint torques (which we have used in this study) are more suitable than EMG signals. Differences in EMG amplitude can for example, be due to different electrode placement on contralateral legs, different skin conductivity or due to different background activity, making these signals more prone to measurement artifacts. In addition, joint torques are more suitable for investigating joint stiffness.
Multisegmental balance control
With a MIMO method and by applying multiple perturbations, multisegmental balance control strategies and the interplay between the joints can be investigated. This approach can be used to test hypotheses about the role of the different joints, and also of sensory information .
It has been suggested that PD patients have a decreased intersegmental coordination [29, 30] or an increased hip stiffness [25, 28]. In this study, the most pronounced difference between the healthy controls and the PD patients was found in the FRF, that is, the PD patient had a higher gain at the lower frequencies, indicating an increased hip stiffness. With our method we can now distinguish between coordination between the upper and lower body and of coordination of the upper/lower body separately. Further investigation of intersegmental coordination could lead to a better understanding of the pathophysiology of balance impairments in PD and possibly improve intervention programs.
The PD patient asymmetrically controlled his balance with both the ankle and the hip joint. This means that one leg contributed more to body stability than the other leg. Balance control asymmetries have been shown before in PD patients, both during quiet stance [32, 49] and with a single DoF approach , but not taking into account the contribution of the hip joint. The balance asymmetry was also present in the intersegmental coordination, i.e. in the FRFs from ankle angle to hip torque and from hip angle to ankle torque. Hence, our new method has the advantage of assessing balance control asymmetries during perturbations, considering the role of the hip joint and the interplay between joints. This creates the possibility of assessing differences in balance control contribution between distal and proximal joints, and it can be investigated whether balance asymmetries influence intersegmental coordination. Further necessary research in a large group of PD patients and healthy matched controls should demonstrate the (potential) clinical value of the new method.
Furthermore, balance control can also be asymmetrical in stroke patients [4, 51]. During the recovering process, restoration of the paretic body side and/or compensation in the non-paretic body side may contribute to improved balance maintenance. Investigating balance control asymmetries provides the possibility of investigating different recovery and compensation strategies during the rehabilitation process.
We presented a new method to identify the multisegmental stabilizing mechanisms in human stance control using non-parametric system identification techniques and evaluated its performance. Model simulations showed that the newly presented method correctly and reliably estimated the balance control contribution of the ankle and hip joints and interactions between the segments. A balance control experiment showed the application in both healthy and pathological participants. Furthermore, the method can distinguish between the balance control contribution of each ankle and hip joint separately. Taken together, this can be used to create insights into the pathophysiology of postural instability and asymmetry in patients and possibly aid to develop and evaluate treatments.
Human balance control model
To test the feasibility of the MIMO identification method, model simulations were performed. A two-DoF mechanical model with a stabilizing mechanism was derived, Figure 8 shows the general model; the derivation of the subsystems is described below.
Derivation of the equations of motion
Kane’s method (TMT method) was used to derive the equations of motions. In this method, the principles of virtual power are used to rewrite the Newton-Euler equations.
The first step is to define the degrees of freedom (generalized coordinates, qt) of the model, which are in this case the support surface (Sx) and segment angles: θ1 and θ2.
With d1 and d2 the distance to the segment’s center of gravity from the distal point, l1 denotes the length of the leg segment. This matrix can be differentiated to the generalized coordinates to obtain the first (Tj,t) and second derivative (Tj,tl) of the transformation matrix.
With mf, m1, m2, I1 and I2, the mass of the foot, the segments masses and inertia’s respectively.
With g the gravitational acceleration (9.81m/s2), τp 1 = - Fp1 * p1 * sin(θ1), the perturbation torques due to the external perturbation force. Note that joint moments (τank and τhip) are the inputs of the model (forward simulation). Note, that the internal forces are not incorporated in the force matrix; these are implicitly incorporated in the Tj,t and Tj,tl matrices.
For which we have used the index notation.
This results in a split up of the Tj,t matrix in a part for the known (Tj,k) and unknown degrees of freedom (Tj,u) and differentiation of these matrices results in and , respectively. Note that the subscripts u and k denote the unknown and known coordinates, respectively.
Linearization and state space notation
Where a is the equilibrium point and x the deviation from the operating point. In our case, the equilibrium position is straight stance (segment angles are 90°, angular velocities, external perturbations, and joint torques are zero). Note that the centripetal term of the movement equations disappears with linearization.
With (thus the derivative of the systems states) and x the states. U are the inputs of the system and represent the joint torques, perturbation force, and the platform perturbation.
Human balance control model parameters
2.2 (Nm s/rad)
27.4 (Nm s/rad)
Time delay ankle
Time delay hip
Eigen frequency (ωn)
With x the system’s states and u the system’s input. Q and R are diagonal matrices; the elements in Q are set to 1 and in R set to 106.
As the sensory signals do not reach the CNS instantaneously (because of neural conduction times) a time delay is included as a pure transport delay, see Figure 8. Parameter values were taken from the literature , see Table 2.
Pink sensor noise was created off line, by scaling the power spectrum of a random timeseries by 1.2 such that at 1Hz, the power of the signal was 1.5*10-7. Subsequently, this signal was added to the joint angles in the model simulations; this lead to a spontaneous sway of around 0.6° (peak-to-peak amplitude; comparable with human quiet stance data). Measurement noise was modeled as white noise (zero mean and 0.0001 variance) and added to the joint angles and joint torques.
TB received the M.Sc. degree in human movement sciences of the Free University (Amsterdam, The Netherlands) in 2006. She is currently working towards her PhD degree at the laboratory of biomechanical engineering at the University of Twente, in close collaboration with the Radboud University medical Centre in Nijmegen. Her research focuses on developing new techniques and methods based on system identification to investigate and quantify human balance control. Concurrently, she is applying these new and adopted methods for patients with Parkinson’s disease in order to create more insight into the pathophysiology of this disease.
AC received the M.Sc. and Ph.D. degrees in mechanical engineering from the Delft University of Technology, The Netherlands, in 1999 and 2004, respectively. He holds a position as an assistant professor at the Delft University of Technology. He is co-founder of the Delft Laboratory for Neuromuscular Control. His research interest is the field of neuromuscular control and includes techniques to quantify the functional contribution of afferent feedback, neuromuscular modeling, haptic manipulators, and system identification. His research focuses on both able-bodied individuals and individuals suffering from movement disorders.
HvdK (1970, Rotterdam, the Netherlands) received his Phd with honors (cum laude) in 2000 and is professor in Biomechatronics and Rehabilitation Technology at the Department of Biomechanical Engineering at the University of Twente (0.8 fte), and Delft University of Technology (0.2 fte), the Netherlands. His expertise and interests are in the field of human motor control, adaptation, and learning, rehabilitation robots, diagnostic, and assistive robotics, virtual reality, rehabilitation medicine, and neuro computational modeling. Dr.ir. H. van der Kooij has published over 60 papers in the area of biomechatronics and human motor control, and has directed approximately € 6 million in research funding over the past 10 years. He is associate editor of IEEE TBME, member of IEEE EMBS technical committee of Biorobots and was member of several scientific program committees in the field of rehabilitation robotics, bio robotics, and assistive devices. At the UT he is founder and head of Rehabilitation robotics laboratory that developed powered exoskeletons for the rehabilitation of upper and lower extremities. He is founder and head of the Virtual Reality Human performance lab that combines robotic devices, motion capturing and virtual environments to asses and train human balance, walking, and hand-eye coordination.
Central nervous system
Auto regressive moving average eXogenous
Frequency response function
Single input single output
Multiple input multiple output
Head arms trunk
Degrees of freedom
Center of mass
Goodness of fit
Linear time invariant
United States of America
The authors gratefully acknowledge the support of the BrainGain Smart Mix Programme of the Netherlands Ministry of Economic Affairs and the Netherlands Ministry of Education, Culture and Science. We thank Jantsje Pasma and Denise Engelhart for their assistance with the measurements. We also want to acknowledge the constructive suggestions of the anonymous reviewers.
- Peterka RJ: Sensorimotor integration in human postural control. J Neurophysiol 2002,88(3):1097-118.PubMedGoogle Scholar
- Kiemel T, Zhang Y, Jeka JJ: Identification of neural feedback for upright stance in humans: stabilization rather than sway minimization. J Neurosci 2011,31(42):15144-53. 10.1523/JNEUROSCI.1013-11.2011PubMed CentralView ArticlePubMedGoogle Scholar
- Peterka RJ: Simplifying the complexities of maintaining balance. IEEE Eng Med Biol Mag 2003,22(2):63-8. 10.1109/MEMB.2003.1195698View ArticlePubMedGoogle Scholar
- van Asseldonk EH: Disentangling the contribution of the paretic and non-paretic ankle to balance control in stroke patients. Exp Neurol 2006,201(2):441-51. 10.1016/j.expneurol.2006.04.036View ArticlePubMedGoogle Scholar
- van Soest AJ, Rozendaal LA: The inverted pendulum model of bipedal standing cannot be stabilized through direct feedback of force and contractile element length and velocity at realistic series elastic element stiffness. Biol Cybern 2008,99(1):29-41. 10.1007/s00422-008-0240-2View ArticlePubMedGoogle Scholar
- Horlings CG: Vestibular and proprioceptive influences on trunk movements during quiet standing. Neuroscience 2009,161(3):904-14. 10.1016/j.neuroscience.2009.04.005View ArticlePubMedGoogle Scholar
- Creath R: A unified view of quiet and perturbed stance: simultaneous co-existing excitable modes. Neurosci Lett 2005,377(2):75-80. 10.1016/j.neulet.2004.11.071View ArticlePubMedGoogle Scholar
- Pinter IJ: The dynamics of postural sway cannot be captured using a one-segment inverted pendulum model: a PCA on segment rotations during unperturbed stance. J Neurophysiol 2008,100(6):3197-208. 10.1152/jn.01312.2007View ArticlePubMedGoogle Scholar
- Horak FB, Nashner LM: Central programming of postural movements: adaptation to altered support-surface configurations. J Neurophysiol 1986,55(6):1369-81.PubMedGoogle Scholar
- Horak FB, Macpherson JM: Postural orientation and equilibrium . In Handbook of Physiology: Section 12, Exercise: Regulation and Integration of Multiple Systems. Edited by: Shepard JT, Rowell LB. New York: Oxford University Press; 1996:255-292.Google Scholar
- Fitzpatrick R, Burke D, Gandevia SC: Loop gain of reflexes controlling human standing measured with the use of postural and vestibular disturbances. J Neurophysiol 1996,76(6):3994-4008.PubMedGoogle Scholar
- van der Kooij H, van Asseldonk E, van der Helm FC: Comparison of different methods to identify and quantify balance control. J Neurosci Methods 2005,145(1–2):175-203.View ArticlePubMedGoogle Scholar
- Pintelon R, Schoukens J: System identification, a frequency domain approach. Vol. 605. New York: IEEE Press; 2001.View ArticleGoogle Scholar
- de Vlugt E, Schouten AC, van der Helm FC: Closed-loop multivariable system identification for the characterization of the dynamic arm compliance using continuous force disturbances: a model study. J Neurosci Methods 2003,122(2):123-40. 10.1016/S0165-0270(02)00303-5View ArticlePubMedGoogle Scholar
- Kuo AD: Effect of altered sensory conditions on multivariate descriptors of human postural sway. Exp Brain Res 1998,122(2):185-95. 10.1007/s002210050506View ArticlePubMedGoogle Scholar
- Speers RA, Kuo AD, Horak FB: Contributions of altered sensation and feedback responses to changes in coordination of postural control due to aging. Gait Posture 2002,16(1):20-30. 10.1016/S0966-6362(02)00003-6View ArticlePubMedGoogle Scholar
- Alexandrov AV, Frolov AA, Massion J: Biomechanical analysis of movement strategies in human forward trunk bending. II. Experimental study. Biol Cybern 2001,84(6):435-43. 10.1007/PL00007987View ArticlePubMedGoogle Scholar
- Alexandrov AV: Feedback equilibrium control during human standing. Biol Cybern 2005,93(5):309-22. 10.1007/s00422-005-0004-1PubMed CentralView ArticlePubMedGoogle Scholar
- Park S, Horak FB, Kuo AD: Postural feedback responses scale with biomechanical constraints in human standing. Exp Brain Res 2004,154(4):417-27. 10.1007/s00221-003-1674-3View ArticlePubMedGoogle Scholar
- Kim S: Postural Feedback Scaling Deficits in Parkinson's disease. J Neurophysiol 2009.Google Scholar
- Fujisawa N: Human standing posture control system depending on adopted strategies. Med Biol Eng Comput 2005,43(1):107-14. 10.1007/BF02345130View ArticlePubMedGoogle Scholar
- Stolze H: Falls in frequent neurological diseases–prevalence, risk factors and aetiology. J Neurol 2004,251(1):79-84. 10.1007/s00415-004-0276-8View ArticlePubMedGoogle Scholar
- Pickering RM: A meta-analysis of six prospective studies of falling in Parkinson's disease. Mov Disord 2007,22(13):1892-900. 10.1002/mds.21598View ArticlePubMedGoogle Scholar
- Hely MA: The Sydney multicenter study of Parkinson's disease: the inevitability of dementia at 20 years. Mov Disord 2008,23(6):837-44. 10.1002/mds.21956View ArticlePubMedGoogle Scholar
- Grimbergen YA, Munneke M, Bloem BR: Falls in Parkinson's disease. Curr Opin Neurol 2004,17(4):405-15. 10.1097/01.wco.0000137530.68867.93View ArticlePubMedGoogle Scholar
- Boonstra TA: Gait disorders and balance disturbances in Parkinson's disease: clinical update and pathophysiology. Curr Opin Neurol 2008,24(4):461-471. 10.1097/WCO.0b013e328305bdafView ArticleGoogle Scholar
- Colnat-Coulbois S: Management of postural sensory conflict and dynamic balance control in late-stage Parkinson's disease. Neuroscience 2011, 193: 363-9.View ArticlePubMedGoogle Scholar
- Carpenter MG: Postural abnormalities to multidirectional stance perturbations in Parkinson's disease. J Neurol Neurosurg Psychiatry 2004,75(9):1245-54. 10.1136/jnnp.2003.021147PubMed CentralView ArticlePubMedGoogle Scholar
- Termoz N: The control of upright stance in young, elderly and persons with Parkinson's disease. Gait Posture 2008,27(3):463-70. 10.1016/j.gaitpost.2007.05.015View ArticlePubMedGoogle Scholar
- Maurer C: Effect of chronic bilateral subthalamic nucleus (STN) stimulation on postural control in Parkinson's disease. Brain 2003,126(Pt 5):1146-63.View ArticlePubMedGoogle Scholar
- Djaldetti R, Ziv I, Melamed E: The mystery of motor asymmetry in Parkinson's disease. Lancet Neurol 2006,5(9):796-802. 10.1016/S1474-4422(06)70549-XView ArticlePubMedGoogle Scholar
- Geurts AC: Assessment of postural asymmetry in mild to moderate Parkinson's disease. Gait Posture 2011,33(1):143-5. 10.1016/j.gaitpost.2010.09.018View ArticlePubMedGoogle Scholar
- van der Kooij H, Peterka RJ: Non-linear stimulus-response behavior of the human stance control system is predicted by optimization of a system with sensory and motor noise. J Comput Neurosci 2011,30(3):759-778. 10.1007/s10827-010-0291-yPubMed CentralView ArticlePubMedGoogle Scholar
- Robinson D: Series Elastic Actuator Development for a Biomimetic Robot, in IEEE/ASME International Conference on Advance Intelligent Mechantronics. 1999.Google Scholar
- van der Kooij H, de Vlugt E: Postural responses evoked by platform pertubations are dominated by continuous feedback. J Neurophysiol 2007,98(2):730-43. 10.1152/jn.00457.2006View ArticlePubMedGoogle Scholar
- Zajac FE: Understanding muscle coordination of the human leg with dynamical simulations. J Biomech 2002,35(8):1011-8. 10.1016/S0021-9290(02)00046-5View ArticlePubMedGoogle Scholar
- Nott CR: All joint moments significantly contribute to trunk angular acceleration. J Biomech 2010,43(13):2648-52. 10.1016/j.jbiomech.2010.04.044PubMed CentralView ArticlePubMedGoogle Scholar
- Pasma JH: Sensory reweighting of proprioceptive information of the left and right leg during human balance control. J Neurophysiol 2012,108(4):1138-48. 10.1152/jn.01008.2011View ArticlePubMedGoogle Scholar
- Koopman HFJM: The three-dimensional analysis and prediction of human walking. The Netherlands: University of Twente; 1989.Google Scholar
- Koopman B, Grootenboer HJ, de Jongh HJ: An inverse dynamics model for the analysis, reconstruction and prediction of bipedal walking. J Biomech 1995,28(11):1369-76. 10.1016/0021-9290(94)00185-7View ArticlePubMedGoogle Scholar
- Brand RA: A model of lower extremity muscular anatomy. J Biomech Eng 1982,104(4):304-10. 10.1115/1.3138363View ArticlePubMedGoogle Scholar
- Chandler RF: Investigation of the intertial properties of the human body. 1975.Google Scholar
- Preuss R, Fung J: A simple method to estimate force plate inertial components in a moving surface. J Biomech 2004,37(8):1177-80. 10.1016/j.jbiomech.2003.12.007View ArticlePubMedGoogle Scholar
- Jenkins G, Watts D: Spectral Analysis and Its Applications. San Francisco: Holden Day; 1969.Google Scholar
- Kim S, Atkeson CG, Park S: Perturbation-dependent selection of postural feedback gain and its scaling. J Biomech 2012,45(8):1379-86. 10.1016/j.jbiomech.2012.03.001View ArticlePubMedGoogle Scholar
- de Vlugt E, Schouten AC, van der Helm FC: Quantification of intrinsic and reflexive properties during multijoint arm posture. J Neurosci Methods 2006,155(2):328-49. 10.1016/j.jneumeth.2006.01.022View ArticlePubMedGoogle Scholar
- Kiemel T, Elahi AJ, Jeka JJ: Identification of the plant for upright stance in humans: multiple movement patterns from a single neural strategy. J Neurophysiol 2008,100(6):3394-406. 10.1152/jn.01272.2007PubMed CentralView ArticlePubMedGoogle Scholar
- Kiemel T, Zhang Y, Jeka JJ: Visual flow is interpreted relative to multisegment postural control. J Mot Behav 2011,43(3):237-46. 10.1080/00222895.2011.568991View ArticlePubMedGoogle Scholar
- Rocchi L, Chiari L, Horak FB: Effects of deep brain stimulation and levodopa on postural sway in Parkinson's disease. J Neurol Neurosurg Psychiatry 2002,73(3):267-74. 10.1136/jnnp.73.3.267PubMed CentralView ArticlePubMedGoogle Scholar
- van der Kooij H: Detecting asymmetries in balance control with system identification: first experimental results from Parkinson patients. J Neural Transm 2007,114(10):1333-7. 10.1007/s00702-007-0801-xView ArticlePubMedGoogle Scholar
- Roerdink M: On the relative contribution of the paretic leg to the control of posture after stroke. Neurorehabil Neural Repair 2009,23(3):267-74.View ArticlePubMedGoogle Scholar
- Winter DA: Biomechanics and motor control of human movement. 2nd edition. New York: John Wiley & Sons Inc; 1990.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.