You are invited to attend the public defence of my thesis
heads up
SenSorimotor control of the head-neck SyStemon 7 March 2014
at exactly 12.30 in the Senaatszaal of the Aula Congress Centre Delft University of Technology,
Mekelweg 5, Delft, The Netherlands.
At 12.00 I will present a brief overview of my research.
After the defence you are welcome to join a reception in Aula.
Patrick Alan Forbes
06 14458462 p.a.forbes@tudelft.nl
Paranymphs
Alistair Vardy 06 10308118 a.n.vardy@tudelft.nl Thomas Edwards 06 22700735 thomas.edwards@ipsos.comHEADS UP
SENSORIMOTOR CONTROL
OF THE HEAD-NECK SYSTEM
PATRICK ALAN FORBES
Proefschrift
ter verkrijging van de graad van doctor aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus prof. ir. K.C.A.M. Luyben, voorzitter van het College voor Promoties,
in het openbaar te verdedigen op vrijdag 7 maart 2014 om 12:30 uur
door
Patrick Alan FORBES
Master of Applied Science in Mechanical Engineering geboren te Oakville, Ontario, Canada
The research described in this thesis was performed within NeuroSIPE (System Identification and Parameter Estimation for Neurophysiological Systems), a program with the goal to develop new diagnostic tools for neurological disorders. This program was supported by the Dutch Technol-ogy Foundation (STW), which is part of the Netherlands Organization for Scientific Research (NWO) and partly funded by the Ministry of Economic Affairs, Agriculture and Innovation.
ISBN: 978-94-6186-274-7
Neck muscle and vestibular organ illustrations: Kathryn A. Forbes (based on illustration from Gray's Anatomy of the Human Body)
Cover and book design: IS Ontwerp / Ilse Schrauwers ~ www.isontwerp.nl Printing: Ridderprint, Ridderkerk ~ www.proefschriftdrukken.nl
©
Patrick Alan Forbes 2014All rights reserved. No part of this book may be reproduced by any means, or transmitted with-out the written permission of the author. Any use or application of data, methods and/or results etc., occurring in this report will be at the user’s own risk.
The printing of this thesis was additionally supported by Twente Medical Systems International B.V. and TNO.
HEADS UP
SENSORIMOTOR CONTROL
OF THE HEAD-NECK SYSTEM
HEADS UP
Dit proefschrift is goedgekeurd door de promotor: Prof. dr. F.C.T. van der Helm
Copromotoren: Dr. ir. R. Happee Dr. ir. A.C. Schouten
Samenstelling promotiecommissie:
Rector Magnificus, Technische Universiteit Delft, voorzitter Prof. dr. F.C.T. van der Helm Technische Universiteit Delft, promotor Dr. ir. R. Happee Technische Universiteit Delft, copromotor Dr. ir. A.C. Schouten Technische Universiteit Delft, copromotor Prof. dr. W.P. Medendorp Radboud Universiteit Nijmegen
Prof. dr. J. van der Steen Erasmus Medische Centrum
Prof. dr. ir. P.M.J. Van den Hof Technische Universiteit Delft/Eindhoven Dr. R.J. Peterka Oregon Health and Science University Prof. dr. ir. H. van der Kooij Technische Universiteit Delft, reservelid
Nil bastardo carborundum
HEADS UP Contents
PART III
Neurophysiology of vestibulocollic reflexes
153
Chapter 8
Frequency response of vestibular reflexes in neck, 155
back and lower limb muscles
Chapter 9
The effect of balance task and descending motor 179
command on vestibulocollic reflexes
Chapter 10
Conclusions, discussion and future directions 199
Appendix
Neuromuscular modelling of head-neck stabilization 215
References 227 Summary 251 Samenvatting 259 Acknowledgements 267 Curriculum Vitae 273 Publications 277
III
8
9
CONTENTS
Contents 6 List of abbreviations 9 Chapter 1 Introduction 11PART I
Experimental methods to investigate sensorimotor postural control
29
Chapter 2EMG feedback tasks reduce reflexive stiffness 31
during force and position perturbations
Chapter 3
Electrical vestibular stimuli to enhance vestibulo-motor 55 output and improve subject comfort
Chapter 4
Galvanic vestibular stimulation elicits consistent 73
head-neck motion in seated subjects
PART II
Modulation of neck reflexes for head-neck sensorimotor control
87
Chapter 5Dependency of human neck reflex responses on the bandwidth 89 of pseudorandom anterior-posterior torso perturbations
Chapter 6
Head-neck stabilization in cervical dystonia: 113
task dependent impairment of reflex modulation
Chapter 7
Vestibulocollic reflexes are modulated to amplitude 131
and bandwidth of lateral torso perturbations
I
II
2
5
3
6
4
7
HEADS UP List of abbrevations
LIST OF ABBREVIATIONS
ANOVA Analysis of Variance
AP Anterior-Posterior
ARCL Anode Right Cathode Left
CCR Cervicocollic Reflex
CD Cervical Dystonia
CNS Central Nervous System
DOF Degree of Freedom
EC Eyes Closed
EMG Electromyography
ESP Erector Spinae
EO Eyes Open
ET Electromyography Task
FFT Fast Fourier Transform
FP Force Perturbation
FRF Frequency Response Function
FT Force Task
GAS Gastrocnemius
GVS Galvanic Vestibular Stimulation
LVNn Linear Vestibular Nucleus Neurons
MUAP Motor Unit Action Potential
MVC Maximum Voluntary Contraction
MVNn Medial Vestibular Nucleus Neurons
MVS Multisine Vestibular Stimulation
NP No Perturbation
NV Non-Vestibular
PP Position Perturbation
PT Position Task
RMS Root Mean Square
SCM Sternocleidomastoid
SD Standard Deviation
SEMI Semispinalis Capitis
SNR Signal-to-Noise Ratio
SOL Soleus
SPL Splenius Capitis
SVS Stochastic Vestibular Stimulation
VAF Variance Accounted For
VAS Visual Analogue Scale
VCR Vestibulocollic Reflex
CHAPTER 1
HEADS UP Introduction CHAPTER 1
HEADS UP
Unless you are reading this thesis to help you sleep, you likely have your head upright. Whether standing or sitting, your neck muscles maintain the posture of your head in order to give your eyes a stable platform to read. Now that you are aware of the state of your head and neck, try to relax your neck muscles. Imagine yourself as a sleepy train passenger making the long journey home, your head creeping slowly down from upright as you drift into dreamland. Let your head sway left to right or forward and backward on top of the inverted pendulum that is your neck. Suddenly, BAM! An abrupt movement by the train jolts you awake, and just as quickly your neck instinctively returns to its vigilant state of maintaining a stable head posture.
The central nervous system (CNS) coordinates neck muscles to maintain stable head posture and generate volitional head movements. Visual and vestibular systems, together with neck somatosensory inputs are integrated by the CNS to provide an estimate of both the external world and an internal representation of ourselves. When our external (or internal) environments change, for example through unexpected or unpredictable disturbances, neck muscles make nimble corrections, through CNS control, to ensure the head remains upright and stable. As a weary commuter, you react swiftly to the train shifting unexpectedly over the tracks, a task that seemingly requires little effort. However, for patients suffering from neck sensorimotor disor-ders, stable head posture is a constant challenge, often leading to complaints of neck pain and fatigue. This thesis investigates the contribution of different sensory feedback mechanisms – with particular focus on vestibular and proprioception sensory systems – to natural head-neck stabilization. The experimental methods developed and the research outcomes obtained from studying healthy controls are in turn used to investigate patients suffering from the neck move-ment disorder, cervical dystonia. This first chapter provides the necessary background on senso-rimotor control of head-neck stabilization such that the goals of the research can be formulated.
SENSORIMOTOR CONTROL OF HEAD-NECK STABILIZATION
Head-neck stabilization is inherently challenging even when stationary, requiring constant vigi-lance to counter the downward pull of gravity. It involves a highly complex biomechanical sys-tem comprised of a large mass (the head) balanced on top of seven vertebrae (the neck), that are in turn connected to a moving base (the torso). Amazingly this multi-degree-of-freedom system is controlled by an equally impressive array of more than 30 bilateral muscle pairs that connect across various combinations of the skull, vertebrae and thorax. Multiple sensory feed-back systems throughout the head and neck provide information regarding movement such that the descending motor commands delivered to neck muscles generate appropriate and compensatory multidirectional forces and movements to sustain upright head posture.
Two particular reflex loops are thought to act as primary contributors to head-neck stabilization and are the primary focus of this thesis: the vestibulocollic reflex (VCR) and the cervicocollic
HEADS UP Introduction CHAPTER 1 Head-neck system (intrinsic properties) Motor circuits Premotor pathways Vestibulocollic pathways Muscle spindles Vestibular organ External perturbations on torso Electrical vestibular stimulation Spinal cord Voluntary head movement Head-in-space Head-on-torso (neck) VCR CCR _ + + + + + + + + _ _ Command Efference copy
Vestibulocollic reflexes – the vestibular system
The vestibular system is a bilateral set of inertial sensors comprised of elaborate interconnected chambers – the labyrinth – that detect head motion in space to provide a perception of self-movement and as well as the orientation of the head relative to gravity. Each labyrinth is located deep in the temporal bone and consists of three orthogonal semicircular canals and two or-thogonal otolith membranes. The semicircular canals (the horizontal, anterior and posterior) sense head rotation velocity and acceleration, and the two otoliths sense head translation accel-eration and head tilt. The two labyrinths are mirrored copies of one another, whereby two canals – one on either side – work in concert as antagonistic functional pairs. For example, within the canals, head rotation movement causes an increased firing rate of vestibular afferents on one side and a decreased firing rate on the opposite side.
Vestibular signals originate from primary afferents and are transferred via the vestibular branch of the eighth cranial nerve to the vestibular nuclear complex for signal processing and rerouting. At this location, vestibular signals are integrated with visual, proprioceptive and other sensory information to estimate head and body orientation (red pathway in Figure 1). The vestibular nu-clear complex shares reciprocal connections with the cerebellum, allowing the CNS to monitor and modulate vestibular reflex contributions to posture control. The vestibular nuclear complex is located within the brainstem and consists of four major nuclei: superior, medial, lateral and inferior (also called the spinal or descending). Although no explicit separation of afferent input occurs across each nucleus, there is a general pattern of function related to the source and destination of the signals being transferred. Of importance for this thesis, the medial nucleus processes information related to neck and eye control and projects via the medial vestibulospi-nal and reticulospivestibulospi-nal tracts to muscles primarily at the cervical levels.
The medial vestibulospinal tract contains the majority of pathways driving the VCR. These path-ways (dark green pathway in Figure 1) are formed by direct three-neuron arcs (vestibular afferent neurons, vestibular nucleus neurons and neck motoneurons) (Wilson and Yoshida 1969b, Wilson and Maeda 1974, Shinoda et al. 2006) and generate response latencies of 8-10 ms (Watson and Colebatch 1998, Rosengren et al. 2010). Typical neck motoneurons receive input from canals and otoliths from both sides. As a result, when the vestibular organ is activated by head mo-tion, neck muscles are capable of generating compensatory reflexes in response to a combined stimulus in the same plane (Suzuki and Cohen 1964). In addition to the direct three-neuron-arc pathways, there is convincing evidence that more indirect pathways also contribute to the reflex response (Miller et al. 1982, Thomson et al. 1995). These pathways are thought to be mediated by the reticulospinal system (Peterson et al. 1980) and neuroanatomical studies support this hy-pothesis since branches from single vestibulospinal and reticulospinal fibres similarly innervate motoneurons from multiple muscles (Shinoda et al. 1996, Shinoda et al. 2006). This innervation pattern further implicates the likelihood of muscle synergies to coordinate and simplify neck muscle control (Sugiuchi et al. 2004).
reflex (CCR) (Keshner 2009). The VCR is driven by activity of the vestibular organ to stabilize the head in space, while the CCR is driven by the activity of neck muscle proprioceptors to stabilize the head on the torso. These two reflex systems combine with voluntary head movements and intrinsic properties to make up the mechanisms contributing to head-neck stabilization control as shown in Figure 1. The next two sections provide brief overviews of vestibular and proprio-ception physiology, as well as our current knowledge on the dynamic characteristics of VCR and CCR, and the section thereafter describes their interaction as related to head-neck stabilization.
Figure 1 also depicts the influence of voluntary head movement on the vestibulocollic path-ways. In voluntary movements, vestibular signals and the VCR are thought to be cancelled at the earliest stages of neural processing since its action would oppose the intended movement (Ezure and Sasaki 1978, Cullen and Roy 2004). During unexpected and unpredictable distur-bances, such as those considered in this thesis, these mechanisms are not likely to affect the VCR and are therefore not discussed here. In the last chapter of this thesis, the effects that voluntary head movements have on vestibular reflexes are discussed further as possible future directions of study.
FIGURE 1
Conceptual diagram of the closed-loop sensory feedback mechanisms contributing to head-neck control. This includes the vestibulocollic reflex (VCR), which is driven by vestibular sensors, and the cervicocollic reflex (CCR), which is driven by proprioception (muscle spindle) sensors. External mechanical perturbations are applied via the torso and electrical vestibular stimulation is applied to the vestibular afferent to modulate firing rates. Vestibulocollic pathways integrate information from both sensory modalities and voluntary mechanisms to generate a desired response that is likely context dependent. Figure adapted from Peterson and Boyle (2004).
HEADS UP Introduction CHAPTER 1
nuclear bag fibres. γ-motoneurons allow the CNS to regulate the sensitivity of the intrafusal muscle fibres. Neck muscles possess a particularly high density of muscle spindles suggesting a high degree of sensitivity for relatively small cervical stimuli (Richmond and Abrahams 1975). The most direct pathways mediating the CCR are from the muscle spindles to the spinal cord and back to the muscle, i.e., short latency monosynaptic spinal reflexes (light green pathway in Figure 1). Although no direct measure of the reflex latency is available in the literature, it is estimated to be between 10-20 ms based on modelling studies examining the dynamic proper-ties of the CCR (Peterson et al. 1985). In addition to monosynaptic pathways, ascending tracts transmit muscle spindle information for processing in higher brain centres and also contribute to longer latency reflex responses. These supraspinal pathways are thought to play a role in the production of the CCR (Goldberg et al. 2012) however their contribution in relation to spinal pathways remains to be determined.
Similar to the VCR, the dynamic properties of the CCR have been well characterized in animal prep-arations (Dutia and Hunter 1985, Peterson et al. 1985). Under these conditions, neck muscle activity is measured while the torso is rotated around various rotation axes. Figure 2B plots muscle activ-ity as a result of neck rotational acceleration. The dynamic behaviour of the CCR is similar to VCR (decreasing gain and increasing phase) with sensitivity to acceleration at high frequencies. At low frequencies however, muscles respond primarily to position (phase = -180°) rather than velocity.
10−2 10−1 100 101 10−2 10−1 100 101 10−1 10−3 101 10−1 10−3 101 Gain [−] −135 −180 −45 −90 0 −135 −180 −45 −90 0 Phase [°] Frequency [Hz] Frequency [Hz]
Open Loop VCR (θH - EMG)
A B
Ezure and Sasaki, 1978 Berthoz and Anderson, 1971 Dutia and Hunter, 1985 Bilotto et al., 1982
Open Loop CCR (θN- EMG)
.. ..
Peterson, 1981 Dutia and Hunter, 1985
Currently, our understanding of how the VCR functions during head-neck stabilization origi-nates primarily from decerebrate animal studies, in which its open-loop (i.e., head motion-to-neck muscle activity) dynamic properties are well established (Berthoz and Anderson 1971, Ezure and Sasaki 1978, Bilotto et al. 1982, Dutia and Hunter 1985). Figure 2A plots the muscle activity as a result of input rotational head acceleration. The VCR response sensitivity is close to velocity (phase = -90°) and position (phase = -180°) at low frequencies (< 1 Hz) and acceleration at high frequencies (phase = 0°). Arguably, the animal models used to describe these responses may not be representative of the VCR in humans, particularly given that the majority of these studies use quadrupeds (typically cats) while humans are exclusively bipedal. However, recent studies examining the VCR in monkeys, a more human-like animal model, have reported similar responses (Sadeghi et al. 2011).
In humans, VCR responses are typically evoked using direct vestibular stimuli such as head taps, electrical stimulation or loud-clicks (Rosengren et al. 2010) (see Research Methods for more de-tail). These stimuli generate short duration reflex responses being < 20 ms (Colebatch et al. 1994, Murofushi et al. 2002, Lee et al. 2008). Short duration responses indicate that high frequencies contribute to the response since a duration of 20 ms corresponds to a frequency of 50 Hz. Con-tributions at frequencies > 20 Hz however are in contrast to current estimates of the dynamic range of vestibular function (0-20 Hz) (Armand and Minor 2001, Huterer and Cullen 2002). Thefore, uncertainty remains regarding the properties of VCR, primarily because the frequency re-sponses have yet to be characterized beyond 10 Hz in animals or humans.
Cervicocollic reflexes – proprioception
Proprioception is the sensation of movement, position and internal forces originating from re-ceptors within the body. Propriore-ceptors include muscle spindles, Golgi tendon organs, joint sensors, skin receptors and strictly speaking the vestibular system. The predominant proprio-ceptive reflex mechanisms within the neck, with the exception of the vestibular system, are me-diated by muscle spindles and Golgi tendon organs. However, the muscle spindles are regarded as the primary receptors responsible for CCR response (Peterson et al. 1985).
Muscle spindles are in parallel with muscle fibres, making their stretch (and stretch velocity) proportional to that of the muscle. Muscle spindles are composed of intrafusal muscle fibres, of which there are three kinds: dynamic nuclear bag fibres, static nuclear bag fibres and nuclear chain fibres. Two types of afferent fibres originate from the muscle spindle: primary (Ia) and secondary (II) endings. Primary endings are most sensitive to stretch velocity and make contact with all three intrafusal fibre types. Secondary endings are most sensitive to stretch and make contact with static nuclear bag and nuclear chain fibres. In addition to afferent fibres, efferent motor fibre endings terminate on all three intrafusal fibre types. Referred to as γ-motoneurons, they are composed of two types: static γ-motoneurons which innervate static nuclear bag fibres and nuclear chain fibres, and dynamic γ-motoneurons which innervate only the dynamic
FIGURE 2
Open loop gain- and phase-frequency response functions for the vestibulocollic (A) and cervicocollic reflexes (B) as mea-sured in cats. Points are experimental data from various references. Units for gain are arbitrary but depict muscle activity as a result of head acceleration (θH) for VCR and neck acceleration (θN) for CCR (adapted from Peng et al. 1996).
HEADS UP Introduction CHAPTER 1
neck system. Moreover, because the CCR interacts substantially with the VCR pathways, their combined modulation may result in an observable change in stabilization strategy.
RESEARCH METHODS
System identification for head-neck stabilization
The primary objective of studying human posture control is to understand how the CNS modu-lates posture to account for changes in tasks, conditions or disturbances. For these purposes system identification techniques are commonly used (Keshner et al. 1995, Fitzpatrick et al. 1996, Kearney et al. 1997, van der Helm et al. 2002) where a model of the system (joint dynamics) is ob-tained by analysing the causal dynamic relationship between input (position, force or current) and output (force, position and electromyography) signals. Frequency response functions (FRFs) describe the magnitude and timing of the output signal with respect to the input signal (i.e., gain and phase) and provide a comprehensive description of the linear component of system dynamics in the frequency domain.
The challenge in analysing neuromusculoskeletal systems is that sensory feedback makes them inherently closed-loop. When analysing the total system input-output behaviour, for example using kinematic measurements, the FRFs unavoidably include effects of the associated feedback pathways (see Figure 3 dark grey box). Therefore kinematic FRFs alone cannot be used to under-stand the reflexive properties within the loop. To overcome this limitation, the joint input-output approach (van der Kooij et al. 2005) is often employed to provide an inferred open-loop esti-mate (Fitzpatrick et al. 1996) of the reflex feedback pathways. With this analysis approach, output muscle activity is correlated with the input head motion to provide an estimate of the reflexive dynamics (see Figure 3 light grey box). These two methods of estimating FRFs (i.e., kinematic and reflexive) are used throughout this thesis to investigate reflexive head-neck stabilization.
To estimate FRFs the system must be perturbed via an independent input. Perturbation signals can be classified as transient or continuous. Transient signals reveal the state of the system at the onset of the perturbation and can be easily analysed in the time domain. However, subject re-sponses from transient inputs may be confounded by voluntary mechanisms on account of the repetitive and predictive nature of these perturbations. Continuous signals on the other hand have several advantages over transients. They can be designed to be unpredictable, thereby mini-mizing voluntary behaviour and isolating reflexive responses. During continuous perturbations, humans adapt the controller for optimal posture maintenance, which is thought to occur with a few seconds into the perturbations and maintained throughout (van der Helm et al. 2002). As a result, the perturbations can be designed to evoke modulation of reflex contributions by modify-ing the specific frequency content in the signal (Mugge et al. 2007, Schouten et al. 2008b). Usmodify-ing continuous perturbations to estimate FRFs inherently assumes that the system is linear, while the human neuromusculoskeletal system is highly nonlinear. As a result, continuous perturbations
Vestibulocollic and cervicocollic interaction
In decerebrate cats, the separate contributions of VCR and CCR are known to combine linearly (Peterson et al. 1985, Dutia and Price 1987) and their interaction can be synergistic or antago-nistic depending on the nature of the movement. During externally imposed rightward (yaw) rotation of the head on a stationary body, the VCR evokes neck muscle activity to produce a counter-rotation to the left. The rotation stretches left-directed muscles and the CCR evokes a synergistic leftward head rotation. In contrast, when the body is rotated to the right in yaw with the head free, CCR will generate a rightward head rotation, which will be opposed by an antagonistic VCR evoked counter rotation to the left. Therefore, the combined function of the VCR and CCR is likely context dependent and modulated by the CNS for a desired outcome (Reynolds et al. 2008).
The modulation of neck reflexes for example is known to occur with changes in head inertia (Goldberg and Peterson 1986, Keshner et al. 1999, Reynolds et al. 2008) and mental set (Kesh-ner 2000), while perturbation amplitude has limited effects on system dynamics (Kesh(Kesh-ner et al. 1995). In humans, reflexive stabilization is thought to provide damping mechanisms for the underdamped mechanics of the head (Keshner et al. 1995, Keshner and Peterson 1995, Gold-berg and Cullen 2011). Numerical models support this hypothesis and indicate that VCR and CCR work together to decrease resonance effects of the passive head -neck mechanics (Peng et al. 1996). The relative contribution of each reflex however depends on the direction of motion: VCR dominates the stabilization response during torso rotations about the horizontal plane (i.e., yaw) (Peng et al. 1996) while both VCR and CCR contribute equally to damping during torso rotation about the vertical plane in the sagittal direction (Peng 1996).
Although the goal of head stabilization will generally be to keep the head upright, there may be conditions where maintaining a head-in-space stabilization tendency is more appropriate than a head-on-torso stabilization tendency. For example, during torso disturbances with the eyes open, a head-in-space strategy is more effective when maintaining visual focus on an earth-fixed target (Goldberg and Cullen 2011) and may occur through an increased reliance on VCR. Indeed, an improved head-in-space stabilization is observed with the eyes open during per-turbations exciting frequencies below resonance (Guitton et al. 1986). On the other hand, with the eyes closed there is no visual requirement to maintain stable head posture in space and a head-on-torso strategy may be more useful, thereby shifting emphasis to CCR contributions.
The properties of a disturbance may also have an effect on reflex contributions. Disturbances that exceed the resonance of neuromusculoskeletal systems, such as the wrist or ankle, evoke a reduction in proprioception reflex contributions (Kearney et al. 1997, Mirbagheri et al. 2000, Schouten et al. 2008b). This is thought to prevent oscillatory behaviour due to the inherent reflex time delays (van der Helm et al. 2002, Schouten et al. 2008b). Since the CCR is driven by the same proprioception sensory mechanisms, reflex modulation may be similarly observed in the
head-HEADS UP Introduction CHAPTER 1
Electrical vestibular stimulation
Electrical vestibular stimulation is a non-invasive experimental technique that is commonly used to probe human vestibular function during posture control. Electric current is applied us-ing electrodes placed over the mastoid processes behind each ear. The applied current results in changes in the firing rate of the vestibular nerve (Goldberg et al. 1984). By controlling the direction of electric current, the afferent firing rate can be increased (anode) or decreased (cath-ode) depending on the pole of the electrode (cathode/an(cath-ode) at the stimulation site. The most common arrangement of electrical vestibular stimulation (and that used in this thesis) is binaural bipolar: electrodes with opposite poles are placed over each mastoid process causing a bilateral but opposing response in the afferent firing rates on either side of the head.
The modulation of afferent firing rates caused by the stimulus induces an artificial sense of mo-tion. The perceived motion is accompanied by compensatory muscular responses and whole-body postural adjustments (Nashner and Wolfson 1974, Britton et al. 1993, Fitzpatrick and Day 2004). These responses are predictable and have been modelled by Fitzpatrick and Day (2004) are typically designed to evoke small deviations around an equilibrium point, so as to minimize
nonlinear contributions and maintain the assumption of linearity. For the purposes of this thesis, perturbation signals were applied almost exclusively as continuous signals and always around a constant equilibrium point.
The continuous signals used throughout this thesis were subsequently classified as mechani-cal or sensory. In general, mechanimechani-cal perturbations are applied to the skeleton as either forces or positions, indirectly evoking sensory feedback mechanisms that reflect conditions humans encounter on a daily basis. Sensory perturbations on the other hand are applied independent of the mechanics and directly affect the afferent output of the sensor. Although sensory perturba-tions are not natural stimuli, they provide a means to examine the role a single sensor (in this case the vestibular organ) plays in muscular and posture control. Both perturbation types are shown in Figure 1 and Figure 3, where the sensory perturbation used in this thesis was electrical vestibular stimulation.
Mechanical perturbations
Continuous mechanical perturbations can be applied either directly to the head or via the torso (see Figure 1 and Figure 3). Compared to torso perturbations, direct head perturbations are argu-ably less representative of experiences in daily life. As a result, few studies have used this approach and those that do typically aim to understand the mechanical properties of the system rather than the neural control maintaining stabilization (Ono et al. 2003, Tangorra et al. 2003, Simoneau et al. 2008). Only one study was found that used direct head force perturbations to investigate head-neck stabilization dynamics (Viviani and Berthoz 1975). These authors suggested that active modulation of neck muscle properties by the CNS could provide the torques necessary to coun-ter the perturbation. An experimental arrangement involving direct head perturbations however requires subjects to maximize neck stiffness in an effort to minimize head motion on the torso. This is likely a different strategy from that used during unexpected torso disturbances, where for example during visual fixation tasks it may be more beneficial to minimize head motion in space.
Perturbations via the torso have been used extensively to investigate the mechanisms controlling head-neck stabilization, including voluntary movement, reflexive feedback and system mechanics (Keshner et al. 1995, Keshner and Peterson 1995, Keshner 2000, Keshner 2003). To isolate reflexive mechanisms subjects are asked to perform mental arithmetic or listen to distracting audio broad-casts while performing a natural stabilization task with the eyes open or closed. Subjects may also be instructed to perform particular tasks including natural stabilization, neck muscle co-contrac-tion or voluntary visual tracking. Comparison across the various tasks helps identify the separate mechanisms responsible for head-neck control. For example, by comparing the natural stabili-zation of elderly subjects to young healthy subjects co-contracting their neck muscles, Keshner (2000) was able to demonstrate that elderly subjects prefer a co-contraction strategy to maintain head-neck stabilization, likely due to an age related decrease in reflex function (Keshner 2000).
Neck muscles Head-neck load
Neck reflexes Vestibular
pathways Motor
command EMG Headmotion
Mechanical perturbation Electrical vestibular stimulation FIGURE 3
Block diagram of the closed-loop neuromusculoskeletal feedback system. The dark grey box depicts the potential esti-mate of system dynamics including closed-loop feedback pathways (i.e., input perturbation/stimulation-to-output head motion). The light grey box is an example of the inferred open-loop estimate of the reflex feedback pathways (i.e., input head motion-to-output muscle activity-EMG).
HEADS UP Introduction CHAPTER 1
CERVICAL DYSTONIA (TORTICOLLIS)
Cervical dystonia (CD) or spasmodic torticollis is a movement disorder, which is characterized by involuntary sustained muscle contractions resulting in abnormal movement or posture and impaired head movement control. CD seriously affects ~7500 people in the Netherlands and the estimated prevalence (diagnosed and undiagnosed) of the disorder in the United States is 0.390% (or 1.17 million people) (Jankovic et al. 2007). Although the pathophysiology of the disor-der is unclear, CD is generally regarded as a motor disordisor-der caused by disturbances in the corti-cal motor loop (Berardelli et al. 1998). In particular, dysfunction in the basal ganglia is thought to be the primary origin (Berardelli et al. 1998, Tijssen et al. 2000, Hallett 2006, Foncke et al. 2007, Breakefield et al. 2008); however, other brain regions including the cerebellum, thalamus, mid-brain and cerebral cortex may be involved in the pathophysiology of the disorder (Neychev et al. 2011, Lehericy et al. 2013).
However, the frequent involvement of abnormal sensory system features suggest that CD is a sensorimotor disorder (Abbruzzese and Berardelli 2003, Kanovsky et al. 2003, Breakefield et al. 2008). For example, disturbed vestibular function is observed under several contexts, viz. a directional preference of the vestibular nystagmus response (Bronstein and Rudge 1986, Stell et al. 1989), delayed responses during combined vestibular and voluntary neck muscle control (Munchau et al. 2001), and altered postural responses during stance under normal conditions (Vacherot et al. 2007) and while applying neck muscle vibration (Lekhel et al. 1997). Similarly, disturbed proprioceptive integration is observed in CD patients, viz. inaccurate knowledge of head posture (Anastasopoulos et al. 2003), trajectory abnormalities during arm reaching tasks (Pelosin et al. 2009), and orienting abnormalities during quiet standing and dynamic stepping movements (Bove et al. 2004, Bove et al. 2007). While these studies implicate CD as a disorder of sensorimotor integration, the effects of CD on the supraspinal modulation of afferent feedback for head-neck stabilization remain unknown.
NEURO-SIPE: TORTICOLLIS PROJECT
Neuro-SIPE (System Identification and Parameter Estimation of Neurophysiological Systems), is a consortium of Dutch medical universities, technical universities and industrial companies, where the aim is to develop and apply diagnostic tools using SIPE techniques for neurological disorders. Combining both technical (SIPE) and medical (Neuro) sciences, Neuro-SIPE projects investigate neurological disorders by accounting for the closed-loop nature of the central and peripheral nervous systems when recording signals and assessing function of the CNS.
Under the umbrella of the Neuro-SIPE consortium, the Torticollis (cervical dystonia) project is a collaboration between Delft University of Technology and Amsterdam Medical Centre, along with industry partners TMSi (Twente Medical Systems International), Motek Medical, TNO based on the anatomical afferent composition of the labyrinth and the assumption that all
af-ferents are affected by the stimulus. In a binaural bipolar configuration the sensation of motion detected by the otoliths is a small linear acceleration towards the cathode and the sensation of motion detected by the semicircular canals is a rotational acceleration in the roll plane about an axis 18 degrees up from the Reids plane also towards the cathode. The evoked compensa-tory postural responses are therefore in the opposite direction, towards the anode. In addition, because the perceived motion is in a head-centred reference frame, postural responses change direction depending on the orientation of the head relative to the body.
Electrical vestibular stimulation is usually implemented using a square-wave transient stimulus and have been combined with mechanical perturbations in various standing studies (Inglis et al. 1995, Cenciarini and Peterka 2006). Recent developments in the use of continuous stochastic signals have demonstrated several advantages over square-wave stimuli (Fitzpatrick et al. 1996, Dakin et al. 2007, Dakin et al. 2010, Mian et al. 2010). Primarily, continuous stimulation can provide more detailed insight into the output muscular and postural responses by examining the frequen-cy behaviour relative to the input stimulus (i.e., FRFs). In addition, continuous signals offer experi-mental advantages by minimizing anticipatory effects (Pavlik et al. 1999), shortening experimen-tal durations (Dakin et al. 2007) and improving signal-to-noise ratios (Reynolds 2011). A significant limitation of both stimuli however is that muscular and postural responses are relatively small. It has been estimated that only 5-10% of the output muscle response is due to the input stimula-tion (Dakin et al. 2010). Although increasing the stimulus amplitude can improve measurement quality, subjects often perceive these higher intensity electrical stimuli as noxious or painful.
Applied to the head-neck system, transient electrical vestibular stimulation is commonly used to evoke the VCR in neck muscles for clinical assessments of vestibulopathy (Rosengren et al. 2010). As a tool to assess the vestibular contributions to head-neck stabilization however, electri-cal stimulation has received little-to-no attention. This is likely due to two complicating factors: 1) the evoked postural responses are expected to be small compared to those reported in stan-ding balance making them difficult to measure, and 2) the close proximity of the stimulation induces substantial artefact in the measured muscle activity that can only be extracted by crude subtraction of the stimulus recorded from a previous trial (Watson and Colebatch 1998). On the other hand, these issues may be easily resolved through careful experimental design. For example, mechanical perturbations can easily be designed to evoke an equivalent magnitude of postural responses making it possible to combine these two techniques. In addition, by using continuous signals over a limited bandwidth, stimulus artefact can be eliminated through ef-fective filtering of measured muscle activity. Electrical vestibular stimulation may therefore be useful in advancing understanding of vestibular contributions to head-neck stabilization.
HEADS UP Introduction CHAPTER 1
tions and was therefore not pursued further. In the absence of movement however, this proto-col was combined with electrical vestibular stimulation in Chapter 9 to investigate the effects of neck muscle co-contraction on VCR.
The second methodological focus in Part I was on the improvement of continuous electrical ves-tibular stimulation used to evoke vestibulo-motor responses as presented in Chapter 3. Previous developments used stochastic signals, which evoke a small response and low coherence be-tween the input stimulus and output muscle activity. In this study, multisine signals were used to increase signal-to-noise ratios, thereby minimizing required input amplitudes, enhancing vestibu-lo-motor output responses and improving subject comfort. In Chapter 4, this multisine vestibular stimulation approach was subsequently used to assess whether electrical vestibular stimulation is suitable to investigate the role of the vestibular system in upright head-neck stabilization.
Part II investigates the modulation of neck reflexes evoked by varying experimental conditions (objective 2) and is comprised of three separate studies. Chapter 5 investigates how VCR and CCR contribute to stabilization and modulate with perturbation properties and visual input. It was hypothesized that neck reflexes would decrease with torso perturbations that progressively excite the system beyond its resonance frequency. Furthermore, this reflex modulation was expected to be maintained across amplitudes and influenced by the presence of visual feedback (i.e., eyes open or closed). In Chapter 6, the knowledge and experimental protocols from Chapter 5 were used to evaluate sensorimotor impairments in cervical dystonia patients. It was expected that patients would demonstrate an impaired ability to modulate reflexes across perturbations varying in frequency content in comparison to healthy controls.
Chapter 7 combines the methods developed in Chapter 3 and 5 to isolate and observe the modulation of the VCR across perturbation amplitude and bandwidth. The objective was to quantify how neck reflexes, specifically VCR, are modulated during high frequency perturba-tions. In addition, this study aimed to demonstrate that VCR and CCR are modulated in oppos-ing directions with amplitude of the torso perturbation. This was expected given the similar dynamic properties observed for the VCR and CCR.
Part III encompasses a neurophysiological investigation of the VCR (objective 3) and is com-prised of two studies. In Chapter 8, the neurophysiological frequency responses of vestibular reflexes were characterized across neck, back and lower limb muscles using electrical vestibular stimulation. It was hypothesized that vestibular reflexes in neck muscles function over a wider bandwidth. The aim was to determine the mechanisms underlying potential between-mus-cle differences. Therefore, a vestibular afferent-motoneuron pool model was used to attribute these responses to either variations in motor unit firing rates or neural filtering mediating the different vestibulomuscular pathways. These results subsequently led to the proposition in Chapter 9 that unlike lower limb muscles, neck muscles receive a constant vestibular input even (Netherlands Organization for Applied Scientific Research) and Ipsen. The project focuses on the
motor control of the neck and explores sensory aspects of the cervical motor disorder, cervical dystonia. The scientific objectives are 1) to develop experimental protocols and closed-loop identification methods to investigate head-neck stabilization and 2) apply these protocols and methods in order to identify deviating motor control in cervical dystonia patients and optimise subject specific diagnosis and treatment methods with botulinum toxin. The work presented in this thesis focuses on the first objective and the first subcomponent of the second objective.
AIMS AND OUTLINE OF THIS THESIS
The general aim of this thesis is to advance the understanding of the sensorimotor control of head-neck stabilization in order to investigate the sensorimotor impairments observed in cervi-cal dystonia patients. To achieve this aim, three major objectives were defined:
To develop new and improved experimental techniques to assess the posture control of head-neck stabilization and test them in healthy controls.
To quantify the modulation of neck reflexes during head-neck stabilization in healthy sub-jects, and possible deviating modulation caused by sensorimotor impairments in cervical dystonia patients.
To characterize the neurophysiological properties and contributions of vestibular reflexes to neck muscle control in healthy subjects.With the exception of Chapter 1 (Introduction) and Chapter 10 (Conclusions, discussion and future directions), all chapters are considered autonomous and can be read individually since the contents have been published or submitted as journal articles.
The studies performed to achieve the aim of this thesis are divided over three parts, where each part addresses a separate objective. Part I has two primary methodological points of focus in addressing objective 1. The first is the development of an electromyography feedback protocol to examine reflex modulation during postural control at varying levels of muscular effort, also known as sub-maximal tasks. This was originally intended to investigate the effects of modulat-ing neck muscle co-contraction on the reflexive stabilization of the head-neck system durmodulat-ing torso perturbations. As a pilot study, the protocol was implemented on the shoulder to examine the effects of perturbation type (force vs. position) on reflex contributions to postural mainte-nance, isolated from the inherently linked perturbation type (i.e., force perturbations-position task and position perturbations-force task). This pilot study is described in Chapter 2. Although the protocol was transferred to the neck in an additional (unpublished) experiment, the need for subjects to perform the task using visual EMG feedback led to unnatural stabilization
condi-HEADS UP Introduction CHAPTER 1
in conditions where the vestibular information and motor command are irrelevant to head-neck posture control. In addition, the effect of neck muscle co-contraction on the VCR was evaluated using an improved EMG feedback protocol based on the developments in Chapter 2. Here, co-contraction was hypothesized to induce an inhibition of VCR contributions to ensure neck spinal stabilization.
Chapter 10 draws the main conclusions and recommends future directions for research on sen-sorimotor physiology of neck motor control.
CHAPTER 2
EMG feedback tasks reduce reflexive stiffness
during force and position perturbations
EMG feedback tasks reduce reflexive stiffness during force and position perturbations CHAPTER 2 PART I Experimental methods to investigate sensorimotor postural control
I
2
I
2
INTRODUCTION
Posture maintenance of the arm is achieved using both muscular (intrinsic) and afferent (re-flexive) contributions. Intrinsic muscle viscoelasticity increases with muscle activation levels (Agarwal and Gottlieb 1977), and co-activation (i.e., co-contraction) of muscles is an effective although energy consuming method of posture maintenance. Reflexive properties are deter-mined by afferent feedback from sensory organs including muscle spindles (motion), Golgi ten-don organs (force) and tactile receptors (touch and pressure). Reflex responses occur only after a perturbation, making reflexive feedback more energy efficient compared to co-activation. However, due to inherent neural time delays associated with afferent feedback, the effective-ness is limited (Schouten et al. 2008b). Under normal conditions, both mechanisms are balanced to maintain posture.
While the intrinsic muscular contribution to posture maintenance is well understood, the contri-bution of reflexes remains a continued focus of research. Experimental studies have shown that the reflexive contribution to posture maintenance depends on several factors, viz., muscle ac-tivation (Matthews 1986, Kirsch et al. 1993, Cathers et al. 2004), disturbance amplitude (Kearney and Hunter 1982, Sinkjaer et al. 1988, Stein and Kearney 1995, Cathers et al. 1999), the mechanical properties of the device with which the subject interacts (de Vlugt et al. 2002), task instruction (Akazawa et al. 1983, Doemges and Rack 1992, Dietz et al. 1994, Mugge et al. 2010) and the band-width of the perturbation signal (van der Helm et al. 2002, Schouten et al. 2008b, Mugge et al. 2010). Although these studies provide valuable insight into the afferent contribution to posture maintenance, the differences in experimental procedures make it difficult to compare results. In particular, the effect of perturbation type (force vs. position) is difficult to assess as it is inherently linked to the task instruction; position perturbations are used for force tasks and force perturba-tions are used for position tasks. The opposing nature of these task-perturbation combinaperturba-tions results in different reflexive behaviour in both single (Akazawa et al. 1983, Doemges and Rack 1992) and multi-joint systems (Perreault et al. 2008).
The goal of this study is to explore the effect of perturbation type on reflexive feedback by using a harmonized task instruction across both position (PP) and force perturbations (FP). Any differ-ences between the two would help identify either task or perturbation as the primary factor in the different observed reflex behaviours in the literature. Subjects performed electromyo-graphical (EMG) biofeedback tasks (ET) representing desired co-contraction levels in the face of both perturbation types.
EMG biofeedback is a technique used for (re)learning of motor control in rehabilitation settings (Basmajian 1981, Holtermann et al. 2010). Subjects are provided with instantaneous feedback on the activity of measured motor units and instructed to perform various control tasks. Cath-ers et al. (1999) used EMG biofeedback as a task instruction to evaluate the effect of amplitude and bandwidth of stretch and voluntary muscle activation on the magnitude and timing of
ABSTRACT
Force and position perturbations are widely applied to identify muscular and reflexive contributions to posture maintenance of the arm. Both task instruction (force vs. position) and the inherently linked perturbation type (i.e., force perturbations-position task and position perturbations-force tasks) affect these contributions and their mutual balance. The goal of this study is to explore the modulation of muscular and reflexive contributions in shoulder muscles using EMG biofeedback. The EMG biofeedback provides a harmonized task instruction to facilitate the investigation of perturbation type effects irrespective of task instruction. External continuous force and position perturbations with a bandwidth of 0.5–20 Hz were applied at the hand while subjects maintained prescribed constant levels of muscular co-activation using visual feedback of an EMG biofeedback signal. Joint admittance and reflexive impedance were identified in the frequency domain, and parametric identification separated intrinsic muscular and reflexive feedback properties. In tests with EMG biofeedback, perturbation type (position and force) had no effect on joint admittance and reflexive impedance, indicating task as the dominant factor. A reduction in muscular and reflexive stiffness was observed when performing the EMG biofeedback task relative to the position task. Reflexive position feedback was effectively suppressed during the equivalent EMG biofeedback task, while velocity and acceleration feedback were both decreased by approximately 37%. This indicates that force perturbations with position tasks are a more effective paradigm to investigate complete dynamic motor control of the arm, while EMG tasks tend to reduce the reflexive contribution. Patrick A. Forbes, Riender Happee,
Frans C.T. van der Helm, Alfred C. Schouten
EMG feedback tasks reduce reflexive stiffness during force and position perturbations CHAPTER 2 PART I Experimental methods to investigate sensorimotor postural control
I
2
I
2
d(t)-f
h(t)
x
h(t)
mounted in front of the subject, where the signal presented matched the task(i.e., a position signal for the position tasks and a EMG signal for the EMG tasks; see Task instructions and test
protocol for more detail).
Voluntary force experiments
Isometric push–pull tests were used to determine the static EMG-to-force ratio, providing a relative measure of the muscle force, following previously described procedures (Schouten et al. 2008a). Visual feedback of the force was given to motivate and assist the subject. A maximum voluntary contraction test was performed first. Following a 5-min break, subjects had to push or pull to fifteen different force levels (0 and ±10, 15, 20, 25, 35, 50, 70 N) and maintain the levels for 10 s. These test series were conducted before and after the main experiments, and the average of the two was used for signal interpretation.
Main experiments
Task instruction and test protocol
The main experiment consists of multiple tasks, lasting 30 s each, while continuous force (FP) or position perturbations (PP) were applied. Additionally, tasks were performed with no perturba-tion (NP). The instrucperturba-tions associated with each task were as follows:
FIGURE 1
Experimental setup showing a subject seated on a chair holding the handle with the right hand. The hand force, fh(t),
ap-plied to the manipulator is measured by a force transducer mounted between the handle and the piston. The manipula-tor controls the position of the handle, x(t), based on the sum of the hand force, fh(t), external force disturbance, d(t), and
the simulated virtual dynamics (environment).
stretch reflexes in the flexor carpi radialis (Cathers et al. 1999, Cathers et al. 2004). Subjects had to maintain specified levels of muscle activation while being exposed to position perturba-tions stretching the muscle. This flexion-only use of wrist muscles effectively made this task equivalent to a force task linked to a position perturbation. The task instruction proposed in the present study uses EMG biofeedback as a representation of co-contraction of antagonist shoul-der muscles making this effectively a bi-directional stiffness task. Such an approach mediates the investigation of the effects of both muscle co-contraction level and perturbation type on muscular and reflex properties. Random continuous disturbances were applied to the arm, and system identification in the frequency domain was performed. Frequency response functions were estimated to capture the arm dynamics, and a neuromuscular control model was used to parameterize the physiological behaviour in the form of intrinsic and reflexive components.
The effect of using EMG as a task instruction (ET) during force perturbations (FP) was first investi-gated by comparison to a position task (PT). Secondly, the effect of muscle (co)-activation level was investigated, a known influential factor on reflexes. Although this effect has been identified previ-ously using position perturbations (i.e., force tasks), the effect has yet to be confirmed during force perturbations and under EMG biofeedback. Thirdly, the effect of perturbation type during this task was evaluated by applying a position perturbation to subjects which mimicked the position re-corded during the force perturbations without informing the subjects that any change was made.
MATERIALS AND METHODS
Subjects
Ten healthy subjects (nine male) with an age range of 23–33 years having no self-reported his-tory of neurological disorders or upper extremity injuries participated in these experiments. The experiment was in accordance with the Declaration of Helsinki, and all subjects gave informed consent prior to the experiment. All tests were performed on the right arm.
Apparatus
Subjects were seated in a comfortable upright posture, holding the handle of the manipulator (see Figure 1) with an elbow flexion of 90° (i.e., the reference position). The manipulator moved in forward and backward directions generating anteflexion and retroflexion of the glenohumer-al joint (Wu et glenohumer-al. 2005). Details of the electro-hydraulic manipulator were described previously (Ruitenbeek and Janssen 1984, van der Helm et al. 2002). The manipulator could be position controlled or force controlled. During position control, the manipulator followed the prescribed motion regardless of the subject’s resistance. During force control, the force measured between the handle and the actuator (i.e., the force imposed on the system by the subject) was used to generate the motion via a virtual mass-spring-damper system. In other words, the handle felt like a mass-spring-damper system to the subject. For this study, a virtual mass of 1 kg, a damping of 0 N/ms and a stiffness of 160 N/m were used. Visual feedback was given on a 17-inch monitor
EMG feedback tasks reduce reflexive stiffness during force and position perturbations CHAPTER 2 PART I Experimental methods to investigate sensorimotor postural control
I
2
I
2
power region. The perturbation signal used in RT trials had a bandwidth of 0.5–20 Hz consisting of 32 evenly distributed clusters with equal power.
PP signals were based on the measured average displacement of the four repeated FP trials at each specific ET level. In this way, the displacement for the position and force perturbations was equivalent at each ET level.
All signals were 30 s in length and had a sample frequency of 2,500 Hz. For analysis, approxi-mately 4 s was removed from the beginning of all recorded data to avoid the inclusion of any transient effect. This resulted in a signal length of 216 samples (~26.2 s) and a frequency
resolu-tion of approximately 0.038 Hz.
To compare the different conditions and to justify a linear modelling approach, the position deviations were kept within a consistent and limited range. For trials using FP, the magnitude of the force disturbance was scaled for each subject and each condition to ensure a root-mean-square (RMS) value for the position of approximately 3.5 mm. This scaling was obtained prior to the actual experiments.
Analysis
Data recording and processing
During each trial, the handle position xh(t), handle force fh(t), disturbance force d(t) and EMG were
recorded at 2.5 kHz. The EMGs of four relevant shoulder muscles, two anteflexors (m. pectoralis
TABLE 1
Experimental matrix summarizing task instruction and perturbation characteristics
Condition Task EMG co-act (%)
(ET level)
Perturbation
Type RMS amp. (mm)
PT_FP Position N/A Force 3.5a
RT_FP Relaxed N/A Force 3.5
ET_FP EMG 40/70/100/120 Force 3.5
ET_NP EMG 40/70/100/120 None 0
ET_PP EMG 40/70/100/120 Position 3.5
a 3.5 mm was the desired displacement
Position task (PT): subjects were instructed to ‘‘minimize displacement’’, while FP were ap-plied. Visual feedback of the handle position was presented together with a horizontal line indicating the reference position.
EMG task (ET): subjects were instructed to match the displayed EMG co-activation signal (see Analysis) to a target level which was displayed as a horizontal line. Four target levels were defined: 40, 70, 100 and 120% of the EMG co-activation signal recorded during PT. Three perturbations were applied during the performance of this task: FP, PP and NP.
Relax task (RT): subjects were instructed to ‘‘relax’’ as much as possible, while exposed to a FP. The monitor was switched off to remove visual feedback. This task was included to improve the estimate of the arm mass.The experiment is summarized in Table 1. The PT and RT trials were conducted first in order to generate the reference co-activation levels required for all ET. The ET_FP trials were mixed randomly with half of the ET_NP trials and performed first. The displacement measured during the ET_FP trials was then used as the disturbance for the ET_PP trials (see Perturbation signals). These were performed mixed with the second half of the ET_NP trials in a random order. During ET trials, subjects were not informed on the type of disturbance (FP, PP or NP). Each test was performed four times resulting in 56 trials.
For all ET trials, subjects were instructed as a pre-condition to co-contract their muscles to the level required. The handle was locked prior to the start of the trial to prevent drift, and ET trials were started only when the absolute handle force was less than 3 N in order to preclude unidirec-tional muscle contraction strategies. The 160 N/m stiffness included in the manipulator helped to prevent drift during the actual experiment. ET trials are described in the remainder of the article as ET_FP40, ET_FP70, ET_FP100 or ET_FP120, where PP replaces FP for position perturbation trials.
Perturbation signals
One perturbation signal was used as the disturbance in all FP trials with the exception of RT trials. The signal had a rectangular power spectrum with dominant power over a bandwidth of 0.5–1.5 Hz and was composed of multiple sinusoids, i.e., a multisine signal (Pintelon and Schoukens 2001). This was supplemented with a reduced power region according to methods developed by Mugge et al. (2007). The reduced power region (1.5–20 Hz) also had a rectangular spectrum and was designed to have 20% of the power in the dominant power region. Such an approach facilitates system identification over the whole bandwidth of 0.5–20 Hz while eliciting responses adapted to low bandwidth perturbations, which enhances reflexive contributions (van der Helm et al. 2002, Schouten et al. 2008b). Frequency averaging of four adjacent points was used for estimating the frequency response functions (see Analysis), resulting in only four clusters of four adjacent frequencies in the full-power region and 23 clusters in the reduced
EMG feedback tasks reduce reflexive stiffness during force and position perturbations CHAPTER 2 PART I Experimental methods to investigate sensorimotor postural control
I
2
I
2
the output relative to the input, and the phase indicates the timing of the output relative to the input. The admittance describes the displacement (xh) of the arm due to an input force (fh) and represents the inverse of the limb impedance. The reflexive impedance describes the muscle activation (a) due to the handle displacement (xh) and reflects the afferent feedback. For FP tri-als, interaction between the subject and manipulator exists and closed loop algorithms were required to estimate the FRFs according to:
ˆ ˆ ˆ H S f S f fx dx df =
( )
( )
ˆ ˆ ˆ H S f S f xa da dx =( )
( )
where Ŝdx(f) represents the estimated cross-spectral density between the disturbance d and the
handle position xh. Likewise, Ŝdf (f) is the cross-spectral density between signals d and fh, and Ŝda(f)
is the cross-spectral density between signals d and a. The cross-spectral densities are averaged across four adjacent frequencies to improve the estimate (Jenkins and Watts 1968). For all PP trials, standard open loop algorithms were used to estimate the FRFs, i.e., d was replaced by x in Eqs. 2-3.
In each condition, the coherence was used to evaluate the quality of the correlation between the input and output according to:
ˆ ˆ ˆ ˆ gdx dx dd xx S f S f S f =
( )
( ) ( )
2
ˆgda ˆ ˆdaˆ dd aa S f S f S f =
( )
( ) ( )
2
where d is replaced by x for PP conditions. Coherence ranges from zero to one and described how much of the input power is related to the output power. The FRFs and coherences were evaluated only at frequencies where the perturbation signal had power.
Parameter estimation of neuromuscular model
A neuromuscular model was employed to identify the physiological relevant parameters un-derlying the experiments (Schouten et al. 2008b, Mugge et al. 2010). The model facilitates the quantification of proprioceptive reflexes, muscle parameters and mass. Figure 2 shows the en-tire model structure including the external force disturbance, environment and arm. The exter-nal force disturbance D(s) minus the force measured at the handle Fh(s) was the input for the environment. The environment was modelled as a second-order mass-spring-damper system which did not vary between conditions where s = j2πf.
major and m. deltoideus anterior) and two retroflexors (m. deltoideus posterior and m. latissimus dorsi) were measured with differential surface electrodes (Delsys Bagnoli System, Delsys, Boston, USA). Before digitizing, the EMG signals were preamplified and bandpass filtered (20–450 Hz).
The EMG signals were used for two purposes: (1) real-time biofeedback of muscle co-activation (i.e., co-contraction) ca(t) during ET and (2) construction of a muscle activation signal a(t) for
system identification.
For biofeedback, the four individual signals were online high-pass filtered (5 Hz), rectified, nor-malized to their separate PT means and low-pass filtered at 0.3 Hz (second-order Butterworth). The co-activation ca(t) was defined as the average of the four muscle signals. A cut-off of 0.3
Hz was chosen to minimize visible variations in the biofeedback signal due to reflexes evoked by the perturbations and thus preventing subjects from possibly reducing reflexive feedback. Lower cut-off frequencies resulted in slow responses of the biofeedback signal and poor task performance. The normalized activation signal of each muscle, as well as the paired anteflexor and retroflexor normalized activation signals, was also analysed to assess task performance and task implementation.
The muscle activation signal a(t) estimates the net amount of force generated using the EMG sig-nals and EMG-to-force ratios and is used in the non-parametric system identification (Schouten et al. 2008a). A prewhitening filter was applied to the raw measurements to improve the quality of the signals following the procedures of Clancy et al. (2002). The parameters for the filter (sixth order) were obtained from the power spectral densities of the 25 N isometric push and pull task. The results of all isometric push–pull tasks were then used to estimate the force-to-EMG relationship by linear regression. The total muscle activation was then obtained by summing the scaled rectified prewhitened EMGs of the four muscles according to:
a t
( )
=1∑
i=K ei w i( )
t2 1
4
,
where Ki is the EMG-to-force scale factors for each muscle, being positive for the two anteflexors and negative for the two retroflexors. This equation assumes that the two muscles generating either push or pull have equal relevance and that their mean value represents the total force being generated in that direction.
Non-parametric system identification
The four repetitions of each test were averaged in time to reduce the effects of noise. The admit-tance and reflexive impedance frequency responses functions (FRF) were then estimated in the frequency domain using system identification techniques described in detail by van der Helm et al. (2002) and Schouten et al. (2008b). In general, FRFs describe the input–output relationship of a system as a function of frequency. At any frequency, the gain indicates the magnitude of
