Overall Objectives
Scientific Foundations
Application Domains
New Results

Section: New Results

modelling and Identification

Force assessment based on evoked EMG in intermittent stimulation

Participants : Qin Zhang, Mitsuhiro Hayashibe, Maria Papaiordanidou, Philippe Fraisse, Charles Fattal.

Muscle fatigue phenomenon and the inadequacy of force sensors limit the application of FES technology. It is essential to monitor muscle state and assess the generated force to compensate the fatigue and achieve the desired trajectory. It is also important to cease the stimulation depending on muscle fatigue to prevent serious muscle damages. Our final purpose is the on-line monitoring of muscle state and assessment of the muscle force to get a more accurate FES control. Evoked EMG signal offers a way of studying the myoelectric features of the neuromuscular activation associated with muscle contraction. This work was concentrated on the development of EMG-torque model to predict force more accurately. In this case, two spinal cord injured (SCI) subjects participated in the experiment for two sessions, recruitment and fatigue session. Tibialis anterior muscle was stimulated with intermittent stimulation, which can reduce muscle fatigue during rest duration, however, this complicated the EMG-torque relationship due to different recovery velocity of EMG and torque. Force assessment in intermittent stimulation is more difficult than the one in continuous stimulation.

With surface EMG acquisition and suppression of stimulation artifact, time domain EMG parameters, peak-to-peak (PTP) amplitude and second phase area (SPA), represent high correlation to the ankle torque. When the muscle is non-fatigued, PTP shows positive correlation with torque when increasing the stimulation level. When the muscle is fatigued PTP shows correlation with intermittent stimulation. SPA is negatively correlated with the torque at constant stimulation level. PTP and SPA can therefore be combined to build an EMG-torque model which can estimate the torque even when the muscle is fatigued. For each subject, the pooled data from both recruitment and fatigue session were used for model inference. Fig.3 illustrates the result obtained from one of the subjects using cross validation [35] .

Figure 3. Non-fatigued (blue) and fatigued (red) PTP, SPA and torque in the same stimulation amplitude. The measured (smoothed line) and estimated (unsmoothed line) torque obtained using EMG-torque model (bottom). The estimation was based on the data which were not used to identify the model.

Force assessment and muscle fatigue detection in implanted subject

Participants : Mitsuhiro Hayashibe, Qin Zhang, David Guiraud, Christine Azevedo-Coste.

For SCI patients, muscle fatigue under FES is an insensible information. In fact, muscle fatigue is difficult to avoid in prolonged movement restoration even with the use of intermittent stimulation. Both for FES control and security of patient, it is essential to observe the time-varying muscle state especially for fatigue and recovery conditions. Implanted FES systems already provide the mobility to be used in private environments, similarly, muscle fatigue should be captured with sensors which have such mobility. EMG requires small electrodes and amplifiers and some wireless EMG systems are commercially available. Evoked EMG can be one of the solutions to detect the time-varying muscle conditions.

In this experiment, a fully implanted SCI subject participated in three sessions. Stimulation patterns were prepared for the recruitment with amplitude modulation, random amplitude modulation and fatigue session with prolonged stimulation. In this initial trial, continuous electrical stimulation was applied to peroneal branch of sciatic nerve by implanted stimulator. The experimental set-up is shown as in Fig.4a. PTP, root mean square (RMS), mean absolute value (MAV) and net area of the ellicited M-wave allow to track efficiently the changes in the joint torque during muscle fatigue. Fig.4b illustrates part of the results. The obtained EMG-torque model will be helpful to achieve more accurate FES control corresponding to muscle fatigue in future work. We are now working on on-line data processing of EMG for real-time assessment of muscle fatigue condition in FES.

Figure 4. (a)Experimental set-up. (b)Upper graphics shows measured torque (black) and PTP of M-wave in measured EMG (red). Bottom graphics shows measured torque (black) and net area of M-wave in measured EMG (green).
(a) (b)

Kinetics of neuromuscular changes during low-frequency electrical stimulation of the abductor pollicis brevis

Participants : Maria Papaiordanidou, Alain Varray, David Guiraud.

The aim of the present study was to examine the time evolution of fatigue components during electrically induced fatigue of the abductor pollicis brevis muscle (APB). Three series of 17 trains (30 Hz, 450$ \mu$ s, 4 s on - 6 s off, at the maximal tolerated intensity) were used to fatigue the muscle. Neuromuscular tests, consisting of electrically evoked and voluntary contractions, were performed before and after every 17-train series. Maximal voluntary force generation capacity and force induced by the trains of stimulation significantly decreased throughout the protocol (-20% and -27% respectively at the end of the protocol, P < 0.001). This decrease was accompanied by significant impairment in the muscle contractile properties (P < 0.05), as assessed by the muscle mechanical response (Pt), as well as by failure in muscle excitability (P < 0.01), studied with the muscle compound action potential (M-wave or Mmax). Central fatigue indices (level of activation, RMS/Mmax and H reflex) were not significantly changed at any point of the protocol, giving evidence of optimal motor command reaching the motoneurons and preserving spinal excitability, ensuring fully central activation of the muscle. The results indicate that a low-frequency stimulation protocol when applied to a fast-fatigable muscle entails peripheral fatigue development, while central fatigue components are not implicated. The nature of the studied muscle (fast-fatigable or fatigue resistant) seems to be an important determinant of the fatigue component development, since, in an earlier study from our laboratory, the same protocol applied to the plantar flexors (fatigue resistant muscles) provoked central activation failure (fig.5).

Figure 5. (a) Mwave evolution throughout the ES protocol. (b) Muscle mechanical response during the ES protocol.
(a) (b)

A computational model of Inner Hair Cell ribbon synapse of the cochlea

Participants : Christophe Michel, Jérôme Bourien, Jean-Luc Puel, Christine Azevedo-Coste.

modelling the Inner Hair Cell (IHC) ribbon synapse is an interesting alternative to evaluate in silico hypothesis which are difficult (or impossible) to investigate in vitro and in vivo. Existing models are generally constituted of a cascade of five stages: i) stapes motion, ii) basilar membrane motion, iii) IHC depolarisation, iv) neurotransmitter release (glutamate) in the synaptic cleft, v) action potentials firing in auditory nerve fibers. In physiological conditions, these models provide a good agreement with both pre- and post-synaptic data recorded in vitro (patch clamp) and in vivo (Compound Action potential, Peri-Stimulus Time Histogram). These models are incomplete at the post-synaptic compartment. This lack does not permit to study pathologies like tinnitus which affect the firing of auditory nerve fibers. The aim of this work is to develop a computational model of the neurotransmission process from the IHC to auditory nerve fibers (fig.6 ). At the post-synaptic level, the release of glutamate vesicles in the synaptic cleft triggers action potential firing through a two-stage model (fig.7 ). Our preliminary results show that the activation of second receptor type which is inactive in normal conditions, leads to a drastic increase of the action potentials firing. This result reinforces previous data which reveal a firing increase of 250% in animals which experienced tinnitus. Beyond this work, we will develop a computational model of the pre-synaptic compartment in order to study in silico the neurotransmission diseases of this ribbon synapse like deafness and tinnitus.

Figure 6. The neurotransmission process from the IHC to auditory nerve fibers. A sound stimulation causes: a cilia motion, a depolarisation of the IHC membrane, an opening of calcium channels, an exocytosis of glutamate vesicles into the synaptic cleft. AMPA receptors (yellow) detect and trigger action potential firing into the auditory nerve fibers
Figure 7. Two-stage post-synaptic model. A. Vesicle exocytosis times are assimilated to an impulse train. B. Each impulse generates through an IIR filter an excitatory post-synaptic current (EPSC). C. EPSCs drive an adapted Hodgkin-Huxley model which generates action potentials firing.

modelling trunk CPG in locomotion

Participants : Jean-Charles Ceccato, Christine Azevedo-Coste, Jean-René Cazalets.

We developped a new model of human locomotion based on a central pattern generator (CPG) mechanism. The CPG is represented by an oscillator network especially dedicated to reproduce trunk rhythmic activities during locomotion, as we have observed in vivo (see previous activity reorts). The model comprises an external input, which allows driving the behavior according to gait context or environment changes (fig.8). The model is able to reproduce walking, running and jumping behaviors. We also demonstrated how the model can be applied to observe the gait of an individual walking. In this case, the external input is a signal from an accelerometer sensor placed on the trunk of the subject. The model output is therefore synchronized with the subject and automatically adapts to changes of his gait pattern. Experimental results show a good synchronization between real and model based simulation behaviors. [20] , [22]

Figure 8. CPG-based trunk modelling and application to locomotion observation

Bladder function modelling

Participants : Jérémy Laforet, David Guiraud, Christine Azevedo-Coste, David Andreu, Maureen Clerc.

We present a bladder model including detrusor and sphincter dynamics. The model focuses on artificially controlled bladder contractions under electrical stimulation. A numeric study of parameter sensitivity shows which ones are required for accurate estimation in order to achieve reasonable patient's dependent simulation. We finally demonstrate the interest of using such models in order to optimize the stimulation profile. Indeed, the choice of On-Off duty cycle influences the efficiency of the bladder voiding and the maximum intravesical pressure. Fine tuning of this duty cycle leads to enhanced urine outflow while maximal pressure is lowered (figure 10 ).

When contracting a muscle using NFES (Neural Functional Electrical Stimulation), the stimulus always activates first the axons of greater diameter. Also selective activation of a given fascicle inside a nerve is not possible with classical cuff electrode as the recruitment is performed uniformly around the nerve. These limits lead to poorly selective muscle recruitment, inducing fatigue and possible pain. To overcome this, selective stimulation strategies can be used. We propose a tool chain to investigate, simulate and tune selective stimulation strategies. It consists of:

This tool-chain can be managed using a graphical interface to make the simulations easier to define and run (figure 9 ).

Figure 9. Screen-shot of the graphical interface developed to manage the simulation tool chain.
Figure 10. Summary of the numerical study results on the influence of the duty cycle of intermittent command for bladder voiding under FES.


Logo Inria