Abstract
Gait event detection is a
well-researched and important area. Gait event detection identifies and/or
predicts, when the lower limbs are/will be in particular positions.
This paper demonstrates
that, in some applications, identifying specific gait events is not strictly
necessary. It is sufficient to produce a single signal once in each gait cycle.
Providing it occurs reliably, at the same point during that gait cycle.
The approach was verified
using simulated and pre-recorded data, the results obtained were extremely
satisfactory.
1. Introduction
Functional electrical
Stimulation (FES) is used in the rehabilitation of patients whose nerve signals
have been impaired due to accident or injury such as strokes or spinal injuries
[1]. Such methods are inherently very
complicated, as they are trying to replace functionality usually provided by
the, very complicated, human body.
Impairments resulting from
stroke or spinal cord injury (SCI) can leave the muscle, and the nerve attached
to it, intact. In both cases, the impediment to the natural stimulation signal
is closer to the brain, or in the brain.
Artificial signals can then
be applied to the still enervated muscle to produce tetanic contractions, and
therefore artificially stimulate movement in the limbs.
When reproducing movement in
the lower limbs, to re-create normal walking gait, timing is very important;
otherwise the body may become unstable.
Therefore, in automatic FES
systems designed to return normal walking to a patient, the stimulation signals
must be timed to coincide with the natural signals they are replacing.
This means that such a
system must have sensors that provide data, which can be analysed to give
specific gait event information [2-4,6]. Sensors that have been used include
accelerometers [5,6], gyroscopes [4], and switches placed under the foot. These
can give gait event timings for heel-strike, toe-off, mid-swing, double-limb
support, single-limb support. The stimulation of the necessary muscles can then
be synchronised using the gait events as temporal points of reference.
The authors present a method of identifying a specific point in the gait cycle, which can then be used to synchronise stimulation timings.
This method does not
identify any of the specific gait events mentioned previously, and therefore is
simple enough to be used in small low-power microcontrollers.
The start and end of a gait
cycle is determined to be the time when a certain waveform shape is identified
over a sample window of x samples.
The length of the sample
window can be varied to ensure that no false activations are made, e.g. by
identifying small limb joint movements as major ones.
In this approach, the hip
angle is used for the synchronisation data as this usually provides a single
oscillation within each gait cycle, with the sample window excluding minor
fluctuations.
The technique is tested
using pre-recorded data and simulated input. The synchronisation method is
accurate enough so that the synchronised recorded data has no observable spikes
or jumps.
2.
Materials and Methods
The sensors used to provide the angle information are variable resistance bend sensors; these have a nominal resistance of 10kW at 0° and 35kW at 90° [7]. The sensors are attached with a nominal bend angle of 90°, calibrated to be 0° of limb angle movement as shown in figure 1.
Figure 1: Location of bend sensors with 90° offset between limb sections
The sensors are linear and
have a high resolution (about 25kW over 90 degrees)
only in one direction of bending. Hence attaching the sensor in the middle of
its range (90° offset) allows
the recording limb flexion and extension in the same linear sensor region.
The analogue voltage data
across each of the sensors is read and converted to a digital value by a PIC
microcontroller running at a clock speed of 16Mhz. This gives a sampling
frequency of approximately 70 Hz per-channel including A/D conversion time.
The gait synchronisation
method uses a sample window of a variable number of samples. The default length
is four samples. However, the sample window length can be adjusted in order to
exclude high frequency variations.
Therefore the same routine
can be applied to both hip sensors.
The hip angle is used for
synchronisation as, in a normal gait cycle, the hip only undergoes one
oscillation.
However, in both normal gait
and abnormal gait, there are minor fluctuations, resulting in noticeable peaks
and troughs; the sample window routine prevents these from signalling an event.
The routine identifies
certain features on the waveform recorded from the hip of both limbs and
signals an event.
This can be used both to
start and stop an automatic data recording procedure; this gives a data file
with values corresponding to one whole gait cycle.
The method can also be used
in a rehabilitation scenario, to give correct stimulation timings for a FES
system.
The method is used simply to
notify one system or another of the start and end of a gait cycle. The
definition of the start and end of a gait cycle can be specified to coincide
with the operation of the system. Subsequently, it is not necessary to derive
specific conventional gait events directly.
Knowing what waveform
sections will signal an event means that it is still possible to calculate
these specific events if needed.
The time lag between the
application of stimulation and the resulting torque at the joint is between
100ms and 300ms. Thus, the stimulation signal necessary at any one time will
need to be applied before it is actually needed. To account for this lag,
knowing the sample period, it is simply necessary to re-associate the
stimulation data with sensor data previous to it.
That is, to link the
activation of stimulation data with sensor data which will occur approximately
200ms prior to the sensor data it is intended to alter.
3.
Results and discussion
The system was tested using simulated input and pre-recorded data.
To simulate input, the hip
sensors are manually oscillated. With each one being approximately 180° out of phase, to simulate
hip movement in normal walking.
The data recording times are
automatically controlled using the synchronising routine. The resulting data
are displayed using the gait analysis simulation interface [8]. The temporal
graph can be seen in figure 2. Note that a 3D display over time is also
available
The difference between the
value of the data recorded at the beginning and the end of the cycle, as seen
on the time graph (Fig. 2) is similar to the difference between consecutive
sequential data sets.
Thus, when looking at a
number of repeated recorded waveforms, there are no abrupt changes in data
values. This indicates that the detection is occurring at the same point on the
gait cycle each time.
Figure 2: PC Display showing
automatically recorded / synchronised hip angle data
Testing using the
pre-recorded data involves loading a data file of a single gait cycle that is
not synchronised with respect to this system.
This file is then run
through the PC software with the event detection routine active. This then
displays the position in the loaded data file that an event is found.
The same position is
indicated each time as this signifies that only one ‘event’ is detected in a
single gait cycle.
This synchronisation allows
the testing of control techniques using data recorded with this system, or
using compatible angle data from any other source.
The system allows the
comparison of data recorded from many different sources as it can essentially
re-sequence external data to coincide with any other data. This will prove
useful and time saving when concerned with statistical analyses.
4.
Conclusion
Although data from testing
on subjects is still being gathered, these results give a good indication of
the actions of the routine in response to real input.
The research detailed here shows that there is merit to the use of a simple synchronisation method with regard to gait analysis.
The simplicity of the approach allows for use on small microcontrollers without sacrificing processor time or requiring an increase in operating frequency, which in turn would increase power consumption. This can produce a stimulation timing signal without having to alter any previously implemented algorithms.
With regard to the offset of each sensor by 90°, another method that can be used to overcome the one-way linearity of the sensors would be to have 2 opposingly oriented sensors attached to the same limb joint.
With this modification, the system will use the whole range of movement for each sensor in one direction of joint movement, while ignoring the overlapping non-linear regions of each.
Further work on this system includes the addition of a routine to monitor the timing of both event signals, and exclude simultaneous events. Such events signify a non-gait movement such as shifting weight or sitting/standing.
Acknowledgements
This work was supported by the University of Reading Research Endowment Trust Fund. Project R2405
References
[1] Benton, Baker et. al. Functional
electrical stimulation: A practical clinical guide. 2nd, 3rd
Ed. Rancho Rehabilitation Engineering Program. Rancho Los Amigos Medical
Centre, 1993.
[2] B.J. Andrews, A. Kostov,
R.B. Stein, ‘Gait event & user intention detection for FES – control:
selecting sensors’, IEEE-EMBC and CMBE, Neuromuscular systems/Biomechanics,
1995.
[3] M.M. Skelly, H.J.
Chizeck, ‘Real time gait event detection during FES paraplegic walking’, Proc.
19th international conference – IEEE/EMBS, pp 1934 – 1937, 1997.
[4] Ion P.I. Pappas, Milos R. Popovic, Thierry
Keller, Volker Dietz, Manfred Morari ‘A reliable gait phase detection system’, IEEE
Trans Neural Syst. Rehabil. Eng. 9(2): pp113-25, 2001.
[5] Michael Whittle. Gait Analysis. (6
May, 1996), Butterworth-Heinemann.
ISBN:0750622229, pp 146-147
[6] Norio Furuse, Imre Cikajlo, Tadej Bajd, ‘Training of foot contact phase during FES assisted walking’, IFMBE Proceedings, Medicon 2001, IX Mediterranean Conference on Medical and Biological Engineering and Computing, Vol II, pp 686 – 689, 2001.
[7] http://www.imagesco.com/catalog/flex/FlexSensors
[8] Greg A. Benedict, Virginie F. Ruiz, ‘A computer simulation for testing and implementing gait controllers’, IFMBE proceedings, IX Mediterranean conference on medical and biological engineering and computing, Vol II, pp 670 – 673, 2001.