Professional Documents
Culture Documents
FES Pareza SPE
FES Pareza SPE
Abstract: We report on our advances in sensory feedback stance phase, which can collapse the subject, and infre-
data processing and control system design for functional quent broken stimulation pulses in the swing phase, which
electrical stimulation (FES) assisted correction of foot can result in unpredictable consequences. In this study, we
drop. We have applied 2 methods of signal purification on have introduced adaptive restriction rules (ARR), which
the bin integrated electroneurogram (i.e., optimized low are initially used as previously reported and then dynami-
pass filtering and wavelet denoising) before training adap- cally adapted during the use of the system. Our results
tive logic networks (ALN). ALN generated stimulation suggest that ARR provide a safer and more reliable stimu-
control pulses, which correspond to the swing phase of the lation pattern than fixed restriction rules. Key Words:
impaired leg when dorsal flexion of the foot is necessary to Foot drop—Functional electrical stimulation—Control
provide safe ground clearance. However, the obtained strategies—Neural recordings—Sensory feedback—
control signal contained sporadic stimulation spikes in the Adaptive logic network.
Functional electrical stimulation (FES) can assist cosmetically undesirable and is exposed to adverse
hemiplegic patients by stimulating the peroneal conditions, causing frequent failures. Fully im-
nerve in the swing phase of the affected leg and planted stimulators can resolve the first 2 problems;
thereby providing dorsiflexion of the foot. Detection however, once the stimulator is implanted it would
of the heel strike and the impaired leg’s foot lift are be desirable to have sensory feedback generated
prerequisites for successful FES use in foot drop cor- from implanted or natural sensory sources. It has
rection. Systems based on surface stimulation and been suggested that afferent neural signals, such as
controlled by a heel switch located in the shoes have those recorded from the sural or calcaneal nerves,
been most successful even if the performance has not which carry mainly cutaneous information, can be
been optimal. Major obstacles in better acceptance used as sensory feedback signals in fully implanted
of these systems include the following. First is the FES systems (1,3). Heel strike, which is important
complicated daily application and removal of the for determination of the stimulus duration, is easy to
FES system. Second is the need to position elec- detect in the rectified and bin integrated electroneu-
trodes precisely to produce correct movement of the rogram (RBI ENG). At the same time, detection of
foot. Third is that the surface stimulation can be un- foot lift, which is important for the start of the stimu-
comfortable at required stimulation levels and may lation, is more difficult to detect because the slower
produce skin irritation. Fourth is that the sensor de- change of skin stretch reduces the firing rate of cu-
tecting heel strike which controls the stimulator is taneous receptors and second because of the larger
influence of the unwanted nerve signal resulting
Received December 1998. from the activation of cutaneous receptors by socks,
Presented in part at the 6th Vienna International Workshop on
Functional Electrostimulation, held September 22–24, 1998, in Vi- footwear, and bending in the metatarsal joints. In
enna, Austria. our previous work, we demonstrated the potential of
Address correspondence and reprint requests to Dr. Aleksan- using adaptive logic networks (ALN) and restriction
dar Kostov, Faculty of Rehabilitation Medicine, 3-48 Corbett
Hall, University of Alberta, Edmonton, AB T6G 2G4 Canada. rules (RR) for gait event detection and FES con-
E-mail: Aleks.Kostov@ualberta.ca trol (2). In this study, we introduced dynamic
443
444 A. KOSTOV ET AL.
sulted in the best ALN training performance. Third, modified version of the rules described previously
a minimum of 10 gait cycles was necessary for suc- (5). The following equation illustrates the concept.
cessful training; we used 15. Fourth, the ALN train-
ing parameters were 5 layers, 0.15 learning rate, and 共Pst − 2 · st兲 ⬍ Pst ⬍ 共Pst + 2 · st兲 (1)
15 epochs. Pbs ⬎ Pbs − 2 · bs
A new measure of evaluating the ALN output was where Pst is the stimulation period (swing) and Pbs is
introduced. In addition to calculating the actual per- the period between stimulations (stance).
centile error, which does not reveal information We developed adaptive restriction rules (ARR)
about the type of error, the output signals were as- by calculating the mean and standard deviations for
sessed by calculating the frequency distribution of the stance and swing phases and using them to elimi-
the transition duration (TD). This distribution was nate control signal transitions considered to be out
calculated by counting the number of samples be- of range for the particular user. The modifications
tween each transition. This approach was selected consisted of continuous adaptation of the rules to
due to the periodic nature of the gait. Comparing the current gait as the ALN were evaluated on new data.
TD distribution of the ALN output to the distribu- A buffer containing transition duration information
tion of the target signal revealed information of its on stance and swing phases during the previous 15
functionality (Fig. 3). gait cycles was used to store and to calculate new
For the elimination of functional errors (2,5), dif- mean and standard deviation values for the restric-
ferent methods using restriction rules were investi- tion rules. The buffer was updated following each
gated. In the first method, we eliminated transitions transition in the output from the restriction rules.
shorter than a specified duration (blanking) while in
the second method, we applied a moving average RESULTS AND DISCUSSION
filter followed by a threshold (smoothing). Both
methods eliminated short duration transitions; how- Figure 3 illustrates the TD distributions of the tar-
ever, they introduced a small delay in the output get FES control signal, ALN output, and the ALN
signal. This delay was considered to be negligible for output processed by the adaptive restriction rules.
foot drop correction. Each method was tested indi- Most of the swing phases were detected correctly by
vidually as well as combined with the other one. ALN while most of the errors occur in the stance
The restriction rules used in this study were a phase (Fig. 3B). Almost none of the complete stance
FIG. 3. Shown are the distribution plots of the swing and stance phase transition durations for the target control signal (A), ALN output
(B), and ALN output processed by adaptive restriction rules (C).