Abstract

A closed-loop controlled gait neuroprosthesis has been designed and evaluated in experiments with paraplegic patients. Walking was accomplished by stimulating 3 muscle groups and the flexion withdrawal reflex at each leg. Ground reactions, joint angles and velocities were recorded by insole pressure sensors, electro-goniometers, and gyroscopes, respectively. Yet, only knee angles and ground reactions are sent to the controller. Improvement of walking performance and safety as well as reduction of muscle fatigue could be achieved by sensor-based transitions, and a knee extension controller to support the patient's desired movement, and to minimize stimulation of the knee extensors.

Introduction

The application of functional electrical stimulation (FES) to restore gait in patients with central lesions has been investigated since the early 1960's, yet, clinical acceptance is poor. Currently, systems in clinical use are open-loop controlled, and apply FES via surface electrodes. Little functional gain cannot compensate for excessive system don/doff time and muscle fatigue, insufficient patient comfort and movement control.

The application of closed-loop control is mandatory to increase walking performance, and patient safety. In addition, patient comfort can be improved, providing them with enhanced, subject-driven controllers. Numerous studies have focused on the control of non-functional single joint movements, standing, and standing-up, both in experiments and simulations [1,2]. However, a lack of studies addressing the closed-loop control of gait has to be constituted.

In this paper, a closed-loop controlled system is presented. Based on the experience with the group's open-loop gait neuroprosthesis, automatic switching of gait phases, and a knee extension controller was introduced. The goal was to provide experimental evidence that closed-loop control can significantly improve walking performance and patient comfort, decrease muscle fatigue, while keeping the system's complexity low.

Methods

A. Experimental Setup

A new neuroprosthetic system was developed, comprised by a multi-channel sensor system, a multi-channel neurostimulator, a sensory substitution system, and controlled by a process control software (Fig. 1).

Figure 1: Process control, a sensor system, a sensory substitution system and a neurostimulator comprise the neuroprosthetic system.

Ground reaction forces from insole pressure sensors (Zebris GmbH, Germany), joint angles from electro-goniometers at ankle, knee, and hip joints, and rotational velocities from gyroscopes at pelvis, thighs, shanks, and feet are recorded at 200Hz by a custom-built sensor system [3]. Stimulation pulses are generated by the current controlled, charge balanced 8-channel neurostimulator ProStim8 (Omicron-Hardtech, Montpellier, France). Modification of each pulse is realized via an RS 232 serial link. A newly developed sensory substitution system is employed to provide the patients with artificial sensory information from their lower extremities. Four miniature vibration motors as used in cellular phones are integrated into a shirt, and positioned closely to clavicles and scapulae. The whole system is controlled by a newly developed process control software package running on a PC (PIII-500MHz, 128MB RAM, WinNT4, C++) at an update rate of 20Hz.

Subject	Sex	Lesion	YPI	Age	Weight	Size
MM	w	Th 7	17	43	47 kg	156 cm
SK	w	Th 9	3	33	55 kg	165 cm

Table 1: Patient data. YPI: years post injury

Stimulation is applied via self-adhesive surface electrodes at quadriceps, hamstrings, and gluteus muscles at 20 Hz, the flexion withdrawal reflex is elicited at 40 Hz. Stimulation current is kept constant and set for each channel individually. Pulse width is used to modulate muscle force. Three patients with complete SCI, well trained and highly motivated, were asked to participate in this study (Tab. 1). They gave informed consent.

B. Control System

Gait is realized by a finite state control scheme (Fig. 2). In each phase a low-level controller is active to generate the required stimulation pattern. Steps are synthesized by three phases: Flexion, KneeExt1, and KneeExt2. During phase Flexion, which is initiated by the patient's finger switches, the flexion withdrawal reflex of the swing leg is activated to cause flexion at hip, knee, and ankle joint. In phase KneeExt1 the knee is extended by activation of the quadriceps, in phase KneeExt2 the gluteus is activated to extend the hip and, thus, to move the patient forward. In these phases, contralateral knee and hip extensors are activated to stabilize the stand leg.

Figure 2: Finite state control scheme for a right step. A left step is realized equivalently.

In the open-loop system, the pattern of each stimulation channel is defined by the phase duration and a value to be set at phase termination, resulting in stimulation ramps or plateaus. Phase durations are fixed, and defined for each patient individually, except phase Flexion which is switched on and off by a finger switch,. To guarantee safe stance in the open-loop system, knee extensors are stimulated with supramaximal pulse widths, yielding excessive muscle fatigue. In addition, the patient's means to control the movement are limited, since they have to comply to the phase durations.

To overcome these shortcomings, time-based transitions have been replaced by sensor-based transitions, and a knee extension controller has been added. To keep sensor system complexity limited, only knee angles and insole pressure signals are utilized by the control system.

1) Transition tFlexion, switching to phase Flexion, was extended by a security mechanism to prevent initiation of a step if the foot load is not shifted sufficiently to the supporting leg.

2) Transition tKneeExt1 was extended to switch when the finger switch is released or a knee flexion threshold is exceeded:

(1)

3) In transition tKneeExt2 foot load of the swing leg is checked and it is switched to phase KneeExt2 only when a certain threshold is exceeded. In this manner, ground contact triggers hip extension:

(2)

4) Knee Extension Controller (KEC). The control objective is to maintain or achieve knee extension. This can be accomplished without detailed knowledge of the neuromuscular system: the quadriceps activation is increased as long as the knee extension is not sufficient, kept constant if it is sufficient, and decreased if the knee is hyperextended. Therefore, the knee angle is utilized as controller input, and mapped to three discrete states knee(k), defined as BUCKLE, extension, and hyperextension. These states are separated by knee angle thresholds and determined for each patient individually.

(3)

The following control law is used specifying the change of PW of the quadriceps muscles at time t =k dt:

Additionally, an upper and lower boundary for was introduced. In all gait phases except for the swing leg during phase Flexion, a KEC is active for each leg. independently.

Results

In a total of 12 experimental sessions, two patients performed 71 runs. One of these sessions was carried out outdoor. 1073 steps were recorded. In the first sessions, the new neuroprosthetic system was tested in open-loop mode. The data obtained (382 steps) served as reference data. Thereafter, the system was used in closed-loop mode.

Succeeding electrode attachment and sensor calibration, patients were asked to stand up, and control parameters were fine tuned, if necessary. Then, they were asked to repeatedly walk a distance of about 4.5m. Control parameters of =30°, = 6°, =0°, =1600ms/s, =400ms/s, =150ms, = 350ms, and = 50N yielded the best results for both patients. With increasing fatigue towards the end of a session, was increased to 200μs.

In Fig. 3, a step taken with the closed-loop system is shown. At t=231.55s (labeled with a circled 1), the patient presses the button, as the left insole signal is larger than right, the controller switches from Stand to FlexionR and the flexion reflex is elicited. As soon as the knee exceeds the flexion threshold (2), phase KneeExt1 is activated and the knee is extended by KEC. When the right insole signal exceeds its threshold (3), it is switched to phase KneeExt2, and hip extensors are activated. As soon as the knee gets into state HYPEREXTENSION (4), KEC reduces the quadriceps pulse width. As the knee stays in HYPEREXTENSION, PW_quad remains at PW_min_..

Figure 3: Right step. Top to bottom: Finger switch state and gait phase, pulse widths, knee angle, insole signals.

In Fig. 4, the behavior of KEC can be observed. At instances of time labeled 1, 4, and 6, the knee state becomes BUCKLE and PW_quadis increased immediately.

Figure 4: Slight knee buckling compensated for by KEC.

At labels 2, 3, 5, and 7, HYPEREXTENSION is achieved and PW_quad can be reduced in all except one (2) cases to PW_quad,min.

Discussion and Conclusions

The new system was accepted very well by our patients. They accustomed to it easily, and appreciated the ability to better control their movements while being supported in as opposed to being forced to a movement.

Transition tFlexion adds a safety feature avoiding accidentally triggering a step. Transition tKneeExt1 unburdens the patient from switching off the flexion reflex, and adapts intrinsically to reflex habituation at the same time. Floor contact detection in tKneeExt2 activates the hip extensors at the appropriate times, thus, prevents back swing of the leg. KEC is capable of compensating for knee buckling before patients even notice, most frequently during single support, and advanced muscle fatigue. Stimulation intensities required to maintain safe knee extension can be reduced significantly. Calculations of the theoretical open-loop stimulation pattern based on the actually recorded closed-loop gait pattern revealed that the integral of PW_quad over time, which can be interpreted as an energy equivalent, can be reduced to approx. 55%, depending on PW_quad,min and PW_quad,max.

Mulder et al. [4] applied a controller similar to KEC to control quiet standing only. They applied only one threshold to increase or decrease PW_quad., and observed the occurrence of limit cycles. KEC is active in both static and dynamic gait phases with loaded and unloaded knees. By specifiying the two thresholds appropriately, we were able to avoid limit cycles.

In this study it could be demonstrated that applying cooperative closed-loop control can improve walking performance and safety while reducing muscle fatigue. The simplicity of the sensor system has the potential to improve patient acceptance. Future work will concentrate on the automatic adaptation of PW_quad,min and PW_quad,max to further optimize energy expenditure.

References

[1] P.E. Crago, R.F. Kirsch, R.J. Triolo (1999) Movement Synthesis and Regulation in Neuroprosthesis (Ch. 42) in: J.M. Winters and P.E.Crago (eds.) Biomechanics and Neural Control of Posture and Movement. Springer, New York

[2] R. Riener and Th. Fuhr (1998) Patient-Driven Control of FES-Supported Standing Up: A Simulation Study. IEEE-TRE 6: 113-124.

[3] T. Fuhr and G. Schmidt (1999) Design of a patient mounted multi-sensor system for lower extremity neuroprostheses. Proc. 21st IEEE EMBS Conf., Atlanta, October 13-16, 1999

[4] A.J. Mulder, P.H. Veltink, H.B. Boom, G. Zilvold (1992) Low-level finite state control of knee joint in paraplegic standing. J Biomed Eng 14:1 3-8

Acknowledgements: The authors would like to thank their patients for participating in this study, Mrs. Grigorean, MSc, PT, and Mr. Kämpf for their valuable support in performing the experiments.

This study was funded in part by the German Research Foundation within the Collaborative Research Center 'Sensory Motor Systems', SFB 462, project A1.