Inertial Gait Phase Detection for control of a drop foot stimulator: Inertial sensing for gait phase detection
Introduction
The effects of stroke (CVA) on the life of an individual can be very dramatic as it disables him physically as well as mentally. Physically, the motor control of one side of the body is often affected. One of the most common consequences is the inability to voluntarily lift the foot (Drop foot).
Since Liberson's [1] proposal in the sixties to stimulate the dorsiflexors for foot lift during the swing phase of gait, many proposals have been made resulting in various commercial systems such as the MikroFES (Josef Stefan Institute, Slovenia), the Odstock (Salisbury DH NHS, United Kingdom), the ActiGait (neurodan/Aalborg), the WalkAide (Innovative Neurotronics, USA) and the Twente two channel implantable drop foot stimulator (Roessingh Res. & Dev., Netherlands/Finetech Medical, United Kingdom). This paper describes the development of an Inertial Gait Phase Detection system (IGPD) that could be used in any of these systems to replace the current triggering method, namely the heel switch. A reliable gait phase detection system is essential for the correct triggering of stimulation and consequently the amelioration of gait.
Many methods have been reported for the control of drop foot stimulators, namely manual switches, force sensitive resistors (FSR's), inclinometers, gyroscopes, accelerometers, EMG and implantable nerve cuff electrodes [2]. An overview is shown in Table 1.
The current design of most drop foot stimulators uses a footswitch to control stimulation as proposed by Liberson et al. [1] which is shown to suffer from a number of deficiencies as described by Lyons and Sinkjaer [2]. This method requires wires to connect to the control box which in turn increases its chances for mechanical failure. Another problem of this design is the necessity to wear shoes in order to keep the switch in position as well as the associated problems of the switch moving out of position. This system is also limited to heel off and heel strike detection on flat surface imposing a limitation to the user. The use of footswitches was also dismissed by Ott et al. [3] and Popovic et al. [4] for having poor detection reliability.
Several researches showed that alternative sensors could be used to replace footswitches providing for more accurate detection and allowing for more flexibility in natural variation such as shoe type.
The use of accelerometers and gyroscopes were found to be very promising in the reliable detection of gait phases and their size was found to be ideal for compact low power systems and possible implantation.
From these papers it can be shown that accelerometers and gyroscopes have been used on many occasions for gait phase detection singularly and in combinations and have shown results to suggest that they are optimum for the intended purpose. Many of the authors report high reliability and reproducibility of their results. Their physical dimensions and operational requirements are minimal allowing for their integration into the drop foot stimulators positioned on the leg and have proven to be able to produce the necessary detection results. Their use can be over a 24-h period. A variable found in all papers was the location of the sensors which although focused mainly on four locations (the pelvis, thigh, ankle/foot and shank) was not the same for each application. The number of sensors was also a variable in most cases. Since many drop foot stimulators have positioned their external control units on the outer side of the upper shank (Fig. 1) this is considered to be the optimum location for the sensors for this application. Although some authors have positioned sensors in this location a complete analysis of the optimal inertial sensor set and algorithms in this position is lacking.
The goal of this project is the distinct and reliable detection of gait phases and transitions to provide a reliable gait phase detection system that is able to operate on all terrains and during all normal daily activities. The system should be small with low power consumption for integration into current drop foot stimulators and potential future implantation. The use of inertial sensors such as accelerometers and gyroscopes is to be evaluated and the optimal configuration of these positioned at the upper shank is to be found. The algorithms evaluated are required to run on the lowest possible computational load for implementation in embedded processing systems. The algorithm system proposed is to be incorporated in current drop foot systems hence the physical dimensions of the sensors and additional hardware was considered. The location of the sensors is to be that of the control unit of most drop foot stimulators allowing the user greater freedom without the necessity of footwear and connecting wires.
Section snippets
Finite state description of gait
Gait is a cyclical movement and can be represented by a simplified state transition diagram as shown in Fig. 2. This model is based on the work of Willemsen et al. [11]. Human Gait can be broken down into three states, namely Stance (ST), Push off (PO) and Swing phase (SW). The instances that indicate transition between these states are Heel Off (HO), Toe Off (TO) and Heel Down (HD) respectively as shown in Fig. 2.
Although this sequence of phases is mostly true for flat surface walking it is
Heel switch performance against Vicon and force plates
The heel switch triggering used to compare the algorithms is done with the use of a threshold applied to the signal using the moving average baseline described before as a trigger level and hence cannot be considered accurate for the timing of heel off and heel down. This can be better seen in the heel switch signal (Fig. 3). This is also the method applied by many stimulators using a heel switch as a trigger. In the case of heel off, the signal is shown to be a slow weight shifting transition.
Discussion
The initial results of four IGPDA's were presented using accelerometers and one gyroscope. The algorithms showed good detection results over a number of activities compared to the current triggering method of using the heel switch. The detection achieved was 100% for all algorithms during flat surface and rough terrain walking. Algorithms IGPDA-1 and IGPDA-2 showed false detections for Carpet (damped) walking where as IGPDA-3 and IGPDA-4 showed full detection during this activity. The worst
Acknowledgements
The authors acknowledge the financial support of the EU Sparsed Research Training Network neuralPRO as well as the EU Project TUBA. Special thanks to Paul Taylor from Salisbury D.H. NHS and Laurence Kenney from University of Salford for their assistance.
Conflict of interest
The authors find no conflict of interest related to this paper and its publication in the Journal of Medical Engineering & Physics.
References (21)
- et al.
Automatic detection of gait events: a case study using inductive learning techniques
J Biomed Eng
(1989) - et al.
Above- and below-lesion EMG pattern mapping for controlling electrical stimulation of paraplegics to facilitate unbraced walker-assisted walking
J Biomed Eng
(1988) - et al.
Spatio-temporal parameters of gait measured by an ambulatory system using miniature gyroscopes
J Biomech
(2002) - et al.
Measurement and reduction of noise in kinematics of locomotion
J Biomech
(1974) - et al.
Functional electrotherapy: stimulation of the peroneal nerve synchronized with the swing phase of the gait of hemiplegic patients
Arch Phys Med Rehab
(1961) - et al.
A review of portable FES-based neural orthoses for the correction of drop foot
IEEE Trans Neural Syst Rehabil Eng
(2002) - et al.
Comparison of foot-switch and hand-switch triggered FES correction of foot drop
- et al.
Gait identification and recognition sensor
- et al.
A fuzzy logic gait event detector for fes paraplegic gait
- et al.
Machine learning in control of FES systems for locomotion
IEEE Trans Biomed Eng
(1995)
Cited by (131)
Abnormal gait partitioning and real-time recognition of gait phases in children with cerebral palsy
2023, Biomedical Signal Processing and ControlThe effect of harvesting the anterior half of the peroneus longus tendon on foot morphology and gait
2024, Journal of Orthopaedic Surgery and ResearchSpeed-Variable Gait Phase Estimation During Ambulation via Temporal Convolutional Network
2024, IEEE Sensors JournalA Method of Using Pressure Insoles for Foot Drop FES Gait Phase Detection
2024, Lecture Notes in Electrical Engineering