Adaptive multichannel FES neuroprosthesis with learning control and automatic gait assessment

Abstract

Abstract Background FES (Functional Electrical Stimulation) neuroprostheses have long been a permanent feature in the rehabilitation and gait support of people who had a stroke or have a Spinal Cord Injury (SCI). Over time the well-known foot switch triggered drop foot neuroprosthesis, was extended to a multichannel full-leg support neuroprosthesis enabling improved support and rehabilitation. However, these neuroprostheses had to be manually tuned and could not adapt to the persons’ individual needs. In recent research, a learning controller was added to the drop foot neuroprosthesis, so that the full stimulation pattern during the swing phase could be adapted by measuring the joint angles of previous steps. Methods The aim of this research is to begin developing a learning full-leg supporting neuroprosthesis, which controls the antagonistic muscle pairs for knee flexion and extension, as well as for ankle joint dorsi- and plantarflexion during all gait phases. A method was established that allows a continuous assessment of knee and foot joint angles with every step. This method can warp the physiological joint angles of healthy subjects to match the individual pathological gait of the subject and thus allows a direct comparison of the two. A new kind of Iterative Learning Controller (ILC) is proposed which works independent of the step duration of the individual and uses physiological joint angle reference bands. Results In a first test with four people with an incomplete SCI, the results showed that the proposed neuroprosthesis was able to generate individually fitted stimulation patterns for three of the participants. The other participant was more severely affected and had to be excluded due to the resulting false triggering of the gait phase detection. For two of the three remaining participants, a slight improvement in the average foot angles could be observed, for one participant slight improvements in the averaged knee angles. These improvements where in the range of 4 c i r c at the times of peak dorsiflexion, peak plantarflexion, or peak knee flexion. Conclusions Direct adaptation to the current gait of the participants could be achieved with the proposed method. The preliminary first test with people with a SCI showed that the neuroprosthesis can generate individual stimulation patterns. The sensitivity to the knee angle reset, timing problems in participants with significant gait fluctuations, and the automatic ILC gain tuning are remaining issues that need be addressed. Subsequently, future studies should compare the improved, long-term rehabilitation effects of the here presented neuroprosthesis, with conventional multichannel FES neuroprostheses.

Keywords

• Multichannel neuroprosthesis
• unctional electrical stimulation
• Iterative learning control
• Automatic gait assessment
• Real-time motion analysis
• Joint angle tracking