Real Time Foot Drop Correction using Machine Learning and Natural Sensors

The objective of this study was to investigate and test a real time system implemented for Functional Electrical Stimulation (FES) assisted foot drop correction, deriving control timing from signals recorded from a peripheral sensory nerve. A hemiplegic participant was attached with a cuff electrode on the sural nerve connected to a telemetry controlled implanted neural amplifier, and a stimulation cuff electrode on the peroneal nerve connected to an implanted stimulator. An input domain was derived from the recorded electroneurogram (ENG) and fed to a detection algorithm based on an Adaptive Logic Network (ALN) for controlling the timing of the peroneal stimulation. The detection system was tested in real time over a period of 392 days, covering a variety of walking tasks. The detection system's ability to detect heel strike and foot lift without errors and to detect the difference between walking and standing proved to be stable for the duration of the study. We conclude that using ALNs and natural sensors provide a stable and accurate control signal for FES foot drop correction.