SmartStim: A Recurrent Neural Network Assisted Adaptive Functional Electrical Stimulation for Walking.

Abstract

According to the Neuro Patience report of the Neurological Alliance, 1 in 6 people in the UK has a neurological condition. With the growth in technology, rehabilitation for neurological problems is one of the fastgrowing fields. Functional Electrical Stimulation (FES) is one of those neuro-rehabilitation methods that uses electrical nerve stimulation to restore functional muscle movements that are lost due to neurological problems such as stroke and multiple sclerosis. This neuroprosthetic device is frequently used to assist walking by treating a condition called Drop Foot, a result of paralysis of the pretibial muscles. This study proposes a two-channel FES device called the SmartStim, which has the ability to modulate its stimulation levels according to various obstacles such as stairs and ramps. This system employs a sensor-based module with a Recurrent Neural Network to classify these different walking scenarios. The module is built with Inertial Measurement sensors embedded in a pair of shoes, and the Recurrent Neural Network uses data from these sensors to predict various obstacles as the user is walking. These predictions are then used by a Fuzzy Logic Controller to control and regulate the stimulation current in two channels of the SmartStim system. In the two channels of the system, one channel will help aid with drop foot, while the other will be used to stimulate another muscle group to help access stairs and ramps by the user. The Recurrent Neural Network module in this system has been trained and tested using the k-fold cross-validation. The evaluation of this trained model shows that it can predict obstacles from sensor data at 97 percent accuracy. Currently, further testing is being performed to assess the workings of the fuzzy logic controller in combination with the Recurrent Neural Network in healthy individuals. It is expected that the SmartStim system may aid users in accessing various walking scenarios more efficiently.