Multitask Lower Limb Classification System for Dorsiflexion Rehabilitation by Using Electroencephalography and Functional Electrical Stimulation

Providing alternatives for lower limb moment control in post-stroke or post-Spinal Cord Injury (SCI) patients who have suffered a severe neural impairment many years ago can further improve the effect of rehabilitation. This work proposes a multitask system for dorsiflexion of the right, left, and both feet through Electroencephalography and Functional Electrical Stimulation (FES). A Butterworth bandpass filter between 8 and 30 Hz is implemented, and subsequently, three strategies are developed by first considering the power spectral density features in mu, low-beta, and high-beta bands responsible for motor task execution. Moreover, the second strategy uses Common Spatial Patterns filters, and the third strategy uses Riemannian Geometry (RG) to extract features. Four distinct classifiers were explored: k-nearest neighbors, support vector machine, Linear Discriminant Analysis (LDA), and decision tree. Based on Accuracy (ACC), True Positive Rate (TPR), and False Positive Rate (FPR) were evaluated the classifiers and the strategies implemented by achieving an ACC of 0.79, a TPR of 0.89 and a FPR of 0.05 employing RG + LDA. The implemented methodology allows the feasibility of a Human-Machine Interface (HMI) which implements a lower-limb MI task recognition system to be used in rehabilitation. Future work will implement the recognition system in real-time to users move the desired limb by activating a FES that generates contraction.