Lower-Limb Motor Imagery-Based Brain-Computer Interface to Control Treadmill Velocities

Lower-limb rehabilitation traditionally relies on physical therapy, but motor imagery(MI)-based brain- computer interfaces (BCIs) promise to facilitate neuroplasticity and adaptation by closing the perception-action cycle. Here, we present a BCI system based on kinesthetic MI that enables treadmill velocity control, establishing a closed-loop feedback mechanism. The system was tested in a healthy participant translating mu (8-12 Hz) and high-beta (18-24 Hz) rhythm modulation into treadmill velocity control commands. Feature extraction techniques, including power spectral density (PSD) and Riemannian geometry (RG), were used for MI- and resting state classification. Additionally, Logistic Regression (LR), k-nearest neighbors, support vector machine, and Linear Discriminant Analysis (LDA) were employed and optimized for accuracy. The results showed increased mu and highbeta activation modulation at the vertex. The online RG+LDA classifier achieving an average accuracy of 72%, while the pseudo-online RG+LR reached 95%. The study's novelty lies in combining kinesthetic MI with treadmill control and employing RG for feature extraction, demonstrating its potential to enhance cortical modulation during rehabilitation. Future work will have to validate the system in poststroke patients for clinical applicability.