A Low-Cost Lower-Limb Brain-Machine Interface Triggered by Pedaling Motor Imagery for Post-Stroke Patients Rehabilitation

• Published in: IEEE transactions on neural systems and rehabilitation engineering Vol. 28; no. 4; pp. 988 - 996
• Main Authors: Romero-Laiseca, Maria Alejandra, Delisle-Rodriguez, Denis, Cardoso, Vivianne, Gurve, Dharmendra, Loterio, Flavia, Posses Nascimento, Jorge Henrique, Krishnan, Sridhar, Frizera-Neto, Anselmo, Bastos-Filho, Teodiano
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.04.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Brain; Brain-computer interface; Brain-computer interfaces; brain-machine interface; Clinical trials; Computational neuroscience; Computing costs; Discriminant analysis; EEG; Electroencephalography; Feature extraction; Image recognition; Imagery; Latency; Low cost; lower-limb rehabilitation; Man-machine interfaces; Mental task performance; motor imagery; On-line systems; Patient rehabilitation; pedaling; Plasticity (neural); Presenilin 1; Presenilin 2; Recovery; Rehabilitation; Relearning; Stroke; Task analysis
• ISSN: 1534-4320; 1558-0210; 1558-0210
• DOI: 10.1109/TNSRE.2020.2974056

A low-cost Brain-Machine Interface (BMI) based on electroencephalography for lower-limb motor recovery of post-stroke patients is proposed here, which provides passive pedaling as feedback, when patients trigger a Mini-Motorized Exercise Bike (MMEB) by executing pedaling motor imagery (MI). This system was validated in an On-line phase by eight healthy subjects and two post-stroke patients, which felt a closed-loop commanding the MMEB due to the fast response of our BMI. It was developed using methods of low-computational cost, such as Riemannian geometry for feature extraction, Pair-Wise Feature Proximity (PWFP) for feature selection, and Linear Discriminant Analysis (LDA) for pedaling imagery recognition. The On-line phase was composed of two sessions, where each participant completed a total of 12 trials per session executing pedaling MI for triggering the MMEB. As a result, the MMEB was successfully triggered by healthy subjects for almost all trials (ACC up to 100%), while the two post-stroke patients, PS1 and PS2, achieved their best performance (ACC of 41.67% and 91.67%, respectively) in Session #2. These patients improved their latency (2.03 ± 0.42 s and 1.99 ± 0.35 s, respectively) when triggering the MMEB, and their performance suggests the hypothesis that our system may be used with chronic stroke patients for lower-limb recovery, providing neural relearning and enhancing neuroplasticity.