Classification of left and right knee extension motor imagery using common spatial pattern for BCI applications

Research on the deployment of various cognitive tasks in the experimental paradigm of a human brain-computer interface (BCI) is on-going in particular upper-limb tasks. Less has been investigated on the lower-limbs, due to its somatotopic arrangement in the sensorimotor cortex compared to that of upper limbs. Hip, knee, foot and toes share spatial proximity with each other. We therefore investigated the possibility to deploy the left vs. right knee extension kinaesthetic motor imagery (KMI) as a cognitive task in a BCI for the control of lower-limbs, primarily for people with neurodegenerative disorders, spinal cord injury or lower-limb amputation. The method involved feature extraction using common spatial pattern (CSP), and filter bank common spatial pattern (FBCSP) algorithm for the optimization of individual spatial patterns. This was followed by supervised machine learning using logistic regression (Logreg) and linear discriminant analysis (LDA) for classification of tasks. The paradigms resulted in four combinations/methods for discriminating between left and right knee tasks. The FBCSP-Logreg outperformed remaining paradigms with a maximum accuracy of 70.00% ± 2.85 and kappa=0.40. The results elicit the possibility to deploy left vs. right knee extension KMI in a 2-class BCI for controlling robotic/prosthetic knee.