Adaptive learning in the detection of Movement Related Cortical Potentials improves usability of associative Brain-Computer Interfaces.

Brain-computer interfaces have increasingly found applications in motor function recovery in stroke patients. In this context, it has been demonstrated that associative-BCI protocols, implemented by means the movement related cortical potentials (MRCPs), induce significant cortical plasticity. To date, no methods have been proposed to deal with brain signal (i.e. MRCP feature) non-stationarity. This study introduces adaptive learning methods in MRCP detection and aims at comparing a no-adaptive approach based on the Locality Sensitive Discriminant Analysis (LSDA) with three LSDA-based adaptive approaches. As a proof of concept, EEG and force data were collected from six healthy subjects while performing isometric ankle dorsiflexion. Results revealed that adaptive algorithms increase the number of true detections and decrease the number of false positives per minute. Moreover, the markedly reduction of BCI system calibration time suggests that these methods have the potential to improve the usability of associative-BCI in post-stroke motor recovery.