Robustness of movement detection techniques from motor execution: single trial movement related cortical potential

Alterations in attention are known to modify excitability of underlying cortical structures and thus the activity recorded during non-invasive electroencephalography (EEG). Brain-Computer-Interface systems for
neuromodulation are based on reliable detection of intended movements from continuous EEG signals, thereby generating real time feedback to induce neuroplasticity. We have shown in past studies that the movement related cortical potential (MRCP), a slow negative potential commencing 1-2 s prior to
movement, can be reliably detected in real time within a single session and used to drive an external device that reproduces the intended movement. In the present study, our main objective was to characterize movement detection performance of two algorithms, when the subject’s attention is altered. In nine
healthy participants the auditory oddball paradigm was used to modulate attention. All subjects completed a set of movement executions prior to and following the oddball paradigm. The locality preserving projection followed by the linear discriminant analysis (LPP-LDA) and the matched-filter (MF) technique were applied offline for detection of movement. Results show that LPP-LDA significantly outperformed MF. The robustness of the LPP-LDA method was demonstrated by a higher true positive rate (TPR), lower false positive rate
(FPR) and a shorter detection latency when attention levels were altered.