Selection of Temporal Features for the Detection of Movement Intention in patients with Amyotrophic Lateral Sclerosis

Detection of movement intention is a critical step for the development of rehabilitation systems and the induction of plasticity using brain-computer interfacing. The movement-related cortical potential (MRCP), obtained from electroencephalography (EEG) signals, is an attractive modality for movement detection as it can be generated 1-2 s prior to the movement execution. In the present study, monopolar EEG signals were recorded from ten channels in five Amyotrophic Lateral Sclerosis (ALS) patients and ten healthy participants while performing hand movement (hand extension/flexion). Movements were detected offline by using classification (Support Vector Machine) between movement and rest epochs with three different groups of time-domain features. The first group of features were time samples of filtered down-sampled EEG (raw features) while the second group (computed features) included the features calculated from extracted MRCPs and rest epochs (e.g., slope, peak negativity and variations in different time segments). In the third condition, the two groups of features were combined. The results revealed that detection accuracy obtained from raw features (88±3%) was higher than either computed features (83±3%) or a combination of the two (84±3%). Therefore, the time samples of EEG signals seem to be a better choice for movement detection using MRCPs.