Background: People suffering from major motor deficiencies such as amyotrophic lateral sclerosis (ALS) can not benefit from usual assistive devices since these devices require some motor function. Therefore a brain-computer interface could be of help to provide an assistive device which do not require voluntary movements, but motor intention. This would help the ALS patients to regain as much independence as well as improving their quality of life. In the present study, a random forest (RF) model was developed in order to detect and classify reaching and grasping motions for control of a robotic arm.
Methods: The RF model was developed using data sets across three sessions which consisted of 1)performing reaching movement 2) performing grasping movement and 3) performing both reaching and grasping movements. EEG signals were recorded across nine channels and motor-related potentials (MRCP) were extracted.
Results: The results showed it was possible to detect both reaching and grasping movements with an average true positive rate (TPR) of respectively 90.63\% $\pm$ 18.43\% (mean $\pm$ standard deviation) and 84.72\% $\pm$ 21.76\% across seven subjects. The classification accuracy for reaching and grasping based on two subjects which participated in session two were an average of 74.82\% $\pm$ 11.44\% across both subjects.
Conclusion: By using a RF model, it was possible to detect and classify reaching and grasping motions. The results shows promising results of utilizing a MRCP-based BCI for ALS patients to control a robotic arm.