Abstract:
Reaching and Grasping is most signi cant component of human life.Translation
of EEG in the form of upper limb movement is of great importance for realization
of natural neuroprosthesis control and restoration of hand movements of patients
with motor disorders. Patients su ering from spinal cord injury (SCI)problems
have lost most of voluntary motor control functions. Such type of loss can be cured
using movement related cortical potentials (MRCPS) analysis. Brain computer
interface with limb neuro-prosthesis is considered as a solution to such problems.
This study anlyzes EEG signals in relation with natural reach and grasp actions.
EEG signals have movement related cortical potentials (MRCPS) which
can be used to decode upper limb movements. This experiment was performed
in Graz University of Technology Austria and they o ered free access dataset
for further exploration.Total 45 subjects were involved in this study, 15 subjects
with every type of electrode:gel,water and dry performed the experiment. All
subjects accomplished self-initiated 80 reach and grasp actions toward a spoon
within the jar (lateral grasp) and toward an empty glass (palmar grasp).EEG
signals are recorded using three types of electrodes: water based, Gel based and
Dry electrodes. In this study signals are classi ed using Deep learning technique
i.e Convulotional Neural Networks.
For analysis, EEG signals were preprocessed using various lteration techniques.
After ltration data is fed into classi er for classi cation of signals. Data
is divided into test set and training set. Grand average peak accuracy calculated
on unseen test data resulted in 54.2% classi cation accuracy i.e Gel based accuracy
approached 56.8.4%, water based 52.7% and dry based 51.8%.
