“Analysis of EEG Recordings During Grasp and Lift (GAL) Trials

• Main Author: Musa, Huwaida
• Format: Final Year Project
• Language: English
• Published: 2018
• Online Access: http://utpedia.utp.edu.my/19173/1/Final%20Report%20FYP.pdf; http://utpedia.utp.edu.my/19173/

EEG stands for Electroencephalography is a medical technique that is used to extract the signal in the brain by using a special medical device known as electroencephalogram. EEG signal is highly noisy as the signal is recorded from a scalp electrode that is influenced by different deeper brain structure. The changes of the signal with time make the signal to be non-stationary. The idea of using the EEG signal to extract the signals which related to object manipulation is reasonable as most of the human cortex involves in basic motor tasks of a person. However, until now there is uncertainty regarding what extent the extraction of signals from the human brain can be used in the monitoring and control purposes. The purpose of analysing this EEG data is to help develop a prosthetic device that can control an upper limb and generate a power grasp or a pinch grasp involving the thumb and index finger. Specifically for this project, we will focus on the process of analysing the EEG signal that includes the extraction of data, filtering data, applying feature extraction and performing classification of the EEG signal. This paper will review the step by step process involved in this project which is to classify the 6 different events or movement of grasping and lifting using the EEG recording of grasp and lift trial. The classification of the signals is done by using various classifiers that differentiate each of the 6 events and compute the percentage of accuracy for the correctly predicted events. There are two types of investigation which are classification with feature extraction and without feature extraction. Throughout this project, we managed to get the classification accuracy of 6 events for 26.6% and 58.5% while for 2 events of up to 67.4% and 96% using 6 channels for both algorithms respectively.