Improving EEG Classification Accuracy of Single Limb Imaginary Movements by Leveraging an Established EMG Pattern Recognition Scheme

Brain-computer interfaces may enable the collaboration between human and machines. They can in fact potentially translate the electrical activity of the brain to understandable commands to operate a machine or a device.In this thesis, a method is proposed to improve the accuracy of a BCI by leveraging an established electromyography (EMG) pattern recognition scheme. Such a pattern recognition scheme extracts time-domain features, which include autoregressive (AR) model coefficients, root mean square (RMS), waveform length (WL) and classifies them using an optimized support vector machine (SVM) with a radial basis kernel function (RBF).Upon validating that such a method can indeed process EMG signals to classify different fifteen movements of the arm with high accuracy (> 90%), this thesis investigates the possibility of implementing it for the design of a BCI based on electroencephalographic (EEG) signals. The discrimination of rest, imaginary grasp and imaginary elbow movement of the same limb was selected as test case to validate the designed BCI. This classification task is particularly challenging because imaginary movements within the same limb have close spatial representations on the motor cortex area.The use of the proposed method was demonstrated to be suitable for identifying imaginary movements using EEG signal. In fact, the investigated method achieved an average accuracy of 91.8 % and 90 % for identifying the user intention corresponding to imaginary grasps and imaginary elbow movements (2-class BCI). An average classification accuracy of 74.2 % was achieved for classification of rest versus imaginary grasps versus imaginary elbow movements (3-class BCI).The investigated method outperformed methods that are widely used in the BCI literature such as common spatial patterns (CSP), filter bank CSP (FBCSP), and band power methods. These results are encouraging and the proposed method could potentially be used in the future in BCI-driven robotic devices to assist individuals with impaired upper extremity functions in performing daily tasks.