سيدموسي حسيني

The main objective of brain-computer interfaces (BCIs) is to translate brain activity patterns into meaningful control commands which enable the users to interact with the environment without using the natural neuromuscular pathways. An important category of BCIs is designed based on extracting information from sensorimotor rhythms (SMRs) and movement-related cortical potentials (MRCP). Hence, decoding movement-related information from brain signals is of utmost importance. The principal objective of this thesis is to decode hand movement trajectories from EEG signals. To this end, a cue-based center-out reaching task with four orthogonal targets was designed. EEG data were recorded from twelve healthy male subjects performing the task. To investigate the generalizability of the model, we incorporated in our center-out task a number of random targets happening in random directions. In the first study, we eva‎luated the capability of recorded data in discriminating four classes of hand reaching using graph theory features extracted from movement and pre-movement intervals. This approach achieved better classification accuracy compared to conventional amplitude features. The second study aimed to investigate the feasibility of using phase-based connectivity features in decoding continuous hand movements from EEG signals. The phase-locking value (PLV) and magnitude-squared coherence (MSC) were exploited as connectivity features along with multiple linear regression (MLR) for decoding hand trajectories. The best channel pairs for extracting features related to hand movements were selected via a proposed search scheme. The results reveal that the regression models based on PLV and MSC features achieve the average Pearson correlations of 0.43 ± 0.03 and 0.42 ± 0.06, respectively, between predicted and actual trajectories over all subjects. The regression model based on amplitude features has achieved the Pearson correlation of 0.31 in contrast. The connectivity features extracted from delta and alpha bands had the most contribution in reconstructing hand trajectories via linear model. In the third study, a state-based decoding scheme based on cascading a discrete state decoder with a continuous decoder was introduced to enhance the prediction of hand trajectories. The binary discrete state decoder comprised common spatial pattern (CSP) features extracted from pre-motor potentials of pre-movement data and support vector machine (SVM) classifier. Two non-parametric continuous decoders based on Gaussian process regression (GPR) have been employed for the continuous decoding of hand movements along each orthogonal axis using the envelope features of EEG signals. The proposed decoding paradigm trained on four orthogonal targets was also generalized to identify and predict the trajectories of random targets. The proposed state-based method achieved a mean correlation coefficient of 0.54 between actual and predicted trajectories for orthogonal targets over all subjects. The trajectories of random targets were also decoded with a mean correlation coefficient of 0.37. In this thesis, new feature extraction techniques and improved decoding paradigms were employed to enhance the results of hand movement decoding based on EEG signals. The proposed methods and approaches could open new possibilities in developing novel types of neuroprostheses for rehabilitation purposes.

واسط مغز-رايانه , الكتروانسفالوگرام , رمزگشايي پيوسته , ارتباطات مغزي , رگرسيون

Brain-computer interface , electroencephalography , continuous decoding , brain connectivity , regression

Seyyed Moosa Hosseini

Vahid Shalchyan