Motor Intentioan Decoding Based On Electroencephalogram Feature Optimization And Synchronization Analysis

Brain-computer interface (BCI) system can provide a direct control or sensory channel that do not rely on peripheral neural and muscular system between brain and external equipments. BCI technologies are expected to restore the motor functionality of patients with paralytic injury or disease, and to provide humans with new way of interaction. The base of BCI technologies are brain intention identification by means of neural signal decoding. For BCIs based on electroencephalogram (EEG), the signal utilized is a synthesis of multiple rhythmic field potentials, so the analysis for understanding the properties of EEG rhythm plays an important role in related BCI decoding methods.This dissertation focuses on the property of rhythmic signal components and methods for its feature extraction. The resulting algorithm is used in motor intention decoding.In the analysis of EEG rhythm components, the relationships between information frequency, temporal and spatial EEG domains have been studied. The dissertation proposed a comprehensive method to optimize the features in all three domains simultaneously. The new method is designed to achieve a better tradeoff between EEG pattern recognition accuracy and generalization capacity of the algorithm, as the proposed method used a data-driven method to achive varied accuracy for different features. The experiments with motor execution and motor imagery EEG data showed that the proposed algorithm can achieve a better user-specific recognition of the motor-related EEG patterns (average rate of correct recognition improved more than 10% and 6% for two experiments), while at the same time remains the system's generalization capacity-which is two important aspects of BCI application.The synchronization feature of multi-channel EEG is also studied in this dissertation, as it embodies spatial properties of EEG besides component distribution. Synchronization between EEG channels reflects neural synchrony between brain regions, and neural synchrony is believed to support a series of important brain functionality. However there are few reports about BCI algorithm using overall synchronization pattern. A feature extraction method based on connectivity network of synchrony is proposed, which first calculates the synchronization value between electrode pairs, then generates the network topography measures to discribe the network synchronization pattern. In order to evaluate this method, the dissertation also proposed a modified rhythmic signal synchonization model. The simulation data generated with the model is used to set the parameters of the feature extraction method.Based on the above studies of rhythmic signal, the synchronization feature method is combined with the EEG component feature optimization to limite signal time-frequency range. Tests with real motor-related EEG data showed that the combination would largely improve the performance of synchronization method.