Analysis and Recognition of Brain Networks for EEG-based Upper-limb Motion

Recent advances in Brain-Computer-Interface (BCI) have allowed us to construct robotic arms for remote control and rehabilitation systems. The analysis and recognition of motor-related brain signals and activities are the key of such BCI systems. However, most previous studies are powerless to deeply reveal the working mechanism of the brain during motion executions and adopt low-accuracy strategies for modeling and pattern recognition. With the introduction and development of network neuroscience in the field of BCI, it allowed us to analyze brain activities for motor-related EEG signals and build principled models to improve the performance of feature extraction and pattern recognition. Therefore, this study concentrates on the construction and analysis of brain networks. Partial directed coherence (PDC) was employed to construct brain networks based on EEG signals for analyzing and recognizing brain activities of different movements (hand opening, hand closing, elbow flexion, and wrist flexion). Features were extracted from brain networks for pattern recognition, and Random Forest (RF) classifier was used to classify different movements. The results indicate that different motion patterns can be related to corresponding network structures and illustrates significant differences among information transmissions. In addition, the results of motion recognition based on RF achieve an accuracy of 75.5%, which demonstrate that the proposed method based on brain networks is reliable to analyze EEG signals and can improve the performance of motion recognition.