Decoding Continue Trajectory of Hand Circular Motion from EEG Signals

Most studies on continuous hand movement decoding rely on center-out paradigms that do not reflect natural motion patterns. This study investigates the feasibility of decoding continuous circular hand movement trajectories in two-dimensional space, including both position and velocity, from EEG signals. Participants performed circular tracking tasks while EEG data were recorded. Three commonly used deep learning models—EEGNet, EEGNet+LSTM, and DeepCon-vNet+LSTM—were employed to decode the motion trajectories. Model performance was evaluated using correlation coefficients (CC) and mean squared error (MSE) between predicted and actual values. Eight healthy adults participated in the experiment. All models achieved statistically significant results above chance level (p< 0.05). The best-performing model achieved average CC of 0.87 and 0.88 for positions (Px and Py), and 0.85 and 0.84 for velocities (Vx and Vy), respectively. These findings demonstrate the potential of EEG-based decoding of continuous circular hand movements and support its application in natural movement prediction for brain–computer interface systems.