Continuous Tracking using Deep Learning-based Decoding for Non-invasive Brain-Computer Interface
Brain-computer interfaces (BCI) using electroencephalography (EEG) provide a non-invasive method for users to interact with external devices without the need for muscle activation. While noninvasive BCIs have the potential to improve the quality of lives of healthy and motor impaired individuals, they currently have limited applications due to inconsistent performance and low degrees of freedom. In this study, we use deep learning (DL)-based decoders for online Continuous Pursuit (CP), a complex BCI task requiring the user to track an object in two-dimensional space. We developed a labeling system to use CP data for supervised learning, trained DL-based decoders based on two architectures, including a newly proposed adaptation of the PointNet architecture, and evaluated the performance over several online sessions. We rigorously evaluated the DL-based decoders in a total of 28 human participants, and found that the DL-based models improved throughout the sessions as more training data became available and significantly outperformed a traditional BCI decoder by the last session. We also performed additional experiments to test an implementation of transfer learning by pre-training models on data from other subjects, and mid-session training to reduce inter-session variability. The results from these experiments showed that pre-training did not significantly improve performance, but updating the models mid-session may have some benefit. Overall, these findings support the use of DL-based decoders for improving BCI performance in complex tasks like CP, which can expand the potential applications of BCI devices and help improve the quality of lives of healthy and motor-impaired individuals. Significance StatementBrain-computer Interfaces (BCI) have the potential to replace or restore motor functions for patients and can benefit the general population by providing a direct link of the brain with robotics or other devices. In this work, we developed a paradigm using deep learning (DL)-based decoders for continuous control of a BCI system and demonstrated its capabilities through extensive online experiments. We also investigate how DL performance is affected by varying amounts of training data and collected more than 150 hours of BCI data that can be used to train new models. The results of this study provide valuable information for developing future DL-based BCI decoders which can improve performance and help bring BCIs closer to practical applications and wide-spread use.