PROJECT TITLE :

Leveraging Deep Learning Techniques to Improve P300-Based Brain Computer Interfaces

ABSTRACT:

The Brain-Computer Interface, or BCI, is a technology that has matured to the point where it can successfully connect the human brain to an external device. The so-called P300 wave can be extracted from electroencephalography (EEG) recordings, which is the basis for one of the most widely used protocols for brain-computer interfaces (BCIs). An event-related potential is known as a P300 wave if it has a latency of 300 milliseconds after the beginning of an unusual stimulus. In this paper, we improved P300-based brain-computer interfaces by making use of deep learning architectures, specifically convolutional Neural Networks (CNNs). P300 classification accuracy performances of the best state-of-the-art classifier were improved by our novel BCI classifier, which we have dubbed P3CNET. In addition, we investigated various pre-processing and training options that made BCI systems more usable. When we were doing the pre-processing of the EEG data, we looked into finding the optimal signal interval that would lead to higher accuracy in the classification. Then, we investigated the bare minimum required for the number of calibration sessions in order to achieve the optimal balance between increased accuracy and decreased calibration time. We analyzed the saliency maps of the input EEG signal that led to a correct P300 classification in order to improve the explainability of deep learning architectures. During this process, we discovered that removing less informative electrode channels from the data did not result in improved accuracy. This was one of the findings that we made. The generalizability of the obtained results was demonstrated by performing all of the methodologies and explorations and validating them on two different CNN classifiers. Finally, we demonstrated the benefits of Transfer Learning by applying the proposed novel architecture to additional P300 datasets and demonstrating its effectiveness. BCI practitioners can improve the effectiveness of their work by making use of the presented architectures and suggestions for practical applications.