The use of artificial intelligence to identify thought messages via non-invasive EEG brain signals

Brain-Computer Interfaces (BCI) have opened up opportunities by advancing the technology, offering new possibilities in both practical applications and theoretical research. Individuals with Completely Locked-in Syndrome, which can result from conditions such as Motor Neurone Disease and Amyotrophic Lateral Sclerosis, stand to gain significantly from BCIs developments aimed at enhancing communication and overall well-being. This PhD research focuses on developing a system to recognise imagined thoughts through Electroencephalography (EEG) brain signals and Artificial Intelligence (AI), with the goal of implementing a novel methodology to establish a direct communication link between brain functionality and computer interfaces. Developing effective systems for transforming EEG signals into practical communication outputs for various mental tasks presents significant challenges in the field of signal processing. A novel approach, termed Automated Sensory and Signal Processing System (ASPS), is introduced for feature extraction and selection in EEG signal data. This method enhances the reliability of EEG-based communication by identifying and selecting the most relevant features for classification. The ASPS approach is initially implemented with an elementary model and tested through bespoke analysis. The study is subsequently scaled up by increasing the number of subjects, forming groups, and incorporating various domains analysis in signal processing and statistical functions. Artificial Neural Networks (ANNs) are employed for classification, simultaneously verifying the performance of the ASPS approach. The extracted features, generated as outputs of the ASPS approach, serve as inputs to the ANN. High-quality features that are consistent and distinguishable for each mental task facilitate high accuracy in brain signal classification, demonstrating the effectiveness of the feature extraction technique. In this study, feature extraction is significantly enhanced by the ASPS approach, leading to more accurate mental imagery recognition. These extracted features are classified using ANN algorithms, specifically Feed Forward Neural Networks (FFNN) and Learning Vector Quantisation, demonstrating high accuracy across bespoke, group-based, and combined analyses. Six different ANN architectures with various combination of neurons and hidden layers are employed. Additionally, Convolutional Neural Network, a widely used image processing technique, is utilised in another experiment to classify signals, demonstrating the capability to recognise imagined thoughts. Based on these architectures, different ANN and CNN models are trained and tested to identify the most optimised classifier for imagination recognition. The performance of these classifiers is summarised and compared to evaluate the robustness of the classification algorithms. Overall, the single-layered FFNN ensures very consistent and high accuracy in imagination recognition. Furthermore, EEG sensor optimisation is explored through extensive analysis, followed by a thorough validation of the optimised sensors. These optimised sensors simplify signal processing and enhance the accuracy of imagination recognition. Finally, an experiment with a novel product, the EEG-BCI prototype, introduces an optimised sensor-based interface that enables EEG recording from the scalp and the identification of two distinct thoughts according to the proposed methodology. The system's upward-trending performance indicates potential for future enhancements, paving the way for an affordable and accessible solution that empowers individuals with disabilities to interact with their surroundings and improve their overall well-being.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.