Accessible and explainable AI for EEG decoding in brain-computer interfaces

The field of brain–computer interfaces (BCIs) is increasingly emerging as a highly relevant domain of research and experimentation, with growing implications for human well-being and daily activities. In this context, electroencephalography (EEG), a non- invasive measure of brain activity, is a crucial tool that enables both the investigation of neural mechanisms that remain largely unexplored and the development of interventions across multiple application areas, including motor rehabilitation, emotion regulation, and mental health. In recent years, technological development has accelerated markedly, driven in particular by the rapid growth of artificial intelligence (AI) models and methodologies. The adoption of machine learning (ML) and deep learning (DL) techniques now enables the interpretation and decoding of highly complex patterns in EEG signals that are largely inaccessible through traditional analytical methods. At the same time, AI is increasingly recognized as a transformative force in healthcare, offering opportunities to improve diagnosis and clinical management and to develop innovative solutions in the BCI domain. This doctoral thesis lies at the intersection of neuroscience and AI and is structured around two complementary research directions. First, it focuses on the development and integration of advanced EEG preprocessing techniques designed to operate in both offline analyses and real-time applications. Second, it investigates state-of-the-art ML and DL models for decoding and analyzing EEG patterns. The contributions presented in this thesis include tools and frameworks that bridge the gap between AI research and its translation into real-world applications. By incorporating principles of explainable artificial intelligence (XAI), the proposed models are designed to be transparent and interpretable. Each contribution is presented in detail in its respective chapter, providing a comprehensive overview of the work conducted and its implications for artificial intelligence and computational neuroscience.