This thesis explores the integration of advanced signal processing techniques and Explainable Artificial Intelligence (XAI) methodologies to enhance the understanding and modeling of functional connectivity of EEG signals in the context of language disorders, particularly Developmental Dyslexia (DD). Developmental dyslexia is a neurobiological condition affecting reading skills linked to neural synchronization deficits, making it a prime candidate for innovative diagnostic approaches.
The approach uses non-invasive brain recording techniques, mainly Electroencephalogram (EEG), to assess connectivity through Cross Frequency Coupling (CFC) metrics. A methodological framework is established, combining classical signal processing with the design of Deep Learning (DL) architectures, including Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). These models are designed to capture complex patterns in EEG data while maintaining interpretability through XAI techniques like SHapley Additive exPlanations (SHAP) or attention layers in the network architecture. These methods are also used to evaluate the lateralization effect in language processing. Key contributions include the development of novel deep learning architectures tailored for Phase-To-Amplitude Coupling (PAC) estimation, classification methods for differential diagnosis, and explainable tools to aid clinicians in understanding EEG based connectivity metrics. The proposed models demonstrate improved accuracy in identifying dyslexic neural patterns, highlighting the role of neural synchronization deficits in language processing anomalies. Moreover, different entrainment profiles to auditory stimuli were found when studying local and long-range neural coupling by
means of CFC. The findings advance the field of neurodiagnostics by offering robust, interpretable AI-driven methods to study functional connectivity.
Buscar en Dimension