A pipeline for decoding visual stimuli from task-evoked and spontaneous activity in EEG

We present a standardized and accessible pipeline designed for EEG data analysis, integrating deep learning techniques within the EBRAINS-Italy framework. Built on PyTorch for seamless GPU support, the pipeline automates EEG data preparation, allowing for efficient machine learning compatibility. It supports multiple analysis pathways, including encoding models for learning latent representations, decoding models for classification, and representational similarity analysis (RSA) to compare neural activity patterns. Case studies on statistical learning in task-evoked and spontaneous activity demonstrate its ability to analyze large-scale high-density EEG datasets. The pipeline's modular structure enables flexible workflows, making it applicable to a wide range of EEG research contexts.