Robust and hardware agnostic EEG signal processing tool
Dan Curăvale, Cristina Andronache, Ana Neacşu · 2024
Advancements in computational power, signal processing, and machine learning have significantly enhanced automatic EEG (Electroencephalogram) processing and interpretation. Historically, EEGs were primarily used clinically for diagnosing epilepsy and other central nervous system disorders. While automatic EEG analysis has expanded into numerous BCI (Brain-Computer Interface) applications, these systems are typically tailored to individual users and lack generalizability. Clinically, EEGs still require manual analysis due to high inter-subject variability, non-standardized recording techniques, and limited public EEG data availability. This article proposes a robust tool for epilepsy detection that is agnostic to acquisition hardware and techniques, addressing issues like electrode referencing, number, and positioning. To achieve robust inter-subject classification performance, our neural network is trained on one of the largest EEG databases. The proposed system includes modules for standardization, visualization, pre-processing, and classification, all integrated into an intuitive graphical user interface (GUI).