Towards ARSPI-NET: Advancing EEG Feature Extraction with Neuromorphic Algorithms

In this research, we present developments to ARSPI-NET, a pioneering hybrid framework aiming to redefine feature extraction methodologies for affective clinical EEG signal analysis through the integration of neuromorphic computing principles. Focusing on the capabilities of Liquid State Machines (LSMs) and Spiking Neural Networks (SNNs), our work addresses the complexities of emotional processing within EEG signals by proposing an energy-efficient and biologically plausible solution. By employing LSMs as a core component, ARSPI-NET interprets spatiotemporal dynamics in EEG data, enhancing the classification of psychophysiological states. Through comprehensive experiments, including the classification of emotional valence from EEG signals induced by the International Affective Picture System (IAPS), we validate ARSPI-NET's effectiveness against traditional methods. Our findings reveal that, besides achieving competitive accuracy, ARSPI-NET stands out for its potential in reducing computational costs and paving the way for the adoption of neuromorphic hardware solutions like Intel's Loihi and IBM's TrueNorth. This study emphasizes the scalability and interpretability of ARSPI-NET, setting a foundation for future explorations harnessing neuromorphic computing for real-world EEG analysis.