Characterizing neural activity from EEG signals during game interaction using persistent homology

Understanding the complex neural activity patterns in the developing brain remains a pivotal challenge in neuro-science. Characterizing neural activity during game interaction offers a unique opportunity to explore brain dynamics in a naturalistic and engaging context. Electroencephalography (EEG) provides a non-invasive method to capture neural signals with high temporal resolution, making it well-suited for studying real-time brain activity during dynamic tasks like gaming. Such signals are non-stationary and exhibit a non-linear manifold nature, which makes it difficult in classifying into different states during interaction. Topological data learning (TDL) framework can help in this context. However, there has been limited study conducted on the application of TDL framework for the purpose of characterizing and classifying EEG data. Therefore, this work utilizes topological attributes like Betti numbers and persistent homology of dimension 0 and 1 extracted from EEG signals to study the neural activity of an individual. Random forest classifier is used to classify the EEG signals during game interaction with accuracy $84.49 \%$. Results show that the performance is at par with deep learning approaches even when less number of topological features are used. The proposed framework shows the efficacy of persistent homology features in EEG signal classification. The robustness of the proposed framework is assessed with publicly available EEG-based Mobile brain-body imaging (MoBI) data of children.