Reproduce “EEG-BBNet: a Hybrid Framework for Brain Biometric using Graph Connectivity” on OpenResearch

About this paper

Payongkit Lakhan, Nannapas Banluesombatkul, Natchaya Sricom, Korn Surapat, Ratha Rotruchiphong, Phattarapong Sawangjai, Tohru Yagi, Tulaya Limpiti and 1 more

Researchers developed EEG-BBNet, a hybrid deep learning framework that combines convolutional neural networks for raw signal feature extraction with graph convolutional neural networks for learning inter-electrode connectivity patterns from EEG data. This framework achieved a 99.26% correct recognition rate for brain biometric identification using Event-Related Potential (ERP) tasks and demonstrated high adaptability to session variability with minimal fine-tuning, while also performing effectively with reduced electrode sets.

Key insights

• A hybrid approach leveraging both raw EEG signal characteristics (via CNNs) and functional connectivity patterns (via GCNNs) provides a more discriminative and robust representation for individual brain biometrics.
• Functional connectivity measures, particularly the Rho index and Pearson's correlation, are highly informative for personal identification, significantly improving performance over models that neglect connectivity or use only physical distance.
• EEG-BBNet demonstrates strong adaptability to inter-session variability, with significant performance recovery (from 29% to nearly 80% CRR for ERP) achieved by fine-tuning with as little as 10% of data from a new session.

Summary by alphaXiv.