A distributed adaptive network framework for ERP-Based classification of multichannel EEG signals

• Published in: Australasian physical & engineering sciences in medicine Vol. 48; no. 3; pp. 1207 - 1224
• Main Authors: Afkhaminia, Fatemeh, Shamsollahi, Mohammad Bagher, Bahraini, Tahereh
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.09.2025; Springer Nature B.V
• Subjects: Adaptability; Adaptation; Algorithms; Biological and Medical Physics; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedicine; Biophysics; Brain-Computer Interfaces; Classification; Collaboration; Cooperation; Datasets; Deep learning; Efficiency; Electroencephalography; Evoked Potentials; Human-computer interface; Humans; Machine Learning; Medical and Radiation Physics; Network topologies; Neural networks; Nodes; Real time; Scientific Paper; Signal classification; Signal processing; Signal Processing, Computer-Assisted; Multichannel EEG signals; Event-Related potential (ERP ); Diffusion strategy; MultiTask Network; Adapt then combine (ATC )algorithm; Distributive adaptive networks
• ISSN: 2662-4729; 0158-9938; 2662-4737; 2662-4737; 1879-5447
• DOI: 10.1007/s13246-025-01578-2

Understanding brain function is one of the most challenging areas in brain signal processing. This study introduces a novel framework for electroencephalography (EEG) signal classification based on distributed adaptive networks using diffusion strategy. Our approach models the brain as a multitask network, where EEG electrodes are considered as nodes of this network. The network parameters are dynamically optimized based on the data from the nodes and inter-node cooperation. The proposed framework, which comprises network modeling and diffusion-based adaptation using the adapt then combine (ATC) algorithm, has been validated on different types of data. Experimental results indicate that the proposed framework outperforms common methods in classifying EEG data based on event-related potential (ERP) pattern identification, particularly in scenarios where machine learning-based models struggle with limited data. Furthermore, its ability to adapt to the non-stationary and dynamic nature of EEG signals and its efficient real-time implementation make this approach ideal for brain-computer interface (BCI), cognitive neuroscience, and clinical applications.