Enhancing classification of a large lower-limb motor imagery EEG dataset for BCI in knee pain patients

Abstract

Abstract Chronic knee osteoarthritis pain significantly impacts patients’ quality of life and motor function. While motor imagery (MI)-based brain-computer interface (BCI) systems have shown promise in rehabilitation, their application to lower-limb conditions, particularly in pain patients, is underexplored. This study evaluates the feasibility of applying an MI-BCI model to a large dataset of knee pain patients, utilizing a novel deep learning algorithm for signal decoding. This EEG data was collected and analysed from 30 knee pain patients, revealing significant event-related (de)synchronization (ERD/ERS) during MI tasks. Traditional decoding algorithms achieved accuracies of 51.43%, 55.71%, and 76.21%, while the proposed OTFWRGD algorithm reached an average accuracy of 86.41%. This dataset highlights the potential of lower-limb MI in enhancing neural plasticity and offers valuable insights for future MI-BCI applications in lower-limb rehabilitation, especially for patients with knee pain.