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

We present the first large-scale, standardized EEG dataset (30 patients, 150 sessions, 15,000 trials) specifically designed for lower-limb motor imagery (MI) in knee pain patients, addressing a critical gap in clinical BCI research. Chronic knee pain alters cortical plasticity, yet our data demonstrate preserved MI capability in patients—a finding with direct implications for rehabilitation BCI development. Our proposed Optimal Time-Frequency Window Riemannian Geometric Distance (OTFWRGD) algorithm achieves 86.41% classification accuracy, significantly outperforming traditional methods (CSP+LDA: 51.43%; FBCSP+SVM: 55.71%; EEGNet: 76.21%). The dataset adheres to EEG-BIDS standards and is fully accessible via Figshare, including raw/preprocessed EEG, stimuli, and analysis code.