BCIwiki (bciwiki.com) — According to Journal of Neuroscience Methods, the journal published a study by Chrisilla S and Selvakumari RS on October 15, 2026, proposing a framework for subject-independent decoding of upper-limb motor imagery EEG. Addressing generalization across users in brain-computer interface (BCI) systems, the researchers evaluated multiple electrode configurations on PhysioNet, BCI IV-2a, and BCI IV-2b.
The framework combines graph convolution, skip connections, and handcrafted time-frequency features with parallel spatial and temporal convolutional pathways. Graph convolution captures functional connectivity between electrodes, while the other convolutional modules learn spatial and temporal representations. Subject-independent performance was assessed using Stratified Group K-Fold (SGKF) and Leave-One-Subject-Out (LOSO) validation. LOSO holds out one participant for testing at a time.
For three upper-limb classes in PhysioNet, configurations with 64, 21, and 5 electrodes achieved SGKF accuracies of 86.42%, 91.01%, and 93.91%, respectively. Corresponding LOSO accuracies were 92.25%, 95.72%, and 96.01%. Lower-electrode configurations performed better among those tested in this study, but this does not establish that reducing electrode count necessarily improves performance in other tasks.
On BCI IV-2a, the three reported configuration results under SGKF were 86.13%, 89.01%, and 92.33%; LOSO accuracy reached 97.12% with five electrodes. On BCI IV-2b, three electrodes yielded 88.60% accuracy under SGKF and 93.35% under LOSO.
The authors report competitive performance against representative graph-based methods while requiring fewer trainable parameters and EEG electrodes. They suggest that the approach could support resource-efficient BCI systems. These findings concern subject-independent dataset evaluations, not validated online interaction performance or clinical efficacy.