BCIwiki (bciwiki.com) — A paper published on 2026-10-17 in Frontiers in Computational Neuroscience reports a unified framework for classifying motor imagery electroencephalography (EEG), evaluated on two brain-computer interface (BCI) competition datasets. Motor imagery means imagining an action; these BCI systems classify the corresponding EEG patterns to distinguish motor imagery tasks.
The framework combines a sliding window (SW), a Fourier decomposition method (FDM), and a one-dimensional convolutional neural network (1D-CNN). Raw EEG is segmented to augment the data and extract additional temporal feature patterns from each trial. Band-pass filtering and FDM then extract frequency-based features. FDM decomposes the signal into eight orthogonal Fourier Intrinsic Band Functions (FIBFs), from which the 1D-CNN learns representations.
Under the subject-dependent evaluation protocol, accuracy reached 93.67% on BCI Competition III Dataset IVa (BCIC III IVa) and 79.14% on BCI Competition IV-2a (BCIC IV 2a). The researchers also used Leave-One-Subject-Out (LOSO) evaluation, holding out one person for testing to assess cross-subject generalization. Average classification accuracy under this protocol was 76.49% and 55.50% on the two datasets, respectively.
On the NVIDIA Jetson Nano, complete pipeline latency was 58.119 ms/trial with throughput of 17.21 trials/s for BCIC III IVa, and 85.055 ms/trial with throughput of 11.76 trials/s for BCIC IV 2a. The authors interpret the findings as indicating potential to assist people with motor impairments. The reported evidence concerns dataset classification and device deployment, rather than therapeutic effects in humans.