BCIwiki (bciwiki.com) — According to research information from the University of Salerno, a preprint uploaded to bioRxiv on 2026-09-15 introduces FUSE, a framework combining electroencephalography (EEG) and magnetoencephalography (MEG) for motor imagery brain-computer interface (BCI) decoding. The preprint has not been peer reviewed. The study evaluated simultaneous EEG and MEG recordings from 20 participants to examine whether shared representations could improve classification and transfer across participants.
FUSE separates representation learning from downstream classification into two phases. First, a convolutional encoder-decoder predicts the power spectral density of EEG and MEG signals directly from time-domain activity, rather than reconstructing the original input. Power spectral density describes how signal power is distributed across frequencies. Second, the encoder is frozen, and its learned representations are reused without further adaptation for downstream motor imagery classification.
The authors report that learned representations increased median classification accuracy from 0.734 with conventional handcrafted spectral features to 0.794. The multimodal framework's median accuracy of 0.794 also exceeded MEG alone at 0.680 and modestly exceeded EEG alone at 0.765. In cross-participant analyses, more than half of cross-subject models performed within 0.01 accuracy of their subject-specific counterparts. This proportion refers to models, not participants.
The authors interpret the findings as evidence that task-agnostic representation learning can capture physiologically meaningful multimodal neural representations that transfer across individuals, potentially supporting more robust and reusable BCI pipelines. The observed results concern classification and transfer performance on recorded data, not clinical efficacy.