BCIwiki (bciwiki.com) — Researchers at Shaanxi Normal University's Key Laboratory of Modern Teaching Technology and Microsoft in Beijing introduced, in a preprint posted to arXiv on September 17, 2026, Bio-MF, a one-step generative framework that synthesizes functional near-infrared spectroscopy (fNIRS) signals from electroencephalography (EEG) for hybrid motor-imagery brain-computer interface (BCI) decoding. The work has not been peer reviewed. Hybrid MI-BCIs combine EEG and fNIRS to exploit complementary electrophysiological and hemodynamic information, but wearing two sensor systems simultaneously is often impractical. Prior diffusion-based approaches to EEG-to-fNIRS generation require up to 1,000 iterative steps, creating latencies incompatible with real-time use.
Bio-MF performs direct signal-space prediction in a single network evaluation, integrating spatial-temporal interactive 4D encoding, cross-modal classifier-free guidance, and noise-level-gated FFT regularization to preserve task-relevant hemodynamic structure across heterogeneous sensor layouts. On Dataset 1, adding synthetic fNIRS to real EEG raised classification accuracy over EEG alone by 3.37 percentage points for deoxygenated hemoglobin (HbR) and 4.15 percentage points for oxygenated hemoglobin (HbO). On Dataset 2, gains of 2.98 and 2.50 percentage points persisted under an unseen 64-channel EEG montage. On an RTX PRO 6000 GPU, Bio-MF generates one fNIRS trial in 7.0 ms, an 857x speedup over the 1,000-step score-based conditional diffusion model baseline. The team has released its code on GitHub. All results come from offline datasets; no closed-loop online experiments have been reported.