Bio-MF: Low-Latency and High-Fidelity EEG-to-fNIRS Cross-Modal Generation for Hybrid Motor-Imagery Brain--Computer Interfaces

Boyuan Zhao; Sifan Zhang; Luping Chen

cs.LGSep 17, 2026

Bio-MF: Low-Latency and High-Fidelity EEG-to-fNIRS Cross-Modal Generation for Hybrid Motor-Imagery Brain--Computer Interfaces

Authors: Boyuan Zhao, Sifan Zhang, Luping Chen

Organizations: Key Laboratory of Modern Teaching Technology, Shaanxi Normal University, Xi’an 710062, China · Microsoft, Beijing, China

Abstract

Hybrid motor-imagery brain-computer interfaces (MI-BCIs) combining EEG and fNIRS can outperform EEG-only systems by exploiting complementary electrophysiological and hemodynamic information. To obtain such hybrid information when paired EEG-fNIRS acquisition is unavailable or inconvenient, recent studies have focused on EEG-to-fNIRS cross-modal generation. However, existing methods still suffer from slow generation and often require pretraining, limiting their use in real-time MI-BCI scenarios. Although one-step generative models offer an attractive route to low-latency synthesis, removing the iterative refinement process can reduce generation fidelity and introduce non-physiological artifacts. To address these problems, this paper proposes Bio-MF, a latent-free one-step MeanFlow framework for EEG-conditioned fNIRS generation. Bio-MF performs direct signal-space x-prediction, converts this signal-space output into MeanFlow velocity supervision, and completes inference with one network evaluation. To preserve task-relevant hemodynamic structure under heterogeneous sensor layouts, Bio-MF integrates Spatial-Temporal Interactive 4D Encoding, cross-modal classifier-free guidance, and noise-level-gated FFT regularization. On Dataset 1, EEG + synthetic fNIRS improves ACC over EEG-only by 3.37 and 4.15 percentage points for HbR and HbO, respectively. On Dataset 2, the corresponding gains remain 2.98 and 2.50 percentage points under the unseen 64-channel EEG montage. On an RTX PRO 6000 GPU, Bio-MF generates one fNIRS trial in 7.0 ms, corresponding to an 857x speedup over the 1000-step SCDM latency. These results show that Bio-MF enables fast EEG-to-fNIRS synthesis while preserving task-relevant generation quality for downstream hybrid MI decoding. Our code is available at https://github.com/psychosiwa/Bio-MF.

Figures & tables

Fig. 1: Overall architecture of Bio-MF. (a) One-step fNIRS generation from the sampled noise endpoint and EEG condition. (b) Multimodal tokenization and shared Transformer blocks. (c) Transformer block structure. (d) Dual-branch MeanFlow velocity supervision and noise-level-gated FFT regularization. (e) CFG velocity target construction using the same shared-stem structure as in (b).

Fig. 2: Diagnostic visualizations motivating two-stage noise-level-gated FFT regularization. (a) Scalp topographies without FFT loss show locally nonsmooth synthetic fNIRS responses compared with real fNIRS. (b) HRF curves without FFT loss show high-frequency jitter and morphology deviation in synthetic fNIRS. (c) HRF curves under global FFT regularization, without the two-stage schedule and noise-level gate, show periodic or impulse-like artifacts.

Dataset	Signal	Model	ACC (%)	SPE (%)	PRE (%)	SEN (%)
Dataset 1	EEG	-	71.92 $\pm$ 2.79	76.85 $\pm$ 3.10	74.36 $\pm$ 3.12	66.98 $\pm$ 3.48
	Syn. fNIRS (HbR)	Bio-MF SCDM [25]	66.42 $\pm$ 2.70 66.38 $\pm$ 2.56	69.25 $\pm$ 2.86 68.27 $\pm$ 2.48	68.52 $\pm$ 3.00 68.32 $\pm$ 2.73	69.59 $\pm$ 2.78 68.26 $\pm$ 2.86
	EEG + Syn. fNIRS (HbR)	Bio-MF SCDM [25]	75.29 $\pm$ 2.36 75.38 $\pm$ 2.23	80.09 $\pm$ 2.61 79.98 $\pm$ 2.33	77.65 $\pm$ 2.72 77.52 $\pm$ 2.61	70.22 $\pm$ 2.79 70.50 $\pm$ 2.71
	Syn. fNIRS (HbO)	Bio-MF SCDM [25]	70.12 $\pm$ 2.82 68.52 $\pm$ 2.97	77.70 $\pm$ 2.70 76.33 $\pm$ 2.71	73.05 $\pm$ 3.12 71.79 $\pm$ 3.24	60.40 $\pm$ 2.76 60.25 $\pm$ 2.74
	EEG + Syn. fNIRS (HbO)	Bio-MF SCDM [25]	76.07 $\pm$ 2.40 75.03 $\pm$ 2.55	82.59 $\pm$ 2.61 81.92 $\pm$ 3.07	80.01 $\pm$ 2.76 79.32 $\pm$ 3.41	69.50 $\pm$ 2.85 67.32 $\pm$ 2.31
Dataset 2	EEG	-	70.78 $\pm$ 5.79	71.83 $\pm$ 5.69	71.56 $\pm$ 5.73	69.32 $\pm$ 6.77

TABLE I: Bio-MF and SCDM classification results.

Model	Sampling state	Generated sample	Steps / NFE	Latency	Speedup
SCDM [25]	Raw fNIRS signal	2 $\times$ 36 $\times$ 256	1000	6.0 s (measured, RTX PRO 6000)	1x
TADM [26]	512-D latent diffusion	$C_{fNIRS}$ $\times$ 1000 sequence; 36 $\times$ 1000 for 36-channel datasets	1000	0.5 s (reported, NVIDIA A800)	$12\times^{\dagger}$
Bio-MF	Raw fNIRS signal, latent-free	2 $\times$ 36 $\times$ 256	1	7.0 ms (measured, RTX PRO 6000)	857x

TABLE II: Generation latency comparison.

Fig. 3: EEG-fNIRS spatial correspondence visualization on Dataset 1. Blue markers denote EEG electrodes, orange markers denote fNIRS recording sites, and purple lines connect each fNIRS site to the EEG electrode with the highest average absolute Pearson correlation.

Fig. 4: HRF-curve visualization on Dataset 1. The panels compare average hemodynamic response curves and cross-trial variation ranges of real and synthetic fNIRS under LMI/RMI conditions.

Fig. 5: Scalp-topography visualization on Dataset 1. This visualization compares real and synthetic fNIRS topographies for HbR/HbO under LMI/RMI conditions, evaluating whether synthetic fNIRS preserves the main activation regions, spatial gradients, and temporal evolution trends of real fNIRS.

Dataset 1
Variant	ACC	SPE	PRE	SEN
Full Bio-MF	75.68 $\pm$ 2.38	81.34 $\pm$ 2.61	78.83 $\pm$ 2.74	69.86 $\pm$ 2.82
Original 4D	74.92 $\pm$ 2.55	80.62 $\pm$ 2.82	78.05 $\pm$ 2.95	69.08 $\pm$ 3.03
w/o 4D PE	74.66 $\pm$ 3.65	78.61 $\pm$ 3.91	76.33 $\pm$ 4.07	68.69 $\pm$ 4.12
w/o CFG	73.44 $\pm$ 2.48	78.30 $\pm$ 2.82	76.15 $\pm$ 2.99	68.46 $\pm$ 3.02
w/o FFT	74.05 $\pm$ 2.40	79.19 $\pm$ 2.74	76.92 $\pm$ 2.84	69.07 $\pm$ 2.94

TABLE III: Classification ablation results (%, EEG + Synthetic fNIRS; HbR/HbO average).

Brain-Computer Interfaces

Brain-computer interfaces translate neural activity into commands or feedback for communication, movement, rehabilitation, and sensory restoration.

See all 130 papers

EEG Decoding Using CNN and LSTM Network
Aug 13, 2026Athanasios KaragounisCross-Subject EEG Decoding MI Classification
School of Science, Department of Digital Industry Technologies GR34400, Psachna, Greece
NEXUS-MI: Communication-Aware Federated Personalization for Gateway-Coordinated Motor-Imagery Brain-Computer Interfaces
Sep 9, 2026Daniel Adu Worae, Aarthy NagarajanElectroencephalography Brain-Computer Interfaces
Department of Computer Science and Engineering, University of Notre Dame, 307 Fitzpatrick Hall, Notre Dame, IN 46556, USA · Department of Computer Science and Engineering and the Lucy Family Institute for Data and Society, University of Notre Dame, Notre Dame, IN 46556, USA
Frequency-Decorrelated Temporal Ensembles for EEG--fNIRS Imagined-Handwriting Decoding
Aug 4, 2026Xiao Fan, Hongbin Guo, Yubo Han +1Cross-Subject EEG Decoding Brain-Computer Interfaces
Xidian University

Aug 13, 2026J/K move · Enter open · S save

EEG Decoding Using CNN and LSTM Network

Athanasios Karagounis

School of Science, Department of Digital Industry Technologies GR34400, Psachna, Greece

Cross-Subject EEG Decoding MI Classification Brain-Computer Interfaces EEG Decoding

Open paper AskPDF arXiv

Motor imagery (MI) brain--computer interfaces (BCIs) have emerged as a promising approach for establishing flexible communication pathways between the human brain and external devices , particularly for individuals affected by stroke or neurodegenerative disorders. Reliable decoding of motor-imagery electroencephalography (MI-EEG) remains challenging because EEG recordings contain substantial noise and exhibit complex, weakly informative relationships with the underlying brain activity. Although deep learning provides an effective means of learning representations directly from EEG signals, its application to MI-EEG feature learning remains comparatively limited. This study introduces a hybrid deep-learning architecture that integrates a convolutional neural network (CNN) with a bidirectional long short-term memory (bi-LSTM) network. The CNN is used to learn high-level spatial and temporal representations directly from raw MI-EEG recordings, whereas the bi-LSTM models temporal dependencies and relationships among the extracted features. The proposed approach is evaluated using both a publicly available dataset and a privately acquired dataset obtained with an EEG acquisition system. The experimental results indicate that the CNN&bi-LSTM architecture provides robust performance for both two- and three-class motor-imagery classification and demonstrates promising subject-independent decoding capability across the evaluated methods.

Abstract

Brain-Computer Interfaces

Explore similar work

EEG Decoding Using CNN and LSTM Network

NEXUS-MI: Communication-Aware Federated Personalization for Gateway-Coordinated Motor-Imagery Brain-Computer Interfaces

Frequency-Decorrelated Temporal Ensembles for EEG--fNIRS Imagined-Handwriting Decoding