Related Experiment Video
Updated: Jun 13, 2026
11:25
Simultaneous Scalp Electroencephalography (EEG), Electromyography (EMG), and Whole-body Segmental Inertial Recording for Multi-modal Neural Decoding
Published on: July 26, 2013
MGFNet: A Multi-Granularity Fusion Network with Coupling-Guided Sparse Routing for Hybrid EEG-fNIRS Decoding.
Yan Zhang1, Xiaoyu Gong1, Xiaoyang Yuan1
1School of Electrical Engineering and Automation, Harbin Institute of Technology, Harbin 150001, China.
Sensors (Basel, Switzerland)
|June 12, 2026
Summary
MGFNet enhances hybrid brain-computer interfaces (BCIs) by effectively fusing electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) data. This novel approach achieves high accuracy in decoding cognitive tasks and demonstrates improved robustness against signal degradation.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Hybrid brain-computer interfaces (BCIs) integrate electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) for improved performance.
- Existing deep fusion methods often use static late-fusion, limiting their ability to capture cross-modal dependencies and handle signal degradation.
- There is a need for advanced fusion strategies that can effectively combine EEG and fNIRS data while being robust to noise.
Purpose of the Study:
- To propose MGFNet, a multi-granularity fusion network for enhanced hybrid BCI decoding.
- To address the limitations of static late-fusion methods in exploiting cross-modal dependencies and mitigating modality-specific signal degradation.
- To evaluate the performance and robustness of MGFNet on benchmark cognitive tasks.
Main Methods:
- Developed MGFNet with intra-modal encoders for modality-specific representations (EEG, HbO, HbR).
- Incorporated cross-modal interaction encoders using dilated convolutions for long-range EEG-fNIRS dependency capture.
- Introduced a Coupling-Guided Sparse Component Routing (CGSCR) module for adaptive routing and a deep supervision strategy for optimization.
Main Results:
- MGFNet achieved high classification accuracies: 99.40% on the n-back task and 99.03% on the word generation (WG) task.
- Outperformed representative comparison methods under a matched within-subject evaluation protocol.
- Demonstrated significant robustness against controlled EEG corruption, outperforming a static-fusion variant by over 9% on the n-back task.
Conclusions:
- MGFNet effectively decodes cognitive tasks using hybrid EEG-fNIRS data.
- The proposed multi-granularity fusion approach significantly improves BCI performance and robustness.
- MGFNet represents a promising advancement for hybrid BCI applications, particularly in challenging signal conditions.
