Improved classification performance of EEG-fNIRS multimodal brain-computer interface based on multi-domain features...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) measure distinct neural activity characteristics.
• Hybrid EEG-fNIRS brain-computer interfaces (BCIs) are a growing research area.
• Existing fusion methods for EEG and fNIRS data lack systematic approaches, limiting BCI performance.

Purpose of the Study:

• To develop a novel multimodal fusion framework for EEG and fNIRS data.
• To exploit the complementary nature of EEG and fNIRS for enhanced BCI performance.
• To establish a systematic approach for fusing multi-domain brain signal features.

Main Methods:

• Proposed a multimodal fusion framework utilizing multi-level progressive learning.
• Implemented multi-domain feature extraction for both EEG and fNIRS signals.
• Employed atomic search optimization for feature selection and progressive machine learning for feature fusion.

Main Results:

• Multi-domain features outperformed single-domain features in classification accuracy.
• Multimodal fusion (EEG-fNIRS) demonstrated superior performance compared to single modalities.
• Achieved high average classification accuracies: 96.74% for motor imagery and 98.42% for mental arithmetic tasks.

Conclusions:

• The proposed fusion framework effectively integrates EEG and fNIRS data.
• The method offers superior classification performance for brain-computer interfaces.
• This framework provides a generalizable approach for future multimodal brain signal fusion.