Neurophysiological data augmentation for EEG-fNIRS multimodal features based on a denoising diffusion probabilistic...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Hybrid brain-computer interfaces (BCI) combining electroencephalogram (EEG) and functional near-infrared spectroscopy (fNIRS) offer advantages over single-modality systems.
• Deep learning significantly improves BCI performance, but is hindered by limited brain signal data.

Purpose of the Study:

• To propose an EEG-fNIRS data augmentation framework (EFDA-CDG) to enhance hybrid BCI system performance.
• To address the data scarcity issue in deep learning for BCI applications.

Main Methods:

• Developed an EEG-fNIRS data augmentation framework (EFDA-CDG) integrating denoising diffusion probabilistic models (DDPM) and Gaussian noise addition.
• Unified EEG and fNIRS data dimensions through feature extraction and spatial mapping interpolation.
• Incorporated EEG feature attention and fNIRS terrain attention in the classification module.

Main Results:

• Validated the EFDA-CDG framework on three public and one self-collected database.
• Achieved high accuracy rates: 82.02% for motor imagery, 91.93% for mental arithmetic, and 90.54% for n-back tasks on public datasets.
• Demonstrated 97.82% accuracy for drug addiction discrimination on a self-collected dataset.

Conclusions:

• The EFDA-CDG framework effectively augments data for hybrid EEG-fNIRS BCI systems.
• This augmentation significantly enhances the performance and accuracy of BCI applications.