Subject-Independent Functional Near-Infrared Spectroscopy-Based Brain-Computer Interfaces Based on Convolutional
Jinuk Kwon1,2, Chang-Hwan Im1,2
1Department of Biomedical Engineering, Hanyang University, Seoul, South Korea.
Frontiers in Human Neuroscience
|March 29, 2021
Summary
A new deep convolutional neural network (CNN) approach enables subject-independent functional near-infrared spectroscopy (fNIRS) brain-computer interfaces (BCIs). This method achieves high accuracy without lengthy individual calibration, improving BCI practicality.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Machine Learning
Background:
- Functional near-infrared spectroscopy (fNIRS) is a promising, non-invasive brain-computer interface (BCI) technology.
- fNIRS signals are subject-specific, necessitating individual calibration for reliable BCI performance.
- Current calibration requirements limit the practical application of fNIRS-based BCIs.
Purpose of the Study:
- To develop a subject-independent fNIRS-based BCI using a deep convolutional neural network (CNN).
- To evaluate the classification accuracy of the proposed CNN approach.
- To compare the performance against traditional methods and reduce calibration time.
Main Methods:
- A novel deep convolutional neural network (CNN) architecture was designed for subject-independent fNIRS data.
- Experiments involved 18 participants distinguishing mental arithmetic from an idle state.
- Leave-one-subject-out cross-validation was used to assess classification accuracy.
Main Results:
- The proposed subject-independent CNN-based fNIRS BCI achieved an average classification accuracy of 71.20 ± 8.74%.
- This accuracy surpassed the effective BCI communication threshold (70%) and linear discriminant analysis (65.74 ± 7.68%).
- Comparable accuracy to subject-dependent BCIs required significantly fewer training trials, reducing calibration time.
Conclusions:
- The CNN-based approach offers a viable solution for subject-independent fNIRS-based BCIs.
- This method significantly enhances the practicality of fNIRS BCIs by minimizing the need for extensive individual calibration.
- The findings pave the way for more accessible and efficient BCI applications.
