Single trial classification of fNIRS-based brain-computer interface mental arithmetic data: a comparison between...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near infrared spectroscopy (fNIRS) is an emerging non-invasive technique for assessing cerebral cortex activity.
• fNIRS is increasingly utilized in brain-computer interface (BCI) research.
• A key challenge in fNIRS-based BCIs is achieving stable and reliable single-trial classification of hemodynamic patterns.

Purpose of the Study:

• To compare the performance of five different classification methods for fNIRS data in BCI applications.
• To evaluate linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), support vector machines (SVM), and their regularized variants (sLDA, sQDA).
• To identify the most effective and computationally efficient classifier for fNIRS-based BCIs.

Main Methods:

• Five classification algorithms were applied to fNIRS data recorded during mental arithmetic tasks.
• Classifiers included LDA, QDA, SVM, analytic shrinkage regularized LDA (sLDA), and analytic shrinkage regularized QDA (sQDA).
• Performance was evaluated based on classification accuracy for oxy-hemoglobin (oxy-Hb) and deoxy-hemoglobin (deoxy-Hb) signals.

Main Results:

• Classification accuracies ranged from 56.1% (deoxy-Hb/QDA) to 86.6% (oxy-Hb/SVM).
• Regularized classifiers (sLDA, sQDA) demonstrated significantly better performance compared to their non-regularized counterparts (LDA, QDA).
• The SVM classifier achieved the highest accuracy, but sLDA offered a strong balance of performance and simplicity.

Conclusions:

• Regularized classifiers are superior for single-trial classification of fNIRS data in BCI.
• sLDA is recommended for fNIRS-based BCIs due to its significant performance improvement, simplicity, and computational efficiency.
• This study provides a comprehensive comparison of classifiers for advancing fNIRS-BCI technology.