Discrimination of Two-Class Motor Imagery in a fNIRS Based Brain Computer Interface

Area of Science:

• Neuroscience
• Biomedical Engineering

Background:

• Motor imagery (MI) is a cognitive process involving the mental simulation of movement.
• Distinguishing between left- and right-hand MI is crucial for brain-computer interfaces (BCIs).
• Functional near-infrared spectroscopy (fNIRS) offers a non-invasive method for measuring brain activity.

Purpose of the Study:

• To discriminate between left- and right-hand motor imagery tasks.
• To evaluate various feature extraction and classification methods for fNIRS-based MI detection.

Main Methods:

• fNIRS data were collected from two participants performing left- and right-hand motor imagery.
• Feature extraction included mean, peak, minimum, skewness, and kurtosis.
• Classification algorithms tested were LDA, QDA, SVM, logistic regression, KNN, and neural networks (LMA, BRANN, SCGA).

Main Results:

• Classification accuracies were below 58% when using skewness and kurtosis features.
• Mean, peak, and minimum features yielded higher accuracies with QDA, SVM, and KNN compared to LDA and logistic regression.
• Bayesian Regularization Neural Networks (BRANN) achieved the highest classification accuracies, exceeding 98%, when using mean, peak, and minimum features.

Conclusions:

• Feature selection significantly impacts the accuracy of fNIRS-based motor imagery classification.
• Simple statistical features (mean, peak, minimum) combined with advanced neural network training (BRANN) are highly effective for discriminating left- and right-hand motor imagery.
• This approach shows promise for developing more accurate fNIRS-controlled BCIs.