Classification of Individual Finger Movements from Right Hand Using fNIRS Signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is an emerging noninvasive brain imaging technique.
• Dexterous tasks like individual finger-tapping remain under-explored with fNIRS.
• Investigating fNIRS for complex motor control is crucial for brain-computer interface (BCI) development.

Purpose of the Study:

• To assess the feasibility of using fNIRS to detect and classify individual finger-tapping movements.
• To evaluate the performance of various machine learning algorithms in decoding fNIRS data from the motor cortex during finger-tapping.

Main Methods:

• Collected fNIRS data from 24 healthy volunteers performing individual finger-tapping tasks.
• Applied a standard fNIRS processing pipeline, including noise reduction with Butterworth and Savitzky-Golay filters.
• Extracted eight spatial statistical features from oxygenated hemoglobin data and applied machine learning classifiers (SVM, RF, DT, AdaBoost, QDA, ANN, kNN, XGBoost).

Main Results:

• Machine learning classifiers, particularly kNN, Random Forest, and XGBoost, demonstrated high classification accuracies (0.75-0.77).
• These algorithms effectively distinguished between different finger-tapping actions based on fNIRS signals.
• The results indicate fNIRS's potential for decoding complex motor intentions.

Conclusions:

• fNIRS, combined with advanced machine learning, shows significant promise for decoding intricate motor tasks.
• This preliminary study opens new avenues for generating control commands in fNIRS-based BCIs.
• Further research involving deeper temporal and spatial signal analysis is warranted to fully understand the underlying neural mechanisms.