Classification of Overt and Covert Speech for Near-Infrared Spectroscopy-Based Brain Computer Interface

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Neuromuscular disorders like locked-in syndrome (LIS) cause paralysis but preserve cognition.
• Brain-Computer Interfaces (BCIs) offer communication pathways for individuals with severe motor impairments.
• Near-Infrared Spectroscopy (NIRS) measures hemodynamic activity in the brain.

Purpose of the Study:

• To investigate the feasibility of using NIRS to detect hemodynamic changes associated with overt and covert speech.
• To develop and evaluate an intuitive BCI system for individuals with LIS.
• To assess the classification accuracy of NIRS signals for spoken words.

Main Methods:

• Eight healthy participants performed overt and covert speech tasks.
• A 12-channel NIRS system measured hemoglobin concentration changes in the inferior frontal gyrus.
• An unsupervised feature extraction algorithm and optimized support vector machine were used for classification.

Main Results:

• High classification accuracies were achieved for both overt (88.19% O2Hb, 78.82% HHb) and covert speech (79.17% O2Hb, 86.81% HHb).
• Overall classification accuracy reached 92.88% for oxy-hemoglobin (O2Hb) and 95.14% for deoxy-hemoglobin (HHb).
• Covert speech classification demonstrated reliable performance.

Conclusions:

• NIRS-based BCIs can effectively detect hemodynamic changes related to speech.
• Covert speech provides a viable control paradigm for future NIRS-based BCIs.
• This technology has the potential to significantly improve communication for individuals with neuromuscular disorders.