EEG Classification with a Sequential Decision-Making Method in Motor Imagery BCI.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Accurate and rapid mental state classification is crucial for brain-computer interfaces (BCI).
• Real-time BCI applications require efficient EEG signal processing for control systems.

Purpose of the Study:

• To develop a subject-specific classifier for fast and reliable mental state recognition in BCI.
• To improve upon existing sequential decision-making strategies for EEG classification.

Main Methods:

• Utilized optimal wavelet analysis with subject-specific parameters for EEG classification.
• Developed a balanced threshold SPRT (BTSPRT) method for sequential decision-making.
• Optimized sequential probability ratio test (SPRT) thresholds based on cumulative evidence and desired stopping time.

Main Results:

• The BTSPRT method achieved an average maximum accuracy of 83.4% with an average decision time of 2.77 seconds.
• This represents an improvement over the sequential Bayesian (SB) method, which yielded 79.2% accuracy and 3.01 seconds decision time.
• BTSPRT demonstrated an explicit relationship between stopping time, thresholds, and error rates.

Conclusions:

• The BTSPRT method offers enhanced classification accuracy and decision speed compared to non-sequential and SB methods.
• This technique provides a valuable tool for explicitly adjusting the speed-accuracy tradeoff in BCI.
• The findings suggest BTSPRT's utility in real-time BCI applications, such as neuroprosthetic control.