Single-trial classification of antagonistic oxyhemoglobin responses during mental arithmetic

Area of Science:

• Neuroscience
• Biomedical Engineering
• Optical Imaging

Background:

• Brain-computer interfaces (BCIs) face challenges in reliably classifying single-trial cognitive task data.
• Near-infrared spectroscopy (NIRS) offers a non-invasive optical method for BCI applications.
• Antagonistic hemodynamic activation patterns in the prefrontal cortex were previously identified during mental arithmetic (MA).

Purpose of the Study:

• To investigate the efficacy of antagonistic hemodynamic features for improving online classification accuracy in optical BCIs.
• To compare the performance of combined antagonistic features against individual hemodynamic features.
• To explore the utility of different hemoglobin measures ([oxy-Hb], [deoxy-Hb], [Hbtot]) and channel combinations.

Main Methods:

• Utilized prefrontal NIRS data from eight subjects performing MA tasks.
• Analyzed [oxy-Hb] responses to identify optimal antagonistic feature combinations.
• Compared classification accuracy using antagonistic [oxy-Hb] features versus single-region features.
• Evaluated antagonistic [deoxy-Hb], [Hbtot], and combined [oxy-Hb]/[deoxy-Hb] features.

Main Results:

• The classification accuracy significantly improved from 63.3% to 79.7% when using antagonistic [oxy-Hb] features.
• Antagonistic hemodynamic patterns proved more effective than individual features for classification.
• Effective BCI control was achieved using only two prefrontal NIRS channels.

Conclusions:

• Antagonistic hemodynamic response patterns represent a viable control strategy for optical BCIs.
• The use of specific prefrontal NIRS channels capturing antagonistic patterns enhances classification performance.
• This approach offers a pathway to more stable and reliable cognitive task-based BCIs.