Dynamic topographical pattern classification of multichannel prefrontal NIRS signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Optical Imaging

Background:

• Near-infrared spectroscopy (NIRS) is an optical imaging technique for brain-computer interface (BCI) applications.
• Current NIRS-BCI studies primarily use temporal features of brain activity for mental state differentiation.
• Spatial distribution of haemodynamic activity also contains valuable information for brain state classification.

Purpose of the Study:

• To investigate the utility of spatiotemporal features in single-trial classification of haemodynamic events for NIRS-BCI.
• To compare classification performance using spatiotemporal, temporal, and combined features.
• To evaluate the impact of different task durations on feature extraction and classification accuracy.

Main Methods:

• Multichannel NIRS data were collected from the prefrontal cortex during a mental arithmetic task and rest.
• Four classification schemes were compared: spatiotemporal features, temporal features, combined features, and a majority vote classifier.
• Feature extraction was performed using two task durations: 20 seconds and 10 seconds.

Main Results:

• The majority voting classifier using features from a 20s task interval achieved the highest accuracy (76.1 ± 8.4%), outperforming temporal features alone (73.5 ± 8.5%).
• For the shorter 10s task duration, spatiotemporal features alone (67.9 ± 9.3%) yielded higher accuracy than temporal features alone (64.4 ± 8.4%).
• These findings indicate that spatiotemporal information significantly enhances classification accuracy in NIRS-BCI.

Conclusions:

• Spatiotemporal information derived from dynamic NIR topograms is valuable for analyzing functional NIRS data.
• Incorporating spatiotemporal features can lead to improved classification rates in NIRS-BCI applications.
• This approach offers a promising direction for advancing NIRS-BCI technology.