Decoding working-memory load duringn-back task performance from high channel fNIRS data.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Functional near-infrared spectroscopy (fNIRS) measures brain activity via blood oxygenation changes.
• Wearable fNIRS devices enable research and occupational applications.
• Machine learning (ML) based brain-computer interfaces (BCIs) can decode cognitive states, but high-channel fNIRS data presents classification challenges due to limited training trials.

Purpose of the Study:

• To develop and evaluate an ML strategy for classifying working-memory load using high-resolution fNIRS data.
• To address challenges in ML classification with limited training data in high-channel fNIRS.
• To determine if state-of-the-art performance can be achieved with novel ML approaches for fNIRS BCIs.

Main Methods:

• Proposed an ML strategy combining spatio-temporal regularization and transfer learning for fNIRS BCI.
• Interpreted the approach as an end-to-end generalized linear model for interpretability.
• Utilized a 3198 dual-channel NIRS device on 43 participants performing the n-Back task.

Main Results:

• Achieved state-of-the-art decoding performance with high-resolution fNIRS data.
• Existing methodologies struggled with high-channel data and were outperformed by the proposed pipeline.
• Demonstrated the effectiveness of the ML strategy in classifying working-memory load.

Conclusions:

• The proposed ML approach establishes high-channel fNIRS devices as a viable platform for advanced BCIs.
• This work enables new applications for wearable fNIRS headsets.
• Facilitates high-resolution model imaging and interpretation for fNIRS data.