Adaptive filtering of physiological noises in fNIRS data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique.
• fNIRS data is susceptible to physiological noise and baseline drift, complicating the analysis of hemodynamic responses (HRs).
• Accurate extraction of HRs is crucial for understanding brain function during various tasks.

Purpose of the Study:

• To develop and evaluate a novel recursive least-squares estimation method with exponential forgetting for noise reduction in fNIRS data.
• To enhance the extraction of hemodynamic responses (HRs) from noisy fNIRS measurements.
• To assess the performance of the proposed method against existing techniques like Kalman filtering and independent component analysis.

Main Methods:

• A recursive least-squares estimation method with an exponential forgetting factor was developed.
• Hemodynamic responses (HRs) were modeled using a linear regression incorporating expected HR, its derivatives, short-separation data, physiological noises, and baseline drift.
• The method was applied to fNIRS data from motor cortex experiments involving finger movements in healthy participants.

Main Results:

• The proposed method demonstrated significant noise reduction in fNIRS data.
• Substantial improvements in contrast-to-noise ratio (CNR) were observed for oxy-hemoglobin (77% reduction in channels with higher CNR) and deoxy-hemoglobin (99% reduction).
• The algorithm proved robust in yielding consistent HR data and was effective for both offline and online processing.

Conclusions:

• The recursive least-squares method with exponential forgetting offers a robust and effective solution for noise removal in fNIRS.
• This approach significantly enhances the accuracy and reliability of hemodynamic response extraction.
• The method shows promise for real-time and post-hoc analysis of fNIRS data in neuroscience research.