Multivariate disturbance filtering in auditory fNIRS signals using maximum likelihood gradient estimation method:...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is valuable for brain cortex assessment but is prone to noise from physiological and instrumental sources.
• Existing filtering methods struggle to effectively remove diverse disturbances like heartbeats, breathing, and shivering.

Purpose of the Study:

• To introduce and evaluate a novel filtering algorithm, maximum likelihood generalized extended stochastic gradient (ML-GESG) estimation, for enhancing fNIRS signal quality.
• To reduce multiple disturbances in fNIRS data by treating them as multivariate parameters.

Main Methods:

• An ML-GESG estimation algorithm was developed to filter complex noise sources in fNIRS signals.
• A comparative analysis was performed against a conventional filtering method.
• Auditory stimuli (12 voice sources) were used with 10 volunteers, employing an 18-channel fNIRS setup.
• Psycho-acoustic factors (loudness, sharpness) served as quality indices for physiological responses.

Main Results:

• The ML-GESG method effectively reduced multiple disturbances, including physiological and instrumental noise.
• Hemodynamic responses, specifically oxygenated hemoglobin concentration, showed a better correlation with psycho-acoustic analysis using the proposed ML-GESG filter.
• The new filtering approach demonstrated superior performance in filtering involuntary signals compared to conventional methods.

Conclusions:

• The ML-GESG estimation is a promising alternative for improving the accuracy of fNIRS-based brain activity analysis.
• This method enhances the reliability of fNIRS by providing a cleaner signal, better correlating with psycho-acoustic stimuli.
• The study highlights the potential of ML-GESG for more precise assessment of brain cortex status in noisy environments.