Linear regression models and k-means clustering for statistical analysis of fNIRS data

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique.
• Accurate statistical estimation of hemodynamic responses is crucial for fNIRS data analysis.
• Existing methods may involve significant assumptions or approximations.

Purpose of the Study:

• To develop a novel algorithm for statistically estimating hemodynamic activations in fNIRS data.
• To minimize assumptions and approximations in the statistical analysis of fNIRS datasets.
• To introduce a clustering method for classifying activated channels in fNIRS data.

Main Methods:

• A linear regression model was employed for statistical estimation of hemodynamic activations.
• A K-means clustering algorithm was utilized to categorize fNIRS channels as activated or non-activated.
• The proposed methods were validated using both simulated and in vivo fNIRS data.

Main Results:

• The algorithm successfully estimated hemodynamic activations in fNIRS data.
• The K-means method effectively clustered activated and non-activated channels.
• Validation on simulated and in vivo data confirmed the algorithm's efficacy.

Conclusions:

• The proposed linear regression-based algorithm provides a robust method for analyzing fNIRS data.
• The K-means clustering enhances the interpretation of brain activation patterns.
• The methodology is applicable across different fNIRS techniques, including time domain (TD), continuous wave (CW), and frequency domain (FD).