Autoregressive model based algorithm for correcting motion and serially correlated errors in fNIRS.

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Signal Processing

Background:

• Functional near infrared spectroscopy (fNIRS) is susceptible to systemic physiology and motion artifacts.
• These artifacts introduce noise, reducing analysis performance and inflating false positive rates in detecting hemodynamic responses.

Purpose of the Study:

• To develop and evaluate a general algorithm for solving the general linear model (GLM) in fNIRS.
• The algorithm aims to optimize pre-whitening filters and use iteratively reweighted least squares to mitigate noise.

Main Methods:

• Developed a general algorithm for GLM analysis in fNIRS, incorporating optimal pre-whitening filters and iteratively reweighted least squares.
• Evaluated performance using receiver operating characteristic (ROC) analyses on synthetic data with controlled noise and simulated artifacts.
• Tested the method on experimental data from children (3-5 years old) to assess real-world noise reduction.

Main Results:

• The proposed method significantly outperformed ordinary least squares (OLS) with existing motion correction techniques.
• False positive rates in experimental data decreased from 37% (OLS) to 5-9% with the new method at a p-value of 0.05.
• The algorithm demonstrated superior control of type I errors in the presence of physiological and motion-related noise.

Conclusions:

• The developed algorithm effectively reduces confounding noise in fNIRS data.
• It improves the accuracy of detecting hemodynamic responses by controlling false positive rates.
• This method enhances the reliability and performance of fNIRS analyses, particularly in noisy conditions.