Learning based motion artifacts processing in fNIRS: a mini review.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is sensitive to motion artifacts (MA) caused by subject movement, compromising data integrity.
• Traditional MA processing methods often reduce the reliability of hemodynamic responses and statistical power in fNIRS studies.
• Limited research exists on advanced learning-based approaches for MA removal in fNIRS.

Purpose of the Study:

• To review and analyze existing learning-based motion artifact processing methods for fNIRS.
• To identify research gaps in the current landscape of learning-based MA correction techniques.
• To propose a novel framework for evaluating the quality of MA correction in fNIRS data.

Main Methods:

• Systematic literature search of 315 studies, identifying seven relevant to learning-based MA removal in fNIRS.
• Analysis of current learning-based MA correction strategies and their limitations.
• Development of a proposed evaluation framework incorporating signal and model quality metrics (e.g., ΔSNR, confusion matrix, MSE).

Main Results:

• Identified a limited but growing body of research on learning-based MA removal in fNIRS.
• Highlighted the absence of standardized metrics for assessing the effectiveness of MA correction techniques.
• Proposed a comprehensive framework for evaluating MA correction quality in fNIRS.

Conclusions:

• Learning-based methods show promise for improving MA correction in fNIRS, but require further development and standardization.
• The proposed evaluation framework can enhance the assessment of MA correction techniques, leading to more reliable fNIRS data.
• This work aims to guide future research and facilitate the application of advanced learning-based methodologies in neurovascular studies.