Are advanced methods necessary to improve infant fNIRS data analysis? An assessment of baseline-corrected averaging, general linear model (GLM) and multivariate pattern analysis (MVPA) based approaches

TL;DR

This study compares baseline-corrected averaging, GLM, and MVPA methods for analyzing infant fNIRS data, assessing their convergence and differences in detecting neural responses to face stimuli in infants.

Contribution

It evaluates the feasibility and differences of advanced statistical methods versus standard approaches in infant fNIRS data analysis.

Findings

Different analysis methods yield converging or diverging conclusions.

MVPA and GLM provide complementary insights compared to baseline correction.

The study informs best practices for infant fNIRS data analysis.

Abstract

In the last decade, fNIRS has provided a non-invasive method to investigate neural activation in developmental populations. Despite its increasing use in developmental cognitive neuroscience, there is little consistency or consensus on how to pre-process and analyse infant fNIRS data. With this registered report, we investigated the feasibility of applying more advanced statistical analyses to infant fNIRS data and compared the most commonly used baseline-corrected averaging, General Linear Model (GLM)-based univariate, and Multivariate Pattern Analysis (MVPA) approaches, to show how the conclusions one would draw based on these different analysis approaches converge or differ. The different analysis methods were tested using a face inversion paradigm where changes in brain activation in response to upright and inverted face stimuli were measured in thirty 4-to-6-month-old infants. By including more standard approaches together with recent machine learning techniques, we aim to inform the fNIRS community on alternative ways to analyse infant fNIRS datasets.