A practical guide to applying machine learning to infant EEG data.

Area of Science:

• Developmental Cognitive Neuroscience
• Machine Learning Applications in Neuroscience
• Infant Brain Activity Analysis

Background:

• Electroencephalography (EEG) is common in developmental cognitive neuroscience, but machine learning (ML) applications lag behind adult studies.
• Infant EEG analysis faces challenges: low trial counts, poor signal-to-noise ratio, and high inter-subject/inter-trial variability.
• Standard ML approaches struggle with the unique characteristics of infant EEG data.

Purpose of the Study:

• To provide a tutorial for applying ML to classify cognitive states in infant EEG data.
• To introduce methods addressing infant EEG data challenges, including Riemannian geometry for connectivity.
• To present adaptable ML pipelines for single- and multi-infant classification.

Main Methods:

• Description of common brain attributes used for EEG classification.
• Introduction of a Riemannian geometry-based approach for robust connectivity estimates.
• Development and demonstration of ML pipelines for infant EEG classification on an auditory oddball dataset.

Main Results:

• Successful classification of perceptual states in 12-month-old infants using EEG data.
• Demonstration of ML pipelines' effectiveness in handling inter-trial and inter-subject variability.
• Validation of the proposed methods on a standard infant EEG dataset.

Conclusions:

• The developed ML pipelines effectively classify cognitive states in infant EEG data.
• The Riemannian geometry approach enhances connectivity estimates, accounting for data variability.
• The provided pipelines and code are adaptable for various infant EEG research designs.