High-frequency broadband activity detected noninvasively in infants distinguishes wake from sleep states: Part 2

Follow-up to High-frequency broadband activity detected noninvasively in infants distinguishes wake from sleep states

High-frequency broadband activity (HFB) indexes local brain activity generated by neurons, but has been limited to invasive measures due to high-frequency signal drop off through the skull. Using scalp EEG, we revealed that HFB distinguishes wake from sleep states in infants with 90% reliability (Holubecki et al., 2025). The presence of thin skull and soft spots (fontanelles) between skull plates that have not yet fused creates the environment needed for HFB detection in infants using noninvasive measures, paving the way to study the infant brain across research and clinical settings.

Our research aims to uncover mechanistic explanations of the neural basis of human behavior, that is, move from where to how. Our goals are multifaceted: (1) advance fundamental science by discovering new knowledge using rigorous, reproducible methods; and (2) advance translational applications in neurotechnology, precision medicine, and product development that are grounded in rigorous science. This project (Part 2) addresses three goals, each linked to a primary result in Holubecki et al., 2025 (Part 1):

	Part 1: Result	Part 2: Goal
1.	HFB analyzed using MATLAB with FieldTrip	Reproduce HFB analysis using Python with MNE
2.	Hierarchical models using all available data revealed greater HFB power in wake vs. sleep states in midline and central channels near fontanelles, and occipital channels over thin skull	Classify wake vs. sleep states using all available data from a single channel: Fz HFB (located over the anterior fontanelle) should discriminate states, whereas P3 HFB (over relatively thick skull) should not
3.	Hierarchical models using limited data detected wake/sleep differences with 90% reliability	Classify wake vs. sleep states using the equivalent limited data from a single channel: Fz HFB should discriminate states, whereas P3 HFB should not

Results demonstrate classification within subjects based on noninvasive HFB detection with minimal training data, achieving above-chance accuracy. Within-subject standardization on the full sample further enabled robust cross-subject generalization.

Publications or other papers using these scripts and/or data should cite the original publication:

• Holubecki, AM, Yarbrough, JB, Rangarajan, V, Kuperman, R, Knight, RT, Johnson, EL. High-frequency broadband activity detected noninvasively in infants distinguishes wake from sleep states. bioRxiv (2025). DOI

Software:

• Python 3.12.12
• Environment: Google Colab
• Package versions:
  • NumPy 2.0.2
  • Pandas 2.2.2
  • SciPy 1.16.2
  • MNE 1.10.2
  • Matplotlib 3.10.0
  • Scikit-learn 1.6.1

Notes:

• Run using the notebook infant_hfb_ml.ipynb with hfb_utils.py and ml_utils.py uploaded to the Colab File panel.