Examining dynamic functional connectivity during sleep in neonates using machine learning and HD-DOT
Poster No:
2432
Submission Type:
Abstract Submission
Authors:
Katharine Lee1, Topun Austin2, Rob Cooper3, Andrea Edwards2, Jem Hebden3, Kelly Pammenter2, Julie Uchitel4, Borja Blanco5
Institutions:
1University of Cambridge, Cambridge, United Kingdom, 2Cambridge University Hospitals NHS Foundation Trust, Cambridge, Cambridgeshire, 3Department of Medical Physics and Biomedical Engineering, University College London, London, Greater London, 4Department of Paediatrics, University of Cambridge, Cambridge, Cambridgeshire, 5Department of Psychology, University of Cambridge, Cambridge, Cambridgeshire
First Author:
Co-Author(s):
Rob Cooper
Department of Medical Physics and Biomedical Engineering, University College London
London, Greater London
Department of Medical Physics and Biomedical Engineering, University College London
London, Greater London
Jem Hebden
Department of Medical Physics and Biomedical Engineering, University College London
London, Greater London
Department of Medical Physics and Biomedical Engineering, University College London
London, Greater London
Introduction:
Sleep is a critical factor in early brain development due to its impact on memory consolidation, synaptic plasticity, and neural network maintenance. High-Density Diffuse Optical Tomography (HD-DOT), a functional near-infrared spectroscopy (fNIRS) technology, has been used to investigate static resting state functional connectivity (FC) during active sleep (AS) and quiet sleep (QS) states in term-aged infants [1]. Dynamic FC analysis investigates time-varying patterns in brain activity to shed light on the non-stationary nature of resting state brain functionality. One functional magnetic resonance imaging (fMRI) method proposed for this objective identifies recurring co-activation patterns (CAPs) using machine learning clustering algorithms [2]. These CAPs represent instantaneous brain configurations at single time points and provide insight into the dynamics of spontaneous neural activity.
Objective: This HD-DOT study examines dynamic functional connectivity during newborn infant sleep to shed light on early brain development in relation to sleep states. Specifically, this study adapts CAP analysis, an fMRI approach that employs unsupervised machine learning, for HD-DOT data.
Objective: This HD-DOT study examines dynamic functional connectivity during newborn infant sleep to shed light on early brain development in relation to sleep states. Specifically, this study adapts CAP analysis, an fMRI approach that employs unsupervised machine learning, for HD-DOT data.
Methods:
In this observational study, HD-DOT data was acquired from a cohort of sleeping newborn infants at the Rosie Hospital, Cambridge UK (n=40, mean postmenstrual age=40+2). These datasets were classified as AS or QS based on behavioural analysis of synchronized video footage. The top 25% of frames for somato-motor and frontal networks were selected for each participant using a seed signal. Activation maps at these frames were clustered using the K-means algorithm into CAPs. CAP consistency was assessed by measuring intra-CAP spatial correlation. Other metrics such as CAP presence and CAP transition rate were compared in AS and QS datasets using rank sum t-tests to investigate how sleep states may modulate resting state networks.
Results:
Distinct CAPs were identified for AS and QS datasets, characterizing unique connectivity dynamics within somato-motor and frontal regions. These CAPs were found to have high consistency scores (left frontal region AS=0.51±0.03 and QS=0.50±0.04; left central region AS=0.53±0.04 and QS=0.51±0.04). Across iterations of CAP analysis, consistent trends emerged. Notably, CAPs with unilateral activation appear more frequently in the QS dataset.
Conclusions:
This study is the first to apply CAP analysis to HD-DOT infant data, demonstrating the utility of unsupervised machine learning in dynamic FC analysis. Preliminary examination of CAP metrics reveals potential differences between AS and QS dynamic FC which may shed light on the function of sleep during early brain development. This study also focuses on dynamic FC as opposed to static FC, providing unique insight into the non-stationary nature of early resting state networks. The CAPs found in this study and the methodology used to identify them will support future infant studies that elucidate the relationship between sleep and neuronal connectivity in the developing brain.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
Novel Imaging Acquisition Methods:
NIRS 1
Perception, Attention and Motor Behavior:
Sleep and Wakefulness
Keywords:
Computational Neuroscience
Machine Learning
Near Infra-Red Spectroscopy (NIRS)
PEDIATRIC
Sleep
Other - HD-DOT
1|2Indicates the priority used for review
Provide references using author date format
Liu, X. (2018). Co-activation patterns in resting-state fMRI signals. NeuroImage, 180, 485-494.