Reference-Relation Guided Autoencoder with Deep CCA Restriction for Awake-to-Sleep Brain Functional Connectome...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Developmental Neuroscience

Background:

• Acquiring consistent resting-state fMRI in developing children is difficult, necessitating separate protocols for sleep and awake states.
• The transition between sleep and awake brain activity poses a challenge for longitudinal brain development studies, as awake-to-sleep connectome prediction remains largely unexplored.
• Existing image translation methods are inadequate for predicting brain functional connectome due to data scarcity and domain differences.

Purpose of the Study:

• To develop a novel method for predicting functional connectome from awake to sleep states in pediatric populations.
• To address the gap in longitudinal brain development research by enabling consistent functional connectome analysis across different states.
• To improve the accuracy and reliability of brain functional connectome prediction for developmental neuroscience.

Main Methods:

• Proposed a novel reference-relation guided autoencoder with deep CCA restriction (R2AE-dCCA) for awake-to-sleep connectome prediction.
• Utilized a reference-autoencoder (RAE) to augment limited paired data by incorporating age-restricted neighboring subjects as references.
• Incorporated a relation network to weigh reference subjects based on source domain similarity and employed deep CCA restriction to preserve relational structures during translation.

Main Results:

• The R2AE-dCCA method demonstrated superior prediction accuracy compared to state-of-the-art approaches.
• The proposed method effectively maintained the modular structure of the brain functional connectome.
• New validation metrics specifically designed for connectome prediction were introduced and utilized.

Conclusions:

• The R2AE-dCCA model offers a significant advancement in predicting functional connectome across different brain states (awake-to-sleep).
• This method facilitates more consistent and accurate longitudinal studies of brain functional development in children.
• The approach holds promise for improving our understanding of neurodevelopmental trajectories by bridging data acquired under varying conditions.