TMN-LAKDE: Characterizing Sleep Instability via Prediction Intervals of Dynamic EEG Spectral Networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Sleep Science

Background:

• Sleep instability, a hallmark of insomnia, signifies the brain's difficulty in maintaining stable states.
• Precise quantification of sleep instability remains a significant challenge in sleep research.

Purpose of the Study:

• To propose and validate a novel interval prediction framework for characterizing sleep instability.
• To investigate whether increased unpredictability in brain network dynamics underlies sleep instability.
• To establish a mechanistic biomarker for sleep instability.

Main Methods:

• Developed an interval prediction framework integrating temporal mobile network (TMN) for state prediction and local adaptive kernel density estimation (LAKDE) for uncertainty quantification.
• Generated prediction intervals (PIs) to measure the uncertainty in brain network dynamics.
• Validated the framework on SIESTA and Sleep-EDF databases.

Main Results:

• The proposed method successfully constructed well-calibrated prediction intervals.
• Subjects with sleep disorders exhibited significantly higher PI normalized average width (PINAW) compared to healthy controls.
• Wider PIs were validated as a mechanistic biomarker for sleep instability.

Conclusions:

• The interval prediction framework effectively quantifies sleep instability by measuring prediction uncertainty.
• Increased PI normalized average width serves as a reliable biomarker for sleep instability.
• Dynamic instability, quantified by prediction uncertainty, shares physiological underpinnings with sleep fragmentation.