Brain Health from Sleep EEG: A Multicohort, Deep Learning Biomarker for Cognition, Disease, and Mortality

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomarker Discovery

Background:

• Objective biomarkers for brain health are lacking, despite sleep's crucial role in cognition and disease prevention.
• Overnight sleep electroencephalography (EEG) presents a potential substrate for a comprehensive brain health biomarker.
• This study investigates a deep learning framework to derive such a biomarker from sleep EEG.

Purpose of the Study:

• To develop and validate a deep learning framework for analyzing sleep EEG data.
• To create a latent representation of brain health from EEG.
• To distill this representation into a single, interpretable brain health score.

Main Methods:

• Analysis of 36,000 polysomnography recordings from 27,000 subjects across six cohorts.
• Utilized a multitask deep neural network trained end-to-end on EEG time series and spectrograms.
• Compared deep learning model performance against demographic baselines, conventional EEG metrics, and classic machine learning approaches.

Main Results:

• Deep learning-derived brain health scores significantly outperformed demographic and expert-defined EEG feature models.
• Correlations with cognitive outcomes improved to r=0.40, and disease classification AUC improved to 0.65-0.75.
• A one-standard-deviation increase in the brain health score was associated with a 31%-35% reduced risk of mortality (P<0.0001).

Conclusions:

• A multitask, end-to-end deep learning approach successfully generated an interpretable, sleep-derived brain health biomarker.
• The framework provides a robust index of brain health by modeling cognition, disease, and mortality.
• Potential for extension to additional data modalities to enhance clinical utility.