Summary: A deep neural network trained on overnight EEG recordings can predict an individual’s brain age with high accuracy, and deviations between EEG-predicted brain age and chronological age are associated with several common medical conditions.
Source: AASM (American Academy of Sleep Medicine)
A recent study demonstrates that a deep neural network can reliably estimate the brain age of healthy adults from electroencephalogram (EEG) data collected during clinical overnight sleep studies. The EEG-predicted age and derived brain age indices showed distinct patterns across populations with diverse medical conditions, suggesting potential clinical utility as a biomarker of brain health.
Using raw EEG signals recorded during polysomnography, researchers developed and validated a deep learning model that estimates a person’s brain age. The model achieved a mean absolute error (MAE) of just 4.6 years when predicting chronological age from EEG data, indicating strong overall accuracy for age prediction based on sleep EEG patterns.
To train and evaluate the model, the team used a large clinical dataset: 126,241 overnight sleep studies for training, 6,638 studies for validation, and a separate holdout test set of 1,172 studies. These study volumes provided diverse EEG examples across age ranges and sleep physiology, allowing the neural network to learn age-related patterns in electrophysiological signals recorded throughout the night.
The investigators derived two complementary metrics from the EEG predictions. Brain Age Index was calculated as the difference between EEG-predicted age and the individual’s chronological age. Absolute Brain Age Index is the absolute value of this difference, quantifying the magnitude of deviation regardless of direction. Analyses adjusted for covariates such as sex and body mass index (BMI) to reduce confounding influences.
Significant associations emerged between higher Absolute Brain Age Index values and several neurological and sleep-related conditions. The study found statistically significant relationships between increased absolute brain age deviation and diagnoses including epilepsy and seizure disorders, prior stroke, and markers of sleep-disordered breathing such as elevated apnea-hypopnea index and frequent arousals. Low sleep efficiency was also linked with greater deviation from chronological age.
Beyond these findings, the researchers observed that people with diabetes, depression, pronounced excessive daytime sleepiness, hypertension, and self-reported memory or concentration problems tended, on average, to have elevated Brain Age Index values compared with the healthy population sample. According to the authors, these patterns indicate that certain health conditions are reflected in deviations between electrophysiological brain age and chronological age.
Lead author Yoav Nygate, senior AI engineer at EnsoData, emphasized the model’s precision: “While clinicians can only grossly estimate or quantify the age of a patient based on their EEG, this study shows an artificial intelligence model can predict a patient’s age with high precision.” He noted that the model’s accuracy allows shifts in predicted brain age relative to chronological age to correlate meaningfully with disease groups and common comorbidities.
The research team suggests that these AI-derived deviations may help identify novel clinical phenotypes embedded in physiologic signals. For example, systematic differences between EEG-predicted age and actual age could flag individuals at higher risk for particular neurological or cardiometabolic conditions, prompting further clinical evaluation or targeted monitoring.
The authors also framed the work as an initial step toward a potential biomarker of brain health. “The results in this study provide initial evidence for the potential of utilizing AI to assess the brain age of a patient,” Nygate said. “Our hope is that with continued investigation, research, and clinical studies, a brain age index will one day become a diagnostic biomarker of brain health, much like high blood pressure is for risks of stroke and other cardiovascular disorders.” The investigators stress that additional prospective studies and clinical validation are needed before widespread clinical adoption.
The research abstract was published in an online supplement of the journal Sleep and was scheduled for presentation as a poster beginning June 9 during Virtual SLEEP 2021. SLEEP is the annual meeting of the Associated Professional Sleep Societies, a joint venture of the American Academy of Sleep Medicine and the Sleep Research Society.
Funding: This study was supported by EnsoData, a healthcare artificial intelligence company. EnsoData’s initial clinical product, EnsoSleep, provides AI-driven scoring and automated event detection for sleep studies.
About this AI and sleep research news
Source: AASM (American Academy of Sleep Medicine)
Contact: Corinne Lederhouse – AASM
Image: The image is in the public domain
Original Research: The findings were presented during Virtual SLEEP 2021