An academic-industrial partnership for AI-based sleep staging in the elderly using an EEG headband and a smartwatch

Investigators:
Joyita Dutta, UMass Amherst

MassAITC Cohort: Year 2 (AD/ADRD)

Project Accomplishments: Through an academic-industrial partnership between UMass Amherst and CGX Inc., this project focused on developing AI-based sleep staging for elderly populations using a CGX EEG headband and an Apple Watch. The project collected concurrent sleep data from 50 adults aged 65 and older (including 20 with at least two Alzheimer’s disease risk factors) using FDA-cleared polysomnography (PSG) as the gold standard alongside wearable devices over 7-night monitoring periods. The three aims were to benchmark wearable device sleep staging performance against PSG, develop a transfer learning framework for improved EEG-based sleep staging, and create a multimodal AI model combining EEG and smartwatch data for more robust sleep characterization in older adults. 

The results demonstrated that the Apple Watch achieved 63.5% sleep staging accuracy compared to PSG, while a transfer learning framework improved the CGX headband’s accuracy from 56% to 68%. The multimodal AI model combining both devices achieved 70% accuracy and reduced estimation errors across key sleep metrics by up to 50% compared to single-device approaches, demonstrating the value of multi-sensor fusion for sleep assessment in aging populations. Key project outputs include a peer-reviewed publication in IEEE Transactions on Biomedical Engineering, the release of the BIDSleep mobile application on the Apple App Store for sleep data collection, and plans to share a de-identified dataset on the PhysioNet repository as a public digital health resource for advancing sleep research and AI model development in senior populations. 

Initial Proposal Abstract: This academic-industrial partnership between UMass Amherst and CGX Systems will develop AI techniques for sleep staging in seniors (>65 years) using multimodal data from two wearable devices. We will validate a wearable EEG headband and a smartwatch for sleep staging and characterization in the elderly. Sleep disturbances are among the earliest observable symptoms of Alzheimer’s disease (AD). While many wearables are available for sleep monitoring, most studies are based on young subjects. Our work will impact early identification of seniors at risk of AD, opening up a large elderly market for wearable devices like EEG headbands and smartwatches.





Outcomes: