Publications
Selected work on sleep, brain health, and clinical AI.
A focused set of papers that define the current research arc: sleep-derived brain health, dementia risk, clinical automation, and continuous sleep physiology.
Brain health from sleep framework
Philosopher’s Stone
Brain Health from Sleep EEG: A Multicohort, Deep Learning Biomarker for Cognition, Disease, and Mortality (NEJM AI, 2026)
This paper introduces a sleep-derived brain health score built from overnight EEG using multitask deep learning. It shows that a single night of sleep contains enough structure to relate meaningfully to cognition, disease burden, and survival, including conditions such as dementia, depression, and hypertension, with higher scores associated with substantially lower mortality risk. From an AI perspective, the model learns a brain health representation directly from raw sleep signals rather than relying only on expert-defined features.
Brain aging
Sleep-based Brain Age and Dementia
Machine Learning–Based Sleep Electroencephalographic Brain Age Index and Dementia Risk (JAMA Network Open, 2026)
This study evaluates a sleep-derived brain age index across five longitudinal cohorts and shows that older-appearing sleep EEG is associated with substantially higher future dementia risk. It advances the case for sleep-based markers as early indicators of neurodegenerative vulnerability.
Clinical sleep AI
CAISR
CAISR: achieving human-level performance in automated sleep analysis across all clinical sleep metrics (Sleep, 2025)
CAISR addresses the full spectrum of clinical sleep analysis, including staging, arousals, breathing events, and limb movements, within a single automated framework. The aim is an end-to-end sleep analysis system that can hold up in clinical practice, not just a collection of strong results on isolated tasks.
Sleep physiology
OSD Sleep Depth
Ordinal Sleep Depth: A Data-Driven Continuous Measurement of Sleep Depth (Journal of Sleep Research, 2026)
OSD treats sleep depth as a continuous physiological quantity rather than a set of discrete labels. By learning sleep depth directly from EEG, it captures variation related to arousal probability, sleep-disordered breathing, age, and cognitive impairment with substantially greater resolution than conventional staging alone.