Science
Sleep as a brain-health signal.
My work asks how overnight physiology can become rigorous measurement for cognition, disease risk, recovery, and response to intervention.
Every night, the brain runs an experiment on itself. Electrical rhythms shift, breathing changes, heart rate rises and falls, muscles quiet down, tiny awakenings appear and vanish, and recovery has to compete with stress, illness, age, light, noise, medications, and whatever else life brought into the bedroom. A sleep study records much of this. The clinical report then compresses a whole night into labels and counts: wake, REM, deep sleep, apneas, arousals, limb movements. Useful labels, yes. Too useful to throw away. Also too thin to be the whole story.
My work has followed two connected lanes.
Improving clinical sleep analysis
The first lane is improving clinical sleep analysis. Sleep medicine depends on expert scoring, specialized equipment, and workflows that can be slow, expensive, and awkward for real patients. I have worked on ways to pull reliable sleep and breathing information from signals that are easier to collect: heart rhythms, respiratory effort, oxygen saturation, wearable bands, and contactless sensors. The aim is practical: more scalable measurement, less friction for subjects, and richer data from the places where sleep actually happens.
CAISR is the cleanest expression of that lane so far. It moved the problem from isolated scoring tasks to the whole clinical sleep report: staging, arousals, respiratory events, and limb movements in one comprehensive system. In validation settings, CAISR reached trained-technologist-level performance across the core tasks, with comparisons matching or exceeding expert reliability, while performance still varies by task and dataset.
From sleep EEG to brain-health biomarkers
The second lane asks what these night signals can tell us once scoring is done. Sleep physiology carries traces of memory systems, vascular and metabolic stress, neurodegeneration, psychiatric disease, and aging. Spindles are brief bursts of coordinated sleep EEG activity, and their shape, density, and frequency can relate to cognition. REM sleep is stranger and harder to summarize, yet REM duration and structure have shown links with cognition, brain anatomy, mood, and neurodegenerative disease. Slow waves, arousals, breathing instability, and continuous sleep depth add other pieces of the picture.
Across this line of work, we linked sleep features to fluid cognition, brain structure, mild cognitive impairment, dementia, future neurologic and cardiovascular outcomes, and mortality. This is where sleep begins to serve as a candidate biomarker in research: a way that may help measure risk, track recovery, evaluate whether interventions are moving physiology in the intended direction, or give clinical trials a more sensitive endpoint.
A brain-health signal
The most complete expression of that second lane, so far, is Brain Health from Sleep EEG, the paper behind what we call the Philosopher’s Stone project. The question was deliberately large: can a single night of EEG be distilled into a brain-health signal that relates to cognition, disease burden, and survival? We trained AI models across large cohorts and found that overnight EEG held a broader, more integrated signal than conventional features alone. I like the alchemy joke because the raw material really is unglamorous: hours of voltage traces while someone sleeps. The serious point is that the conversion is measurable. Messy overnight physiology can become an index of brain health.
From measurement to decisions
That scientific arc also points toward a more personal decision cycle. If sleep carries brain-health information, then sleep can become part of an individualized protocol loop: measure, interpret, adjust, measure again. A person might combine wearable data for regularity and recovery with occasional sleep EEG, cognitive context, subjective experience, and carefully chosen changes in timing, light, exercise, travel schedule, or clinical follow-up when needed. The questions become concrete. Did REM rebound after a stressful month? Did spindle features or sleep depth shift after a training block? Did breathing stabilize with treatment? Did the person feel sharper, steadier, or better recovered?
That future has to stay disciplined. A dashboard full of numbers is easy. Interpretation is harder. The useful version is restrained: sleep as one rigorous entry point into brain health, tied to physiology, uncertainty, and repeated decisions.
I am interested in that bridge from the sleep lab to decisions people care about. The night has terrible handwriting, but it keeps writing. My job is to make more of it readable.