Abstract
<jats:p>Cerebrospinal fluid (CSF) dynamics are clinically important but can presently be measured only with expensive, non-portable imaging or invasive monitoring, precluding scalable assessment. We investigated whether CSF flow in the cerebral aqueduct can be reconstructed from noninvasive physiology recorded during sleep. In 22 healthy adults, with independent validation in 14 additional adults, regularized linear and ensemble regressors predicted the cardiac-cycle CSF flow waveform from electrocardiography, photoplethysmography (PPG), respiration, and electroencephalography acquired during overnight polysomnography, using 7-Tesla phase-contrast MRI as ground truth. Waveforms were reconstructed with high fidelity and generalized to the validation cohort, and deriving physiological metrics from the reconstructed waveform estimated peak velocity more accurately than predicting it directly. A minimal, wearable-oriented PPG feature set retained performance. These results support the feasibility of a noninvasive, potentially wearable digital biomarker of CSF dynamics and motivate evaluation in neurological disease populations.</jats:p>