Video-based eye movements are linked to cortical and pupil-based arousal in human sleep

Abstract

Study ObjectivesTo quantify high-resolution video-based eye kinematics across sleep macro- and microstructure and determine their coupling with pupil-based and cortical arousal markers. MethodsWe recorded polysomnography and utilized infrared video-based eye-tracking in 17 healthy adults. Computer vision was employed to extract eye position and speed, which were related to pupil size (subcortical arousal marker) and EEG spectral slope (cortical arousal marker) across sleep stages, rapid eye movement (REM) substates, and K-complexes. ResultsEye kinematics varied significantly across sleep stages (p<0.001), except for horizontal pupil position (p=0.192). Video-based eye position and speed are sufficient to classify stages with above chance level accuracy (47%). Eye movement speed positively correlated with pupil size during non-REM sleep (R[&ge;]0.21, p<0.001) and with spectral slope during wakefulness and REM sleep (R[&ge;]0.11, p<0.018). These strengths of the correlations differ depending on the direction of the eye movement. Phasic REM exhibited faster eye movements, larger pupil size (p<0.001), and a flatter spectral slope (p=0.014) compared to tonic REM, indicating elevated subcortical and cortical arousal. K-complexes were accompanied by increased eye movement speed (p<0.05) and a steeper spectral slope (p[&le;]0.010), suggesting a transient shift toward a sleep-protective cortical state despite concurrent oculomotor activation. ConclusionsVideo-based eye-tracking reveals that eye movements are quantitatively coupled to brain-wide arousal fluctuations in a state-dependent manner. This methodology provides a ground-truth framework for resolving fine-grained eye dynamics during sleep, offering a high-fidelity tool for sleep phenotyping and clinical assessment. Statement of significanceTraditional sleep monitoring relies on low-resolution electrical signals that often confound true eye movements with brain or muscle activity. This study uses high-resolution video tracking to establish a physical ground truth for eye kinematics across human sleep. We demonstrate that eye movements are coupled with both cortical and subcortical arousal levels in a state-dependent manner, providing a clearer neurophysiological distinction between, for example, sleep substates like tonic and phasic REM. This framework reveals how the eyes serve as a non-invasive window into the brains internal arousal state. Such insights address critical gaps in understanding sleep microarchitecture and offer a novel pathway for developing high-fidelity biomarkers for neurological disorders, such as Parkinsons disease, characterized by disrupted sleep-related arousal.