Frequency-Domain fNIRS Biomarkers for Non-REM Sleep Staging: Euclidean Peak Distance Validated with EEG

Polysomnography remains the gold standard for sleep staging but is cumbersome and impractical for routine real-world use. This study evaluates functional near-infrared spectroscopy (fNIRS) as a wearable modality for sleep monitoring and examines frequency-domain features across a broad physiological spectral range. Eighteen healthy adults underwent simultaneous fNIRS and EEG recording during sleep, with fNIRS optodes positioned over the right dorsolateral prefrontal cortex, during which two-hour sleep data were collected. After quality inspection, data from 14 participants were retained, and a subset of eight participants who exhibited all non-REM stages (i.e., N1-N3) was used for repeated-measures analysis. Although the final analytic sample was small (n = 8), post-hoc power analysis indicated sufficient power to detect the dominant cardiac-related effects. Concentration changes of oxy-hemoglobin (𝛥HbO) and deoxy-hemoglobin (𝛥HbR) signals were preprocessed, segmented into 30-s epochs (to align with EEG epochs), and transformed into the frequency domain. Repeated-measures ANOVA and Friedman tests showed that Euclidean Peak Distance (EPD) and peak value/frequency features in the cardiac and harmonic bands significantly differentiated adjacent non-REM stages N1-N2 and N2-N3. The strongest effect was observed for Cardiac EPD (p GG = 0.017; p Friedman = 0.021). The dominant cardiac peak frequency shifted toward lower frequencies with increasing sleep depth, while its peak amplitude increased during N3, indicating slower but stronger cardiac-driven hemodynamic oscillations. N3 spectra also showed a secondary sub-peak between the respiratory and cardiac bands, suggesting enhanced autonomic-vascular coupling during deep sleep. 𝛥HbO features showed stronger discrimination across sleep stages than 𝛥HbR, although both exhibited similar depth-dependent trends. These findings establish frequency-domain fNIRS features, particularly EPD, as physiologically interpretable markers of sleep depth and support the potential of wearable fNIRS for non-invasive sleep monitoring.