This study details the development and potential application of a soft wearable near-infrared spectroscopy (NIRS) system for real-time, long-term monitoring of brain water dynamics during sleep. This device is designed for comfortable use in the home environment and contributes to deepening our understanding of sleep and brain health by revealing how changes in brain water are linked to glymphatic system activity. In particular, it presents new possibilities for assessing sleep quality and monitoring neurological diseases through analysis of changes in brain water dynamics and physiological rhythms during each sleep stage. 💡

1. Introduction: Importance of sleep and brain health and limitations of existing research

Sleep is an essential physiological process for memory processing, cognitive function, and neural recovery. If sleep is disturbed, metabolic wastes can accumulate in the brain, causing problems with cognitive function and memory formation. The glymphatic system plays an important role in removing these metabolic wastes. The glymphatic system removes waste products and regulates central nervous system activity during deep sleep through cerebrospinal fluid (CSF) circulation. Therefore, monitoring brain water dynamics, which reflects cerebrospinal fluid redistribution, may be a very important biomarker for understanding the relationship between sleep disorders and neurological diseases.

However, there are currently several difficulties in directly tracking cerebrospinal fluid flow. 💉 For example, invasive methods such as two-photon microscopy have limited clinical application due to safety concerns, while MRI has the disadvantages of being expensive, limiting patient movement, and being difficult to use in a natural sleeping environment. Polysomnography (PSG), the standard for sleep assessment, is also bulky, expensive, and can only be used in hospital settings, making it unsuitable for long-term home monitoring.To overcome these limitations, the research team developed a fully integrated, gentle, wireless, skin-adherent brain water near-infrared spectroscopy (NIRS) system. The device offers a high signal-to-noise ratio (SNR) and excellent fit, enabling reliable, long-term monitoring during sleep. It is specifically designed to use three wavelengths to quantify changes in brain water content and simultaneously capture physiological rhythms such as breathing, heart rate, and slow oscillations. The goal of this study is to demonstrate that this soft NIRS system allows continuous, non-invasive monitoring of brain water dynamics and associated physiological signals across sleep stages. 🌟

2. Soft wearable NIRS system overview and design

The soft wearable NIRS system developed in this study is designed to provide continuous, real-time monitoring of brain water dynamics during sleep at home. In particular, we wanted to overcome the limitations of MRI, which include cost and limitations in a natural sleep environment.

2.1 System overview and functions* What we measure: Measure brain water dynamics in both waking and sleeping states.

• Validation: Direct correlation between sleep stage classification and brain water changes was analyzed by comparing electroencephalogram (EEG) and electro-oculogram (EOG) signals simultaneously acquired from a commercial device.
• Glymphatic system connection: Cerebrospinal fluid plays an important role in removing metabolic waste products such as amyloid-β during deep sleep. Taking into account the difficulty of directly measuring cerebrospinal fluid dynamics throughout the brain, this device aims to capture cerebrospinal fluid changes associated with glymphatic system activity by tracking state-dependent changes in brain water content measured at the forehead.
• Comfortable fit: Lightweight, flexible polymer that adheres closely to the skin has been used to increase user convenience. 😌
• Multiple Wavelengths: Obtain a variety of optical signals by integrating LEDs with 640nm, 680nm, and 950nm wavelengths and photodetectors that can detect multiple wavelengths.
• WIRELESS & WEARABLE: LEDs mounted on a flexible circuit board allow the device to bend to the curves of your skin for reliable data acquisition even when you move, and operate wirelessly for natural sleep monitoring in the home environment.

  "This is the first time a wearable device has detected brain water dynamics under home sleep conditions."

• AI-based analysis: Recorded EEG signals are classified by a hybrid model that combines threshold-based detection and AI machine learning algorithms. This allows you to accurately classify sleep stages (wake, NREM, and REM sleep) and simultaneously capture various physiological rhythms such as breathing, heart rate, and slow oscillations. Brain water is quantified via the Lambert-Beer law. This multimodal analysis not only increases the accuracy of sleep stage classification but also helps identify broad physiological mechanisms that influence sleep quality.

2.2 Device design and characteristicsThis soft device consists of three main layers:

1. Adhesive layer: Attached to the skin.
2. Flexible printed circuit board (fPCB): Contains LEDs and detectors. 💡
3. Silicone encapsulation layer: Provides mechanical protection and comfort.

• Optical Components: Integrated LEDs with wavelengths of 640nm, 680nm, and 950nm and a photodetector (AS7341) allow precise tracking of signals at multiple wavelengths simultaneously.
• Electronic Components: Integrated Bluetooth low energy (BLE) microcontroller, voltage regulator, battery, and antenna enable wireless communication and efficient signal processing.
• Battery Life: Powered by a 110 mAh Li-polymer battery, operating for approximately 5.5 hours under continuous LED lighting and wireless data streaming conditions. We plan to extend operating time through power optimization in the future.
• Mechanical Stability: Finite element analysis confirmed that the device operates reliably without mechanical damage under typical adult forehead curvatures as well as harsher infant scalp curvatures. 🛠️
• Communication and Signal Processing: The device emits near-infrared light via multiple LEDs and detects the reflected signal with an integrated photodetector. The acquired optical signals are digitized by an onboard BLE microcontroller and transmitted wirelessly to an external computer for further analysis.
• Signal-to-Noise Ratio (SNR): The flexible device demonstrated significantly better signal stability than the traditional rigid device under a variety of movement conditions, including sitting, walking, and climbing.
• Thermal Safety: Thermal safety has been verified through simulations and experiments to ensure that skin surface temperature remains below the clinically acceptable 41°C safety threshold even during extended operation. 🔥

3. Verification of NIRS system performance for brain water monitoringThis NIRS system is designed to effectively capture the operating principles of the glymphatic system and brain water dynamics.

3.1 Principle of the glymphatic system

• Cerebrospinal fluid (CSF) circulation: Cerebrospinal fluid enters the brain along the perivascular space around the arteries, moves to the interstitial space through Aquaporin-4 (AQP4) channels, exchanges with interstitial fluid (ISF), and then removes metabolic wastes through veins. This is essential for maintaining brain homeostasis and preventing neurological diseases.
• Non-invasive monitoring: Evaluation of cerebrospinal fluid dynamics via MRI is possible, but is not suitable for continuous monitoring during sleep. This soft NIRS device noninvasively captures physiological processes associated with glymphatic activity by tracking state-dependent changes in brain water measured at the forehead.

3.2 Verification of depth detection ability through Monte Carlo simulation

• Photon propagation simulation: Through Monte Carlo simulation, we analyzed how photons pass through multiple layers of head tissue, including skin, skull, and cerebrospinal fluid.
• Cortical Depth Sensing: Photon density heatmaps showed that although photon concentration is highest in superficial tissue, photon trajectories are not limited to superficial layers but extend beyond the cerebrospinal fluid layer into deeper brain tissue. This means that the NIRS system is suitable for measuring water changes in cortical brain tissue.

3.3 Measurement of brain water dynamics during daily activities and respiratory control* Water changes during activity and rest: Brain water changes (Δ[H2O]) were measured during alternating exercise and rest cycles using an NIRS device.

* **During exercise**: Δ[H2O] initially increased and then decreased mid-activity, likely reflecting increased cerebral and peripheral blood flow due to increased metabolic demands and increased plasma hydration to tissues.
* **At rest**: Δ[H2O] initially decreased and then increased again, possibly reflecting rebalancing of interstitial and cerebrospinal fluid following decreased vascular tone and perfusion. 🏃‍♀️↔️ 😴

• Water changes during controlled breathing: Δ[H2O] was measured during alternating cycles of normal breathing and intentional breath holding.
  • During breath holding: Δ[H2O] showed a characteristic pattern of initial increase followed by a gradual decrease before returning to baseline. This reflects the combined effects of hypercapnia-induced vasodilatation and intracranial fluid shift, including cerebrospinal fluid redistribution.
  • During normal breathing: Δ[H2O] showed a small, consistent increase, reflecting stable ventilation, carbon dioxide removal, and maintenance of vascular tone.
• Stability and reliability: These state-dependent changes were reproducible even over repeated cycles and did not require device recalibration, demonstrating the robustness of the system for long-term monitoring.

"Monte Carlo simulations, activity/rest validation, and respiration validation show consistent conclusions that Δ[H2O] reflects physiologically meaningful brain water dynamics and can be detected noninvasively by wearable systems."

4. Determination of sleep stages and assessment of brain water dynamics

To assess the relationship between brain water dynamics and sleep stages, the team combined EEG and EOG with a NIRS device to record overnight.

4.1 Sleep stage classification methodologyThe research team used three sleep stage classification approaches:

1. Machine Learning (ML) Model: We compared the performance of traditional machine learning algorithms and deep learning models (MLP, CNN, LSTM).
2. Threshold-based model: wakefulness, non-REM sleep (NREM), and REM sleep states were distinguished based on sigma (σ) band power. 😴
   • If the sigma band power (10-15 Hz) was above 60% of the moving average, it was classified as NREM, and if it was below that, it was classified as REM.
   • If total power exceeds 50%, it is classified as awakened and takes precedence over sigma band determination.
   • This method was strong for NREM detection (95.9%), but weak for REM detection (approximately 70%).
3. Hybrid model: Combines sigma band-based thresholding with a supervised learning classifier trained with the full feature set.
   • This model leverages the strengths of both approaches to significantly improve accuracy, especially in detecting REM sleep, which is difficult to classify.
   • Comprehensive classification accuracy improved to 80-90% and consistently outperformed existing ML and deep learning models.

4.2 Association between brain water signals and sleep stages* Multi-taper spectrogram analysis: found oscillatory patterns in the Δ[H2O] signal that are temporally consistent with EEG-based sleep stage dynamics.

• Distinct spectral peaks: In the spectral analysis of the Δ[H2O] signal, several distinct peaks were observed:
  • Low frequency brain water oscillations (~0.05 Hz)
  • Respiratory activity (RS) (~0.3 Hz)
  • Slow oscillation-related NIRS oscillation (SONO) (0.6–0.7 Hz)
  • Cardiac Cycle (CC) (~0.8–1.2 Hz)
  • In particular, SONO shows a frequency band that overlaps with the slow oscillations (0.5 Hz) of the EEG, suggesting a potential link with electrophysiological activity.
• Temporal alignment: We found that changes in brain water dynamics are tightly synchronized with EEG-defined sleep stage transitions, with virtually no time delay. This reflects rapid coordination between brain water dynamics and neural state transitions. ⏰

  "Median latencies were −1.45 seconds in wakefulness, 0 seconds in REM sleep, and 0 seconds in NREM sleep."

  "These distributions are centered near zero across all states, showing minimal temporal offset between Δ[H2O] transitions and EEG-defined sleep stage boundaries, demonstrating tight temporal alignment between NIRS-derived brain water dynamics and neural state transitions."

5. Analysis of brain water dynamics and physiological connectivity by sleep stage

Overnight recording places greater demands on device stability, user comfort, and signal robustness. This study demonstrated the temporal relationship between sleep stages and brain water dynamics through overnight recordings from a single participant. 🌙

5.1 Analysis of overnight recorded data* Stable Measurements: The wearable device maintained stable, high-quality measurements throughout the night.

• Sleep stage Δ[H2O]: Sleep stages categorized by EEG/EOG signals were aligned with the Δ[H2O] signal, which captures band-limited spectral features corresponding to respiratory signals (RS), slow oscillation-related NIRS oscillations (SONO), and cardiac cycle (CC).
• Change in SONO: Average SONO peak power was lowest during wakefulness, increased slightly during REM sleep, and was highest during non-REM sleep (NREM). NREM sleep is known for its slow, synchronized neural activity, which is associated with changes in brain fluid dynamics.
• Changes in RS: RS dynamics showed marked differences across sleep stages.
  • During REM sleep: RS peak was centered at approximately 0.35 Hz and was low in power and wide, indicating irregular breathing. 🌬️
  • During NREM sleep: RS slowed to about 0.25 Hz, peaks narrowed and power increased, reflecting more regular and forceful breathing.
• Changes in CC: During NREM sleep, the 0.9 Hz peak was higher in power and narrower, indicating a more regular heart frequency range. Upon transition to REM sleep, this peak broadened and decreased in power, reflecting increased variability in cardiac frequency range dynamics. ❤️

  "These changes were tightly synchronized with EEG-defined transitions, demonstrating physiological sensitivity to Δ[H2O]."

• Sleep state separation: We found that sleep states can be partially separated by projecting the features for slow oscillations, RS, and CC into a joint feature space. This suggests that wearable optical devices may provide complementary support for basic sleep monitoring.

5.2 Analysis of cumulative brain water changes

When we analyzed cumulative brain water trajectories from multiple overnight recordings of participants, we discovered the following characteristics:* Increase upon entry into NREM: Cumulative brain water increased upon entry into NREM from wakefulness and REM sleep, and continued to increase throughout the initial and continued duration of NREM sleep.

• Decreased upon entering REM: Cumulative brain water decreased during the transition from NREM to REM sleep.
• Sleep stage dependence: These direction-specific patterns suggest that brain water dynamics are dependent on sleep stage transitions. The results are consistent with existing models that NREM sleep is associated with greater fluid accumulation than REM sleep.

6. Conclusion

This study demonstrated the feasibility of a wireless soft wearable NIRS device for real-time, long-term monitoring of brain water dynamics associated with sleep stage transitions in a home environment. The device integrates multi-wavelength LEDs and photodetectors with flexible circuitry and soft encapsulation to maximize skin fit, minimize motion artifacts, and increase user comfort.

An in vivo study with multiple subjects in a home environment demonstrated the ability to monitor sleep throughout the night and identified persistent changes in brain water dynamics across different sleep stages. Spectral analysis revealed Δ[H2O]-derived spectral features related to respiration, cardiac activity, and slow oscillatory range dynamics (RS, CC, SONO).

Although this study did not directly measure cerebrospinal fluid flow or solute clearance, the observed sleep stage-dependent brain water dynamics occur within the physiological range in which glymphatic activity is regulated. Glymphatic function is known to be enhanced during NREM sleep and decreased during REM sleep, which is also associated with changes in neural activity, vascular dynamics, and intracranial fluid regulation.

The ability of the wearable NIRS platform to capture brain water dynamics at high temporal resolution during natural sleep suggests its potential utility as a noninvasive tool for long-term monitoring of glymphatic-related brain fluid dynamics.Ultimately, this soft wearable NIRS device could be utilized for long-term sleep monitoring and complementary assessment of sleep physiology, and would be particularly useful in neurological diseases where sleep and brain fluid dynamics are altered. Furthermore, the device has the potential to be integrated into next-generation wearable healthcare platforms to enhance personalized home healthcare solutions. In the future, further studies are needed to improve spatial coverage and account for surface contributions, while establishing its clinical role through comparisons with reference measurements such as contrast-enhanced MRI. 🚀