World Sleep Industry ENCYCLOPEDIA

near-infrared brain function monitoring devices for sleep research (fNIRS)

2.2.3

Definition, measurand, and research boundary

Near-infrared brain-function monitoring equipment for sleep research generally refers to fNIRS systems that place near-infrared sources and detectors on the scalp and estimate changes in oxygenated and deoxygenated hemoglobin from wavelength-dependent changes in tissue light absorption and scattering. Continuous-wave, frequency-domain, and time-domain instruments use different emission, detection, and quantification methods. Common continuous-wave systems use the modified Beer-Lambert relationship to estimate concentration changes relative to a baseline; these relative changes should not be mislabeled as precise absolute cerebral oxygenation values.

fNIRS is primarily sensitive to hemodynamics near the superficial cortex and relates only indirectly to neural activity through neurovascular coupling. Its signals also include scalp, skull, and systemic circulatory contributions. A change in HbO, HbR, or total hemoglobin is therefore not a measurement of neuronal firing, and an increase or decrease cannot simply be equated with excitation or inhibition. Coverage, depth sensitivity, and spatial resolution jointly depend on probe geometry, tissue anatomy, and analysis; a scalp channel label alone does not prove an exact cortical origin.

This entry is limited to adjunctive brain-function or cerebral-hemodynamic measurement in sleep research. It is distinct from cerebral tissue oximeters, pulse oximeters, therapeutic near-infrared devices, and generic sleep-stage classifiers. Synchronized fNIRS and PSG can investigate hemodynamic changes associated with stage, arousal, respiratory events, or interventions, but fNIRS generally does not replace EEG, EOG, EMG, airflow, respiratory effort, or oxygen saturation, and research capability does not automatically confer a diagnostic or clinical-monitoring indication.

Optical hardware, probe arrays, and localization

A system includes one or more near-infrared sources, detectors, drive and acquisition electronics, fibers or direct-contact probes, a holder or cap, synchronization interfaces, and analysis software. Documentation should identify wavelengths, source power or drive scheme, sampling rate, detector type, gain, dynamic range, multiplexing, source-detector separation, and the source-detector pair for every channel. A brand name or channel count cannot reconstruct the measurement. The same source-detector pair may also produce separate measurement channels at different wavelengths.

Array design should follow the research question and target cortex, with optode positions recorded through head landmarks, a standard placement system, three-dimensional digitization, or image registration. A hairless forehead is convenient but does not represent the whole brain. Hair, hair color, skin pigmentation, head circumference, skull thickness, and anatomy alter coupling and photon paths. If groups systematically differ in coverage or signal loss, deleting low-quality channels can turn a measurement bias into an apparent physiological difference.

Long-separation channels contain both deeper and superficial contributions. Short-separation channels are more sensitive to superficial scalp layers and can help model that interference, but short must be defined for the instrument, age, and head geometry. Too few or poorly matched short channels, or short channels that are themselves saturated, cannot guarantee a pure cortical result after regression. Probe pressure should preserve coupling without compressing skin or changing local blood flow. Adhesives, springs, cap loading, material compatibility, and overnight migration all require evaluation.

Sleep installation, synchronization, and recording

Before sleep, setup should verify participant identity, array orientation, optode positions, channel coupling, ambient-light exclusion, raw intensity, synchronization triggers, and remaining storage, with hair preparation and abnormal channels documented. An fNIRS cap can compete for space with EEG electrodes, mastoid references, chin leads, respiratory tubing, or the pillow. The arrangement should preserve diagnostic PSG channels while minimizing hard points, traction, and pressure. Good contact while awake does not guarantee acceptable quality after side sleeping, head turns, or sweating.

PSG, fNIRS, video, and systemic physiological measurements need a traceable common timeline. Matching computer clocks or pressing start at the same time is not evidence of event-level synchronization. Trigger transport, device buffering, sampling clocks, and reconnection can add a fixed offset and overnight drift. The study should preserve original triggers, acquisition times, and alignment methods, verify common events at the beginning, end, and after interruptions, and label intervals that cannot be aligned reliably.

Sleep stage should be scored under the applicable PSG rules and then linked to predefined fNIRS windows. Sleep transitions, brief arousals, apnea, limb movement, turning, and technologist contact can occur close together; an averaged curve cannot by itself assign causation to one of them. Recording oxygen saturation, heart rate, respiration, and position—and, where the question requires it, blood pressure, end-tidal carbon dioxide, or other systemic variables—supports interpretation of hemodynamics. Those variables should not automatically be treated as noise to remove.

Signal quality, confounding, and preprocessing

Quality assessment should begin with raw light intensity and examine saturation, dark noise, signal-to-noise, cardiac components, optode detachment, steps, spikes, and slow drift channel by channel. Software, versions, and parameters should be retained for every transformation from raw intensity to optical density and then HbO and HbR estimates. Channel rejection, motion correction, filtering, baseline choice, path-length factors, short-separation regression, and interpolation all change the result. A plausible processed waveform cannot retrospectively prove that acquisition quality was good.

Heart rate, respiration, blood pressure, carbon dioxide, oxygen saturation, Mayer waves, and scalp blood flow can overlap with brain-origin changes. Respiratory events and arousals during sleep change these variables together, so indiscriminate regression may remove the phenomenon under study. Conversely, a low-pass or band-pass filter does not prove that superficial influence was removed. The protocol should state in advance which systemic measures are confounders, covariates, mediators, or primary outcomes and report the different controls and sensitivity analyses used with and without short-separation measurements.

Turning, facial movement, REM-related traction, periodic limb movement, PAP adjustment, and technologist handling can shift probes relative to the scalp. Different motion-correction methods may preserve, attenuate, or create event-related changes, so video, accelerometry, technologist logs, and source traces should be used together. Rejected channels, intervals, and participants and the governing rules should be reported. If one sleep stage loses more data because of motion or coupling, the remaining observations no longer represent the complete night.

Study design, statistics, and interpretation

The protocol should prespecify primary chromophores, regions or channels, sleep stages, event anchors, baseline, analysis windows, and statistical models. Overnight signals are autocorrelated, and many events from one person are not independent participants. Treating hundreds of events as hundreds of subjects exaggerates precision. Multiple channels, wavelengths, stages, and time points also create multiplicity and call for suitable hierarchical models, correction, permutation methods, or constrained hypotheses, with effect sizes and uncertainty reported.

HbO and HbR provide complementary information, and reporting only one requires justification. A group average can conceal individual responses in opposite directions, while selecting a best channel or most significant interval creates bias. Connectivity and machine-learning findings depend strongly on filtering, systemic-signal treatment, channel selection, leakage between training and test data, and validation splits. Classification accuracy does not prove that a model learned brain-origin sleep physiology or that it transfers across instruments, laboratories, ages, or disorders.

Simultaneous PSG-fNIRS studies demonstrate feasibility for exploring sleep-stage and obstructive-respiratory-event hemodynamics, but many have small samples, limited frontal coverage, and heterogeneous devices and pipelines. A single study cannot define an individual diagnostic threshold, and cerebral hemoglobin changes do not replace AHI, PSG staging, or pulse oximetry. A proposed screening, diagnostic, treatment-decision, or real-time alarm use requires separate validation of clinical performance, human factors, and risk for that exact intended use.

Safety, comfort, and participant protection

Optical safety depends on wavelength, source type, power per source, spot size, duty cycle, the number of simultaneously active sources, probe contact, exposure duration, and fault conditions. Invisible near-infrared light is not inherently risk free, and low average power does not replace an eye and skin exposure assessment under the worst credible combination. LEDs or lamps and lasers follow different safety pathways. Emission state must not be judged by whether a participant can see the light, and servicing should prevent direct viewing of an uncontrolled source.

Practical overnight risks include focal pressure, erythema, heat, allergy, adhesive injury, headache, sleep disruption, cable entanglement, and tripping when leaving bed. Infants, children, people with fragile skin or reduced sensation, and participants unable to communicate discomfort need more conservative fixation and observation. Pain, burning, numbness, persistent redness, or skin injury under an optode calls for stopping and inspection, not tighter cap pressure in pursuit of higher intensity.

When fNIRS and PSG form a medical electrical system, electrical isolation, electromagnetic compatibility, fluids, cleaning, connectors, power, and fault recovery need assessment. Consent should explain the added instrument's purpose, possible discomfort, data use, and non-diagnostic research status. Combining audio-video, hemodynamic, and clinical data increases re-identification risk. Unvalidated individual brain-health scores should not be returned as research results, and incidental findings and post-withdrawal data handling should be specified in the protocol.

Validation, data governance, and reproducibility

Equipment acceptance should test dark noise, source stability, detector linearity and saturation, interchannel crosstalk, timing accuracy, trigger latency, long-term drift, battery or mains behavior, ambient light, known optical phantoms, and the complete array. Coupling and comfort should also be evaluated with the actual pillow, positions, combined EEG cap, and overnight duration. Replacing a source, detector, cap, firmware, analysis software, or synchronization path calls for impact-based revalidation rather than relying on performance from the previous configuration.

Raw intensity, probe geometry, wavelengths, source-detector indices, coordinates, auxiliary physiology, triggers, exclusion markers, and processing code should be retained separately from derived HbO and HbR. SNIRF and NIRS-BIDS can help organize data and metadata, but a successful export does not prove lossless preservation of every vendor field, timeline, and coordinate. Before sharing, files, versions, and de-identification should be verified, with provenance from a result back to the source acquisition and exact processing environment.

Reproducible research reports participants, sleep schedule, instrument architecture, array diagram, source-detector separations, localization, raw quality, exclusions, preprocessing order, systemic-physiology control, statistical model, and software versions. Ongoing maintenance also monitors optode aging, fiber damage, holder deformation, calibration drift, consumable lots, and cleaning effects. The important outcome is not a colorful brain map, but traceable, quality-transparent, reproducible hemodynamic data obtained without materially disrupting sleep and interpreted within the method's limits.