Name, definition and product boundary
A closed-loop smart eye mask is a sleep wearable that combines a light-blocking, periocular form factor with sensing, real-time decision logic and an active output, so that a subsequent output can change in response to a recently measured state. Inputs may come from electroencephalographic or electrooculographic electrodes, movement, contact pressure, ambient light or a pulse waveform. Outputs may include sound, dim light, vibration, temperature adjustment or another non-invasive cue. An individual product ordinarily uses only a subset of these options. The defining feature is not the number of functions but whether measured feedback actually changes the timing, intensity, content or termination of a later stimulus.
A conventional sleep mask only reduces light reaching the eyes. A device that plays audio, fades light or executes a fixed program on a preset schedule uses timed or open-loop control. A device that records sleep data without using those data to alter output during the same night is a monitor. An application, an artificial-intelligence label or a personalized playlist does not by itself create a physiologic closed loop. The term is appropriate only when the input signal, decision rule, delivered output and feedback update form a chain that can be tested.
This entry covers closed-loop systems whose principal wearing form is an eye mask and whose intended role is general sleep support or research. It does not make them equivalent to a brain-computer interface, a standard polysomnographic system or a medical neuromodulation device. If a product uses electrical current, a magnetic field or another form of energy to alter neural activity, or claims to diagnose, prevent, mitigate or treat insomnia or another disease, the risks, evidence and regulatory requirements of an ordinary consumer mask cannot simply be carried over. A product name and a soft appearance do not replace review of intended use, actual output and status in the relevant jurisdiction.
Sensing, decision and output pathway
A complete pathway commonly includes skin-contact or non-contact sensors, an analogue front end, signal-quality assessment, feature extraction, state or event estimation, a control rule, an actuator and an overnight log. EEG can support detection of a specified waveform or estimation of a sleep stage, EOG can reflect eye activity, movement and pressure can indicate posture or displacement, and a pulse waveform can provide beat-related trends. These channels measure potential, motion, pressure or optical change; they do not directly measure deep sleep, recovery or sleep quality.
When confidence is sufficient and safety conditions are met, the controller may play a sound, energize a defined light source, change a vibration pattern, or reduce or stop output after detecting sleep onset, arousal, movement or signal corruption. In a phase-locked system, the combined delay from sampling, filtering, event detection and computation through actuator response changes where a stimulus actually lands. In a stage-driven system, the scoring window and delay in state transitions likewise affect the trigger. A real-time claim therefore needs a stated timescale and acceptable error rather than a single idealized trace.
Closed-loop value also lies in withholding stimulation when contact is poor, motion artefact is present, a wireless link fails, the battery is low, algorithm confidence is insufficient or an output limit has been reached. The design should define stop, degraded and fallback states and should let the wearer interrupt output without first unlocking a phone. If a device continues a fixed program after its sensor disconnects, the interface should make clear that closed-loop operation has ended. Open-loop continuation must not be presented as if physiologic feedback were still controlling it.
Mask structure, fit and signal quality
An eye mask has to provide light exclusion, stable contact and all-night comfort at the same time, and these objectives can conflict. A loose strap permits light leakage, migration and electrode separation; an overly tight strap can create sustained pressure at the brow, bridge of the nose, cheekbone, periocular tissues or behind the ears. Eye cups and pads should avoid direct pressure on the globes and lashes. A rigid controller, connector or battery should not occupy a location that is loaded during side sleeping. Mass, centre of gravity, ventilation, temperature rise and size adjustment are parts of performance, not merely styling details.
Head and facial biopotentials are small, and blinking, facial muscle activity, turning, perspiration, hair, skin oil and electrode impedance can all generate artefact. Speaker vibration, light-source drive, power switching and wireless transmission may also contaminate the input used for triggering. For light, wavelength, luminance or irradiance, pulse width, frequency, closed-eyelid condition and actual distance matter. For sound, sound pressure, spectrum, episode duration and cumulative exposure matter. These quantities should be measured on the finished product under realistic wearing conditions rather than inferred solely from a component specification.
Before first use, the wearer should confirm that the manufacturer permits the intended overnight duration and should practise a short awake trial, manual stop and rapid removal. Before bed, the pad and skin can be checked for cleanliness and dryness, components for secure attachment, battery for sufficient charge, and the program and audio or light limits for correct settings. On waking, the user should inspect pressure marks, skin response, signal gaps, trigger count and unexpected stops. One comfortable night does not establish long-term fit, and an uninterrupted recording does not mean that every segment was suitable for decision-making.
Algorithms, data and interpretation of effects
Sleep stage, slow-wave event, sleep-onset time and arousal probability are algorithmic inferences from limited signals. An evaluation should identify the sensors, sampling and filtering, label definitions, reference record, target population, separation of training and test data, and sensitivity, specificity or agreement for each state. Overall accuracy alone can hide errors in uncommon stages and periods of poor signal. An algorithm should retain an unknown or no-trigger state instead of forcing every segment into a class.
Trigger evaluation has to cover both recognition and delivery: whether the system fired in the correct state, whether the output landed in the target phase or time window, how many false and missed triggers occurred, whether stimulation produced an arousal, and whether delivered dose matched the setting. An all-night composite score, average sleep-onset time or next-day impression cannot replace these pathway measures. When a software update changes staging, thresholds, stimulus content or data-cleaning rules, the version should be recorded and longitudinal comparability reassessed.
Research has shown that selected EEG-guided closed-loop acoustic protocols can increase slow-wave or spindle activity in some healthy samples under specified conditions, but behavioral, memory, sleep-improvement and age-group results have not been consistent. In a small study, closed-loop dim red light delivered through an eye mask also did not produce the same slow-wave power effect as acoustic stimulation. These findings belong to the tested device, timing, dose, sample and endpoint. They do not establish that every audio-light mask is effective, and an EEG change cannot replace validation of insomnia symptoms, daytime function or long-term safety.
Performance evaluation and purchasing acceptance
Procurement information should identify the exact model, hardware and firmware version, sensor locations, available raw and derived data, closed-loop target, stimulus modality, dose range, contraindications, supported population, and functions that remain when the phone or cloud is unavailable. Evidence should correspond to the same construction and algorithm version and report the reference system, sample composition, exclusions, control or sham condition, valid nights, attrition and adverse events. Prototype feasibility, qualification of one component or good laboratory performance does not automatically establish that a production system is effective in the home.
Acceptance testing should examine the sensor, control algorithm and actuator separately and then evaluate their integrated end-to-end behavior. Relevant measures can include valid-signal proportion, event or stage error, end-to-end latency, stimulus-timing error, output ceiling, cumulative dose, arousal rate, false triggers, communication loss, low-battery behavior, over-temperature protection, manual stop, fallback modes and log completeness. Foreseeable conditions such as side sleeping, turning, perspiration, a weak network, an offline phone and all-night battery demand belong in the test plan.
A home trial should also record burden at sleep onset, removal during the night, morning eye and skin comfort, headache or dizziness, device temperature, noise and light leakage, effects on a bed partner, cleaning time and multi-night completion. When comparing products, preference should be given to systems with transparent parameters, output limits, detectable faults and exportable quality flags rather than to the largest number of stimulation programs, a proprietary sleep score or unexplained AI personalization.
Cleaning, power and software maintenance
The mask directly contacts the forehead, periocular skin and hair, so sweat, sebum, cosmetics and skin-care products can affect hygiene, materials and sensor contact. Batteries, controllers, audio-light modules and cables that are designed to be removed should be detached before the textile and pad are cleaned and completely dried according to the instructions. Wipeable, sweat-resistant or assigned a water-resistance rating does not mean suitable for immersion, machine washing, tumble drying or disinfection. A damaged electrode, split pad, exposed wire, unusual odour, deformation or altered performance after cleaning calls for inspection and discontinuation until resolved.
Charging should occur off the head with a dry interface. A damaged, swollen, unusually hot or unverified battery or charger should not be used. Charging leads, straps and small detachable parts need to be kept away from infants and young children to prevent entanglement, ingestion and choking. During use, the battery compartment should not be trapped between the pillow and the head or combined with an additional heating product. Endurance testing should cover the most demanding simultaneous sensing, computation, wireless and stimulation mode, not standby time alone.
The application, phone and cloud service are components of the closed-loop system. Users should be able to learn which raw signals are collected, which inferences are uploaded, how long data are retained, who receives them, and how they can be exported and deleted. Purchasers should also check account permissions, encryption, firmware signing, the update mechanism, vulnerability-support period and end-of-service plan. Versions and settings should be retained across updates, with a new baseline established when necessary. If secure updating is no longer possible or a cloud service closes, the device should not continue automated stimulation in an indeterminate state.
Safety, abnormal events and medical boundary
The device should not be self-applied over damaged, infected or markedly inflamed periocular skin, after recent eye or facial surgery, or when a clinician has advised avoiding occlusion, light, heat, vibration or pressure. Eye pain or pressure, altered vision, persistent headache, nausea, dizziness, numbness, red or itchy skin, abnormal heating, harsh sound, repeated startling or inability to stop immediately calls for mask removal and help proportionate to severity. Recurrent signal loss should not be corrected simply by tightening the strap.
A person with photosensitive epilepsy or another seizure history, migraine-related light sensitivity, serious eye disease, auditory hypersensitivity, an implanted electronic device, or treatment that affects sleep or neural activity should first check the specific output, contraindications and professional advice. Children, cognitively limited users and anyone unable to remove the mask independently also face supervision, entanglement, small-part and misuse risks. Any electrical stimulation, magnetic stimulation or high-energy optical output requires a separate assessment of the actual finished product; a soft eye-mask form does not make such technology low risk.
Persistent difficulty initiating sleep, marked daytime impairment, repeated choking awakenings, loud snoring with breathing pauses, unusual nocturnal behavior or another continuing symptom warrants professional assessment. Consumer mask estimates of stages, slow waves, recovery or stress can neither diagnose nor exclude a sleep disorder and should not direct self-discontinuation or adjustment of medication or replace established treatment. Selection should consider intended use, independent validation, controllability of output, wearing safety, fault fallback, privacy, cleanability and support rather than treating closed loop itself as a guarantee of benefit.