World Sleep Industry ENCYCLOPEDIA

sleep EEG systems

2.2.3

Definition, system composition, and limits of capability

A sleep electroencephalography system is a medical electrical system centered on acquiring, displaying, preserving, and analyzing scalp EEG across sleep and wake transitions. A typical system includes electrodes and leads, a patient input unit, differential amplifiers, isolation and analog-to-digital circuitry, synchronized clocks and storage, an acquisition workstation, and review and scoring software. It may also incorporate synchronized video, audio, an event button, or remote data services. The EEG equipment may be a subsystem of polysomnography, a stand-alone sleep EEG platform, an ambulatory recorder, or part of long-term video EEG. The shared name does not make these configurations clinically interchangeable.

Within polysomnography, EEG is ordinarily interpreted together with electro-oculography and chin electromyography to stage sleep and identify arousals, while respiratory, oximetry, cardiac, and limb signals supply other parts of the sleep study. When the question is epilepsy or another episodic nocturnal disorder, the task changes to detecting, localizing, and characterizing interictal or ictal activity and relating it to behavior. That task may require wider scalp coverage, additional electrodes, synchronized video, and a longer observation period. A validated reduced-frontal montage can support a defined staging purpose, but the mere presence of EEG does not make it a complete PSG or a diagnostic clinical EEG examination.

Capability must therefore be defined by intended use, patient population, setting, operator, actual electrode set, accessories, and software version. A consumer brainwave headband, anesthesia-depth index, amplitude-integrated EEG display, stand-alone spectral map, or neurofeedback device is not automatically a sleep EEG system in this clinical sense. Changing a headband, input module, or software license on the same host may materially change its claims. Orders, procurement specifications, and reports should state whether the examination is sleep staging, routine EEG incorporating sleep, ambulatory EEG, or synchronized video EEG instead of allowing a broad product label to substitute for a protocol.

Electrodes, montages, and the digital signal chain

Scalp electrodes should be placed according to the applicable standardized position nomenclature and the individual head measurement, with the reference, ground, supplemental electrodes, laterality, and any forced displacement documented. Cup, adhesive patch, needle, dry, and textile electrodes differ in contact mechanism, securement duration, and suitable setting; acceptable performance for one cannot be presumed for another. Hair, skin injury, allergy, skull defects, and implants should be considered before preparation. A numeric impedance within range does not by itself exclude electrode bridging, misplacement, reversal, or a poor mechanical connection that becomes intermittent during sleep.

Each input passes through differential amplification, common-mode rejection, patient isolation, anti-aliasing, sampling, quantization, digital filtering, and storage before it appears on a screen. Sampling rate, effective resolution, input range, noise, bandwidth, saturation recovery, and channel crosstalk can determine whether slow activity, low-amplitude fast activity, or sharp transients remain visible. Low-frequency, high-frequency, and notch filters should not be increased merely to make a tracing look clean, because they can attenuate slow waves or spikes and alter phase and morphology. Acquisition settings and sufficiently raw digital data must be retained so a qualified reviewer can change the display later without inventing information that was never recorded.

A montage is a way to display voltage differences, not an additional measurement source. Digital systems should retain inputs that can be re-referenced and permit approved bipolar, referential, or other displays while making clear that failure of a shared reference can contaminate many derivations. A reduced forehead device that has been validated for a specific staging task can lower application burden, but it necessarily sacrifices spatial coverage and cannot algorithmically reconstruct unrecorded temporal or occipital activity. Bad-channel substitution, interpolation, or automated reconstruction must be visibly labeled, and the underlying segment must remain available for review.

Setup, calibration, and artifact control

Before acquisition, staff should verify the patient and order, relevant medication and sleep-deprivation status, device and accessory identifiers, electrode map, clock, sampling and storage settings, power status, and the scope of consent for video or audio. Eye opening and closure, eye movements, jaw contraction, swallowing, breathing maneuvers, or other protocol-appropriate biocalibrations help confirm that EEG, EOG, EMG, ECG, and event markers have the correct direction and timing. Seeing a moving line on every channel is not a sufficient check. If the camera, infrared system, microphone, bedside event button, and external PSG host use separate clocks, their common time base and recovery after disconnection need explicit verification.

Common sleep-recording artifacts include sweat drift, movement and cable tug, electrode pop, intermittent contact, 50- or 60-Hz interference, ECG or pulse contamination, eye movement, frontalis and masseter activity, ventilator or humidifier interference, and reference-electrode failure. Dry electrodes and headbands can also be affected by contact pressure, pillow friction, hair, strap relaxation, and side-sleeping compression. Artifact may resemble a spike, arousal, or high-frequency cerebral activity. An amplitude threshold alone cannot reliably distinguish it; spatial distribution, other physiological channels, video, behavior, and temporal context all contribute to interpretation.

The system may flag poor contact, clipping, a flat line, motion, or mains interference, but an automated signal-quality score remains an aid. Attended protocols should define surveillance frequency, authority to repair electrodes, and methods that minimize awakening. Unattended protocols need robust application feedback, suitable backup derivations, and post-study reporting of valid time for each channel. A bad channel must not be silently smoothed or copied from a neighbor. If a shared reference, input unit, clock, or power failure disables several essential channels, the study should be judged technically inadequate under a predefined rule rather than forced to completion from the residual hypnogram.

Matching the protocol to the clinical question

Sleep can activate some EEG abnormalities, so a routine EEG may deliberately include natural sleep, sleep deprivation, or an appropriate induction method. That examination differs from overnight PSG, whose principal outputs concern sleep architecture and the temporal relationship of sleep to respiratory and other events. Suspected obstructive sleep apnea requires airflow, respiratory effort, and oxygenation information; EEG alone cannot classify apneas. Conversely, the limited EEG montage of a standard PSG may be inadequate for episodic impaired awareness, unusual motor activity, or other nocturnal behavior. The clinical question should select the examination and montage, rather than allowing available hardware to constrain the diagnosis by default.

For suspected nocturnal epilepsy, nonepileptic events, parasomnias, or REM sleep behavior disorder, the protocol may need expanded scalp coverage, synchronized video and audio, ECG, selected EMG, an event button, and an observation window long enough to capture representative behavior. Camera placement should cover the patient and bed environment while respecting privacy and preserving the onset and offset of movement. Witness accounts, an event diary, and medication information belong on the same timeline. Failure to capture a typical event, or absence of an abnormality in reduced leads, does not exclude epilepsy; correct automated sleep staging also does not establish the cause of an unusual movement.

Infants, children, older adults, and people with neurologic disorders have different normal backgrounds, sleep organization, movement patterns, and artifact profiles. Electrode size, securement materials, and interpretation rules may need population-specific adaptation. A ward, home, and sleep laboratory also differ in supervision, electrical and network conditions, noise, and emergency support. A person who may require immediate intervention, repeatedly remove leads, fall when mobilizing, experience prolonged seizures, or be unable to cooperate safely should be studied in a setting with the corresponding capabilities. Protocol choice and escalation must follow applicable professional qualifications, labeling, and local clinical rules.

Sleep scoring, EEG interpretation, and automation

Sleep staging ordinarily assigns consecutive fixed-duration epochs as wake, N1, N2, N3, or REM using EEG together with EOG and chin EMG, then derives total sleep time, efficiency, latency, and stage proportions. EEG supplies features such as the posterior dominant rhythm, vertex waves, sleep spindles, K complexes, and slow waves, but boundaries such as REM versus quiet wake often depend heavily on eye movement and muscle tone. If required channels are missing or dominated by artifact, the segment should be marked unscorable or its uncertainty made explicit. Body motion or a prior probability should not be used to fabricate a complete night.

Sleep stages, arousals, epileptiform discharges, and clinical seizures are different levels of annotation. A sharp waveform may be a physiologic transient, muscle activity, electrode artifact, or a true cerebral discharge, and requires assessment of field, morphology, background, evolution, and video context. Spectral power, compressed EEG indices, and color trends are useful navigation tools but do not replace the source waveform. The report should distinguish machine suggestions, technologist edits, and physician conclusions and should state recording limitations, actual electrodes, medication and sleep obtained, and whether the patient's typical event was captured.

Automated staging and event-detection models need validation with the intended electrodes, device, age group, population, environment, and artifact burden. High performance in a training dataset does not guarantee performance in the home, in children, in neurologic disease, or after a software update; an overall agreement statistic can also conceal errors in uncommon stages or clinically important events. The system should preserve algorithm and parameter versions, confidence or rejection information, human edits, and an audit trail so the result can be reproduced. Automation can reduce workload and improve consistency, but it cannot independently diagnose a sleep disorder, epilepsy, or a psychiatric or neurologic condition.

Safety, data governance, and maintenance

Patient-connected inputs must meet the applicable requirements for electrical isolation, leakage current, electromagnetic compatibility, and operation with defibrillation or other concurrently used equipment. Accessories and the acquisition unit must be used in an approved combination. Leads should be routed away from the neck and walking path and secured to reduce entanglement, falls, and connector strain. Conductive paste, abrasive preparation, adhesives, and sustained pressure can cause irritation or injury and require controlled quantity and contact time. Flammable or volatile materials require appropriate ventilation, ignition control, and institutional procedures. Ordinary EEG electrodes must not enter an MRI environment unless the complete system is specifically designed for it.

Most sleep EEG systems are record-and-review tools and do not inherently provide continuous real-time alarming. A visible waveform, remote upload, or algorithmic detection does not mean that a person is continuously watching, nor does it guarantee timely notification of a seizure, hypoxemia, or arrhythmia. The device and service model must define surveillance responsibility, notification latency, network-loss behavior, nighttime response, and escalation. Without an appropriate intended use and validation, the system must not be relied upon as an ICU, emergency, or injury-prevention monitor. A prolonged seizure, breathing difficulty, or other emergency requires the relevant emergency response rather than waiting for the final report.

Raw EEG, video, audio, identity data, annotations, and reports are highly sensitive. Acquisition, transfer, remote access, export, teaching use, and archiving need least-privilege access, identity checks, encryption, integrity verification, audit logs, retention rules, and secure deletion. EDF and other exports must be validated together with montage labels, units, sampling and filter metadata, events, and time-zone information. Maintenance includes input and calibration checks, cable and electrode inspection, battery health, clock synchronization, storage capacity, validated cleaning, software and security updates, and continued readability of prior studies. Updates require regression testing of display, measurement, export, and algorithm behavior.

Acceptance testing, failure criteria, and development

Acceptance testing should exercise the entire path from electrode input to final report. Bench evaluation includes input noise, common-mode rejection, gain and linearity, frequency response, sampling and quantization, crosstalk, clipping and recovery, calibration, event latency, clock drift, restart after power loss, storage, and export. An overnight simulation should also test progressively deteriorating contact, movement, switching of power or network state, video synchronization, and operation near storage limits. A startup self-test or one attractive waveform cannot establish that the system is fit for its intended protocol.

Study acceptability should be specified in advance according to the clinical question, including common valid time for essential derivations, coverage of relevant sleep or target events, video visibility, and completeness of supporting channels. Shared-reference failure, mislabeled electrodes, time misalignment, saturation during the decisive interval, or destructive filtering can make a long record uninterpretable, whereas limited local artifact may be retained with clear annotation. A technical failure record should identify the root cause, conclusions affected, and remedy. Repetition, a wider montage, attended acquisition, or escalation to long-term video EEG may be necessary; algorithmic confidence cannot replace missing evidence.

Development directions include more comfortable dry and around-the-ear electrodes, wireless patient units, multi-night home EEG, automated quality control, individualized staging, and remote specialist review. Lower application burden may enable longer observation in a natural environment, but fewer channels, variable contact, battery limits, network dependence, and model drift introduce new uncertainty. New systems should be judged by valid signal duration, sensitivity to target events, false alerts, human revision, repeat-study burden, and effect on clinical decisions, not only by correlation coefficients or comfort scores. Every new feature still requires an explicit intended-use boundary and a traceable path back to the original evidence.