The artifact atlas
Identify by eye every common physiological and non-physiological artifact, then prove it on a 20-item drill.
Prerequisites: L0.4 · Reading raw traces
1 claim on this page is unverified. TODO(confirm) marks a specific statement the author has not yet checked against a
primary source. Everything else on this page has been reviewed. Treat a marked claim as
provisional and go to the cited source rather than quoting the sentence.
Objectives
- Identify by eye blinks, horizontal saccades, EMG, ECG and pulse, drift, electrode pops, cable and motion artifacts, line noise, bridging, flat channels and drowsiness-related slowing
- Describe each artifact by channels affected, morphology, duration, frequency content and polarity relative to the reference
- Score at least 85% on a 20-item labelling drill
Why this matters
Most of what is large in a raw EEG trace is not brain. Blinks are ten times bigger than an ERP; a clenched jaw covers the whole gamma band; one loose electrode can dominate an average reference. Every cleaning method in Level 2 assumes you already know what these things look like, because the methods are only as good as your check on their output. This lesson is the vocabulary; the drill at the end is the test.
Concepts
Each artifact below is described on the same five axes — channels affected · morphology · duration · frequency content · polarity relative to the reference — because those are the five questions that separate one artifact from another and from brain activity. Polarity statements assume a referential montage with a reference that is far from the artifact source (mastoids, average); with a different reference the sign can flip, which is itself a useful check.
Blinks
- Channels: frontal-polar (Fp1, Fp2) largest, falling off steeply toward F3/F4 and Fz, essentially absent occipitally.
- Morphology: a smooth, rounded, monophasic deflection, roughly symmetric between the two sides.
- Duration: about 200–400 ms.
- Frequency content: low; most energy below a few Hz.
- Polarity: positive at Fp1/Fp2 with respect to a posterior or mastoid reference, because the eyeball is a dipole with the cornea positive and the eyelid sliding over it changes the potential seen from the forehead. Amplitude at Fp is typically well over 100 µV (TODO(confirm) typical range).
- Distinguish from: frontal slow waves (which are not confined to Fp and do not come in the stereotyped shape) and from vertical saccades (sharper onset).
Horizontal saccades
- Channels: lateral frontal (F7, F8; also Fp1/Fp2 and AF sites), with opposite polarity on the two sides.
- Morphology: a step — a rapid onset followed by a plateau that lasts until the next eye movement (a “square wave”).
- Duration: onset within tens of ms; plateau from hundreds of ms to seconds.
- Frequency content: very low (a step is mostly DC and low frequencies), with a brief broadband edge at the onset.
- Polarity: the side the eyes move toward goes positive (the cornea approaches that electrode), the other side goes negative; F7 and F8 are mirror images. Reading is a train of small saccades in one direction with a large return saccade.
- Distinguish from: blinks (same polarity on both sides, no plateau) and from lateral EMG (fast, spiky, no step).
EMG (muscle)
- Channels: wherever a muscle is under the electrode — temporal and lateral frontal (T7, T8, F7, F8, FT sites) for jaw clenching and the temporalis; occipital and posterior temporal for neck tension; frontal for the forehead; any site during swallowing or talking.
- Morphology: dense, spiky, irregular fast activity that thickens the trace; it starts and stops with the muscle contraction.
- Duration: as long as the contraction — a swallow (a second or two), a sustained clench (tens of seconds), a whole tense recording.
- Frequency content: broadband, from about 20 Hz to well above the EEG range, with no single peak; it raises the whole high-frequency floor of the spectrum.
- Polarity: none meaningful; it is high-frequency noise superimposed on the trace.
- Distinguish from: beta (lower amplitude, rhythmic, does not track jaw or neck tension) and from line noise (a single sharp spectral line, sinusoidal). The pitfall of reporting EMG as gamma is previewed below and returns in L7.4 and L7.5.
ECG and pulse
- ECG (electrical): small, sharp, regular spikes — the QRS complex — at the heart rate (roughly once per second), widespread across channels with a consistent shape, most visible when the reference is near the heart’s field (an ear or mastoid reference) and in low-amplitude segments. Frequency content: sharp transients, so broadband but brief.
- Pulse (mechanical): a slow, rounded, rhythmic wave at the heart rate confined to one electrode that sits over a scalp artery (often temporal); it is the electrode moving with the pulse, not an electrical signal from the heart. Frequency content: about 1 Hz and its harmonics.
- Distinguish from: brain rhythms by the metronome regularity and the match to heart rate; ECG from pulse by spike vs wave and by widespread vs single-channel.
Sweat and skin-potential drift
- Channels: one or a few electrodes, or all channels at once if the whole scalp is sweating.
- Morphology: slow, large, wandering baseline shifts; the trace tilts and drifts off the slot.
- Duration: seconds to minutes.
- Frequency content: below about 0.5 Hz.
- Polarity: either; sweat changes the electrode–skin interface potential.
- Distinguish from: slow eye movements (frontal and mirror-symmetric) and from delta activity (rhythmic, physiological amplitude). On a recording with no hardware high-pass —
ds-eegbciwas recorded with none — drift is plainly visible and must not be mistaken for a physiological slow wave.
Electrode pops
- Channels: a single channel.
- Morphology: an abrupt step or spike followed by an exponential return toward baseline, sometimes repeated as the contact flickers.
- Duration: onset in one or a few samples; recovery over tens of ms to seconds.
- Frequency content: the sharp edge is broadband; the recovery is slow.
- Polarity: either.
- Distinguish from: everything else by its confinement to one channel and its edge sharpness. Pops are also the canonical demonstration of filter ringing in L1.5, because a filter turns the sharp edge into oscillations on both sides.
Cable and motion artifacts
- Channels: several channels at once, often a group that shares a cable bundle, or all channels during head movement.
- Morphology: irregular, large, non-stereotyped deflections that coincide with movement; may include pops and drift.
- Duration: matches the movement.
- Frequency content: mostly low, with broadband edges.
- Distinguish from: physiological artifacts by their irregularity and their coincidence with events you can see (subject shifts, cable tugged, cap adjusted).
Line noise
- Channels: all channels if the ground or common-mode rejection is poor; single channels with high impedance otherwise.
- Morphology: a constant sinusoid at the mains frequency; at 10 s per page it appears as a uniform thickening or “fuzz” on the trace, not as individual cycles.
- Duration: continuous, though it can wax and wane as equipment switches on and off.
- Frequency content: a sharp line at 50 Hz or 60 Hz (by country) plus harmonics. In
ds-eegbci(60 Hz mains, 160 Hz sampling) only the 60 Hz fundamental is below the 80 Hz Nyquist frequency, which L1.6 returns to. - Distinguish from: EMG (broadband, irregular) by looking at the spectrum: a line, not a floor.
Electrode bridging
- Channels: two (or more) neighbouring electrodes.
- Morphology: the two traces are near-identical, including their noise; their difference is close to zero.
- Duration: the whole recording (a gel bridge between electrodes, usually from over-application of gel), or from the moment of the bridge.
- Frequency content: whatever the shared signal has; the tell is the identity, not the frequency.
- Distinguish from: genuinely correlated neighbours by looking at the difference channel or a correlation matrix. Bridged channels look fine in isolation, interpolation “works”, and the montage has silently lost a channel — the pitfall is previewed in L0.2 and treated in L2.2.
Flat and dead channels
- Channels: one channel (or the recording reference, which is flat by construction in some systems).
- Morphology: a straight line, or a line with only line noise on it.
- Duration: whole recording, or from the moment of disconnection.
- Distinguish from: a channel that is merely quiet by checking its variance against neighbours; a truly zero-variance channel is either dead, disconnected, or the reference.
Drowsiness-related slowing
Drowsiness is a state, not an artifact, but it lands in the atlas because it is the commonest reason a “resting” file is not what its label says. Look for the pattern from L0.4: alpha fragments and drops out, the background slows toward theta, slow roving eye movements appear frontally, and vertex sharp waves may appear at Cz. Channels: posterior for the alpha dropout, frontal for the eye movements, Cz for vertex waves. Duration: tens of seconds to minutes.
The same artifacts in a spectrum and a topography
The notebook reproduces each artifact type as a figure with its PSD and topography. The point of that exercise is that each artifact has a spectral and a spatial signature as well as a temporal one: blinks and saccades are frontal and low-frequency; EMG is lateral and broadband; line noise is a single line; pops and drift are single-channel and low-frequency. Level 2 will lean on these signatures when it classifies ICA components — a component’s topography, time course and spectrum are the same three views.
A label is only useful with a reason. When you tag a segment “EMG”, say which channels, what the waveform looked like, and what in the spectrum backs it up. The drill asks for the label, but the feedback shows the tell-tale features so that you learn the reasons; a correct label with the wrong reason will fail on the next file.
Explore
The item bank holds at least 60 labelled 5-s segments with 8 or more channels each. Items are drawn from ds-eegbci (no hardware filters, so drift, pops and line noise are real) and ds-lemon raw (which has a VEOG channel for blink and saccade items). Labels are label_source: algorithmic until the author reviews every item (§4.5); labels that Phase 0 data cannot supply (for example ECG and bridging) may be absent from the bank, and the drill says so.
Practice
The artifact atlas: trace, spectrum and topography for each artifact type nb-0-5-artifact-atlas
Downloads from ds-eegbci.
The notebook reproduces each artifact type as a figure with its PSD and topography, so that the temporal, spectral and spatial signatures are side by side.
Exercises
Exercise ex-0-5-artifact-drill
Drill (widget-graded)Complete a 20-item drill in Spot the Artifact and score at least 85% (17 of 20).
Graded by the w-spot-the-artifact above: complete a session there and the score is recorded here. Pass mark 85%.
No session recorded yet.
Exercise ex-0-5-ambiguous-item
Free responsePick one item you got wrong or found ambiguous. Which two labels were you choosing between, and which of the five axes (channels, morphology, duration, frequency content, polarity) would have settled it?
Pitfalls
EMG reported as gamma
- Symptom
- Broadband high-frequency power over temporal/frontal sites tied to jaw or neck.
- Cause
Scalp muscles (temporalis, frontalis, masseter, the neck extensors) produce electrical activity that is large compared with cortical gamma, broadband from roughly 20 Hz upward with no single peak, and sits directly under the electrodes that “show the effect”. Its amplitude is modulated by anything that changes muscle tone — and many experimental manipulations do. The spectrum of EMG overlaps the…
- Detect
- Topography: cortical gamma from a focal source is rarely largest at the edge of the montage; EMG is. Look at T7/T8, F7/F8, FT sites and the occipital rim. - Spectral shape: EMG raises the floor across a wide range with no peak; a genuine oscillation has a peak above the aperiodic background (L1.7, L4.6). - Time course: EMG turns on and off with muscle events (swallows, jaw movement, blinks with…
- Fix
- Instruct and monitor: relaxed jaw, no talking, breaks; record a facial EMG channel where high frequencies matter. - Clean before measuring: ICA with muscle-component removal, or rejection of segments with high-frequency power above a threshold (L2.5, L2.6); state what was removed. - Report a peak, not a band: use spectral parameterization to show a gamma peak above the aperiodic background befo…
Eye movements contaminate lateralized effects
- Symptom
- N2pc / lateralized gamma tracks HEOG.
- Cause
The eye is a dipole with the cornea positive. A horizontal saccade toward a lateral target moves that dipole so that electrodes on the side the eye moves toward go positive and those on the other side go negative (L0.5). Because a lateralized ERP is computed as a difference between hemispheres, an eye movement that is systematically toward the target produces a difference wave with a clean latera…
- Detect
- Plot the HEOG (or F7 minus F8) averaged the same way as the effect (contralateral minus ipsilateral): any residual lateralized deflection in that average is a confound. - Compare the time course and topography of the effect with the HEOG average; an eye-movement confound is largest at the most lateral frontal sites and falls off toward the midline. - Split trials by whether a saccade was detect…
- Fix
- Record EOG (or eye-track) and reject trials with horizontal eye movements above a stated threshold, applied identically across conditions (L2.5). - Use a step-function artifact detector on HEOG rather than a simple amplitude threshold, because small saccades are steps, not spikes. - Correct with ICA or EOG regression and verify on the residual HEOG average that no lateralized deflection remains…
In other tools
In other toolsEEGLAB · FieldTrip — names only
The equivalents of what this lesson does, for a reader who works in another toolbox. Function names only: their own documentation is the place to learn how to call them.
EEGLAB
pop_eegplotEEGLABpop_rejcontEEGLAB
FieldTrip
ft_databrowserFieldTripft_artifact_zvalueFieldTrip
Names checked 2026-09-18 against EEGLAB 2026.0.0 (plugins at the versions in EEGLAB’s own plugin list) and FieldTrip 20251218.
Reading
- Luck (2014). An Introduction to the Event-Related Potential Technique, 2nd ed.. unverified
- Kane et al. (2017). Revised glossary of clinical EEG terms. unverified