Wiki

Wiki: Basics of EEG

Basic knowledge to prevent EEG from becoming a magical mind-reading technique

Mind Uploading Research Project

Public Page Updated: 2026-04-01 Beginner guide

How to use this page

Read this first to avoid getting lost

This page is a wiki that explains from the basics what EEG measures. Beyond complicated formulas, the goal is to understand ``what kind of signals are mixed together and where'' and ``why preprocessing and QC are important.''

  • EEG is a mixed electrical signal observed at the scalp.
  • While strong against temporal changes, it has limitations in spatial identification and deep estimation.
  • Reference system, electrode layout, device chain, and protocol are part of the observation model rather than background implementation details.
  • Being able to predict something with your scalp is different from being able to uniquely determine the source in your brain.
  • EEG source-imaging progress is not one ladder; field visibility, forward-model / conductivity burden, solver-family uncertainty, and validation class remain separate.
  • Source-space connectivity and directed connectivity are even stronger claims than source localization and need separate stopping rules.
  • Same-session multimodal EEG acquisition is not self-validating; synchronized streams, shared cross-modal components, and target biological variables remain different achievements.
  • Wearable OPM-MEG is not a drop-in naturalistic readout; shielding class, field control, calibration / coregistration, anatomy route, crosstalk, and task regime still shape the claim ceiling.
  • A shared EEG-fMRI or EEG-PET-MRI factor can still mix neural, autonomic, and vascular contributions.
  • QC and pretreatment records greatly determine the reliability and ceiling of the result.
Helpful next pages
Introduction to EEG Data & Bench Hands-on
Best for
People learning EEG for the first time, people who want to create a premise for public data and introductory pages
Reading time
10-15 minutes
Accuracy note
We highlight the limitations of EEG so as not to overestimate it, but that doesn't mean it's useless.

Relatively clear at this stage

What we know now

  • EEG is good at looking at time changes in milliseconds.
  • The observed signal is the result of a mixture of many activities, and interpretation requires assumptions.
  • Including individualized MRI and external references improves source imaging, but uncertainty remains for deep and weak sources.
  • A stronger EEG source-imaging paper can improve detectability, conductivity calibration, inverse-family uncertainty reporting, or direct validation without solving the others.
  • Reference choice, recording setup, and channel layout can materially change ERP, connectivity, and decoding conclusions.
  • Artifact cleanup does not by itself solve source leakage or turn directed connectivity into causal proof.
  • Adding fMRI, PET, or other modalities can strengthen one audit layer while still leaving fusion validity, shared-factor specificity, and bundle robustness unresolved.
  • Wearable OPM-MEG strengthens movement-tolerant macro electrophysiology only under named shielding, field-control, calibration, anatomy, crosstalk, and task-regime conditions.
  • Public data provides plenty of practice with preprocessing and baseline comparisons.

Still unresolved beyond this point

What we still do not know

  • It remains unresolved whether non-invasive EEG alone is sufficient to reconstruct detailed causal structures within the brain.
  • The choice of preprocessing does not uniformly change the conclusion for each task.
  • It is not yet fixed which external benchmark will be the standard validation set for source imaging.
  • How far wearable OPM-MEG generalizes beyond shielded proof-of-concept and narrow task regimes remains unresolved.

Learn the basics

Check the basics in the wiki

What the wiki is for

The wiki is a learning aid. For the project's official current synthesis, success criteria, and operating rules, always return to the public pages.

What does the EEG see?

EEG is a method that measures electrical potential differences using electrodes placed on the scalp. In other words, we are not looking directly into the brain with a camera, but are reading the mixed signals that are transmitted to the outside as a result of many overlapping activities.

Switch to stop on this page first

With EEG, it is different to observe scalp signals, conditionally estimate brain sources, estimate interactions, and uniquely identify internal states. If these are confused, beginners will read "seen" and "estimated" as if they were the same thing.

2026-03 correction for the beginner route

The older beginner route on this site stopped at "EEG is mixed and source imaging is hard." That was too weak. For EEG, measurement condition itself matters: reference system, electrode layout, device chain, and protocol can change what the scalp signal even means. It was also necessary to say more clearly that a connectivity map or directed graph is a stronger claim than a sensor trace or even a source estimate.

Adding another modality does not make EEG a solved state meter

The next beginner shortcut to stop is the word multimodal. Kothe et al. (2025) describe LSL as synchronization infrastructure rather than device-side delay truth, Vafaii et al. (2024) show that simultaneous multimodal recordings retain both common and divergent organization, and Chen et al. (2025) show that simultaneous EEG-PET-MRI can contain coupled global dynamics together with modality-specific structure. Therefore, even when EEG is combined with other stacks, the site still separates synchronized acquisition, shared statistical structure, and one externally calibrated biological variable rather than treating them as one achievement.

2026-04-01 correction: wearable OPM-MEG is not a portable free pass

The beginner route still had one coarse object left: wearable MEG. The current primary literature does not support treating OPM-MEG as if movement tolerance automatically removed shielding, field-control, anatomy, and calibration burden. Boto et al. (2018) established wearable feasibility but also showed saturation risk without background-field control. Rea et al. (2021) and Mellor et al. (2022) show that precision field modeling and nulling are part of the route, Holmes et al. (2025) show that lightly shielded operation still depends on active compensation plus tSSS, Rhodes et al. (2025) show that pseudo-MRI is useful but does not replace individual MRI as the gold standard, Wu et al. (2025) show that crosstalk remains an array-level burden, and Spedden et al. (2025) show whole-body stepping feasibility in only three healthy adults under a narrow sensorimotor beta task. Therefore, even on this beginner page, the minimum safe question is whether a paper strengthens field control, lighter-shield deployment, MRI-light anatomy substitution, array engineering, or only a narrow task proof-of-concept. Without that split, wearable OPM-MEG is too coarse an object.

What EEG is good at

What I'm good at Reason
Seeing changes over time Because EEG can capture signals in milliseconds, it is easy to track when changes occur.
State transition and event detection It is suitable for observing conditions that change over time, such as sleep stages and seizure events.
Reproduction practice with public data PhysioNet has standard data and it is easy to start practicing L0.

Things that EEG is bad at

Things I'm not good at Why is it difficult
Accurately stating "somewhere in the brain" This is because the signal becomes blurred while passing through the skull and scalp, and the inverse problem cannot be solved uniquely.
Knowing the deep structure in detail Activities far from the scalp and weak signals are difficult to observe.
Making a strong claim of identity using EEG alone EEG is an important clue, but it alone cannot confirm memories, values, or causal continuity.
Treating a connectivity map or directed graph as discovered causal wiring Reference choice, sensor mixing, source leakage, parcellation, and missing external validation can all change the network result even after the waveform looks clean.

Observation, estimation, and identification are different

stage What can be said with EEG Things I can't say yet
Observation The mixed potential on the scalp can be taken on the ms scale. It is good at tracking state transitions and event times. We cannot say that we have directly seen which deep source is the sole cause, including cell types and neuronal modifications.
Conditional estimation Including individual MRI, electrode coordinates, and forward model will improve the estimation of near-cortical and some deep activities. Being able to detect when the conditions are severe is different from being able to restore uniqueness in general.
Network / directed-connectivity estimate With source modeling, parcellation, and explicit metrics, one can estimate conditional interaction structure more strongly than at pure sensor level. That still does not prove leak-free inter-areal coupling or causal direction. Connectivity and directed connectivity need their own validation and abstention rules.
Identification External criteria such as intracranial stimulation, simultaneous SEEG/ECoG, phantoms, and postoperative outcomes allow for error auditing. Without an external standard, it is impossible to say ``I have found the source'' or ``I have achieved a sufficient state for WBE.''
2026-03 actual measurement evidence

Seeber et al. (2019) demonstrated detectability of subcortical signals with 256ch scalp EEG and simultaneous DBS recordings, but did not claim general unique reconstruction. Unnwongse et al. (2023) reported that localization error depends on cranial conductivity and source depth in direct validation for intracranial stimulation, and Hao et al. (2025) showed that source power and source depth strongly influence error in 29 cases of simultaneous HD-EEG/SEEG. Therefore, the correct way to read it is ``partially auditable if the conditions are strictly fixed'', not ``the brain source can be uniquely read using EEG alone''.

2026-03-30 re-audit: better EEG source imaging is still four different advances

This beginner page still needed one more correction. Current primary literature does not support reading all EEG source-imaging progress as one continuous ladder. Seeber et al. (2019) strengthen field-formation visibility for a specific subcortical regime, Vorwerk et al. (2024) and Vorwerk et al. (2026) show that tissue and skull conductivity assumptions still move the result materially, Luria et al. (2024), Tong et al. (2025), and Feng et al. (2025) strengthen uncertainty exposure inside a stated inverse family, and Pascarella et al. (2023), Unnwongse et al. (2023), and Hao et al. (2025) validate different source regimes and error objects rather than one universal source-recovery claim. Therefore, on this site, a "better ESI result" is now read through four floors rather than as one progress bar.

Four-floor split for EEG source imaging What got stronger Representative primary papers What it still does not buy
1. Field-formation visibility Whether a target source class reaches the sensors under a specific depth, orientation, extent, and montage regime. Seeber et al. (2019) It does not by itself fix conductivity sensitivity, inverse-family spread, or general source recovery.
2. Forward-model / conductivity burden How much skull or tissue conductivity and geometry assumptions still move localization, depth, or amplitude. Vorwerk et al. (2024); Vorwerk et al. (2026) Reducing conductivity-driven spread does not by itself prove that the solver family or validation regime is sufficient.
3. Solver-family uncertainty How clearly the inverse family reports posterior width, alternative configurations, debiased intervals, or uncertainty maps. Luria et al. (2024); Tong et al. (2025); Feng et al. (2025) Better uncertainty exposure does not by itself prove that the reported candidates are externally correct.
4. Validation class Which external standard was actually passed: focal-source comparison, intracranial stimulation, simultaneous invasive recording, or another regime-specific ladder. Pascarella et al. (2023); Unnwongse et al. (2023); Hao et al. (2025) A direct-validation result in one source regime does not automatically transfer to all depths, source extents, or clinical settings.
Connectivity is not just one more EEG output

It is tempting to think that once a source estimate exists, a connectivity graph is just the next summary. That is too strong. Vinck et al. (2011) made wPLI safer against some zero-lag mixing, but Haufe et al. (2013) showed that sensor-space connectivity remains strongly limited by volume conduction, Palva et al. (2018) showed that even source-space measures can create ghost interactions, and Miljevic et al. (2025) showed that sensor-space network results move with rereferencing, epoch design, and metric choice. On this site, EEG connectivity is therefore read as a model- and pipeline-conditioned estimator, not as automatically discovered wiring.

Why QC and pretreatment are important

EEG is a measurement that is susceptible to noise, but the important correction is that the issue is not only noise. Results can move with eye blinks, myoelectricity, body movements, power supply noise, reference choice, electrode layout, device-side filtering, and site-specific setup. Therefore, it is not enough to keep only a clean-looking figure; one must also record the measurement condition that made that figure possible.

What you want to keep as a minimum

  • Reference method:What standard was used to measure the potential difference?
  • Recording setup:Which device chain, sampling policy, and electrode layout were used?
  • Filter:Which frequency band is passed through?
  • Artifact processing:Which noise was removed and how?
  • Exclusion criteria:Which data were excluded and why?
Same task does not guarantee the same measurement condition

Xu et al. (2020) showed that cross-dataset deep-learning results move with environmental variability such as amplifier, cap, sampling rate, and filtering. That is why this site does not treat setup as a background nuisance. It is part of the observation model and has to be logged before the score is interpreted.

Shared multimodal factors can still be mixed physiology

Even when a paper reports one common EEG-fMRI or EEG-PET-MRI factor, that factor can still mix neural and non-neural contributions. Gold et al. (2024) show that fMRI-autonomic covariance grows as vigilance decreases in simultaneous EEG-fMRI-autonomic recordings, Özbay et al. (2019) show sympathetic contributions to the fMRI signal, and Epp et al. (2025) show that BOLD changes can oppose oxygen-metabolism changes across a large fraction of cortex. On this site, a common factor is therefore not promoted automatically to the target neural variable.

References

  1. Pernet, C. R., Appelhoff, S., Gorgolewski, K. J., et al. (2019). EEG-BIDS, an extension to the brain imaging data structure for electroencephalography. Scientific Data, 6, 103. doi:10.1038/s41597-019-0104-8
  2. Pernet, C., Garrido, M. I., Gramfort, A., et al. (2020). Issues and recommendations from the OHBM COBIDAS MEEG committee for reproducible EEG and MEG research. Nature Neuroscience, 23, 1473-1483. doi:10.1038/s41593-020-00709-0
  3. Michel, C. M., & Brunet, D. (2019). EEG source imaging: a practical review of the analysis steps. Frontiers in Neurology, 10, 325. doi:10.3389/fneur.2019.00325
  4. Mikulan, E., Russo, S., Bares, M., et al. (2020). Simultaneous human intracerebral stimulation and HD-EEG, ground-truth for source localization methods. Scientific Data, 7, 127. doi:10.1038/s41597-020-0467-x
  5. Seeber, M., Cantonas, L.-M., Hoevels, M., et al. (2019). Subcortical electrophysiological activity is detectable with high-density EEG source imaging. Nature Communications, 10, 753. doi:10.1038/s41467-019-08725-w
  6. Unnwongse, K., Achakulvisut, T., Wu, J. Y., et al. (2023). Direct validation of EEG source imaging by intracranial electric stimulation in human patients. Brain Communications, 5(2), fcad023. doi:10.1093/braincomms/fcad023
  7. Hao, S., Zhao, H., Feng, Z., et al. (2025). HD-EEG source imaging with simultaneous SEEG recording in drug-resistant epilepsy. Epilepsia, 66(11), 4451-4464. doi:10.1111/epi.18552
  8. Pascarella, A., Mikulan, E., Sciacchitano, F., et al. (2023). An in-vivo validation of ESI methods with focal sources. NeuroImage, 277, 120219. doi:10.1016/j.neuroimage.2023.120219
  9. Vorwerk, J., Wolters, C. H., & Baumgarten, D. (2024). Global sensitivity of EEG source analysis to tissue conductivity uncertainties. Frontiers in Human Neuroscience, 18, 1335212. doi:10.3389/fnhum.2024.1335212
  10. Luria, G., Viani, S., Pascarella, A., et al. (2024). The SESAMEEG package: a probabilistic tool for source localization and uncertainty quantification in M/EEG. Frontiers in Human Neuroscience, 18, 1359753. doi:10.3389/fnhum.2024.1359753
  11. Tong, P. F., Yang, H., Ding, X., et al. (2025). Debiased Estimation and Inference for Spatial-Temporal EEG/MEG Source Imaging. IEEE Transactions on Medical Imaging. doi:10.1109/TMI.2024.3506596
  12. Feng, Z., Mishne, G., Hashemi, A., et al. (2025). Block-Champagne: Imaging extended E/MEG source activation with empirical Bayesian uncertainty quantification. IEEE Transactions on Medical Imaging. doi:10.1109/TMI.2025.3642620
  13. Vorwerk, J., Köhler, T., Güllmar, D., et al. (2026). Potential of EEG and EEG/MEG skull conductivity estimation to improve source analysis in presurgical evaluation of epilepsy. Journal of Neural Engineering, 23(1), 016007. doi:10.1088/1741-2552/ae2f01
  14. Xu, M., Yao, S., Wei, Z., et al. (2020). Cross-dataset variability problem in EEG decoding with deep learning. Frontiers in Human Neuroscience, 14, 103. doi:10.3389/fnhum.2020.00103
  15. Vinck, M., Oostenveld, R., van Wingerden, M., Battaglia, F., & Pennartz, C. M. A. (2011). An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and sample-size bias. NeuroImage, 55(4), 1548-1565. doi:10.1016/j.neuroimage.2011.01.055
  16. Haufe, S., Nikulin, V. V., Müller, K.-R., & Nolte, G. (2013). A critical assessment of connectivity measures for EEG data: a simulation study. NeuroImage, 64, 120-133. doi:10.1016/j.neuroimage.2012.09.036
  17. Palva, J. M., Wang, S. H., Palva, S., et al. (2018). Ghost interactions in MEG/EEG source space: a note of caution on inter-areal coupling measures. NeuroImage, 173, 632-643. doi:10.1016/j.neuroimage.2018.02.032
  18. Miljevic, A., Murphy, O. W., Fitzgerald, P. B., & Bailey, N. W. (2025). Estimating sensor-space EEG connectivity PART 1: Identifying best performing methods for functional connectivity in simulated data. Clinical Neurophysiology, 174, 73-83. doi:10.1016/j.clinph.2025.03.043
  19. Kothe, C., Shirazi, S. Y., Stenner, T., et al. (2025). The lab streaming layer for synchronized multimodal recording. Imaging Neuroscience, 3, IMAG.a.136. doi:10.1162/IMAG.a.136
  20. Vafaii, H., Mandino, F., Desrosiers-Grégoire, G., et al. (2024). Multimodal measures of spontaneous brain activity reveal both common and divergent patterns of cortical functional organization. Nature Communications, 15, 581. doi:10.1038/s41467-023-44363-z
  21. Chen, J. E., Lewis, L. D., Coursey, S. E., et al. (2025). Simultaneous EEG-PET-MRI identifies temporally coupled and spatially structured brain dynamics across wakefulness and NREM sleep. Nature Communications, 16, 8887. doi:10.1038/s41467-025-64414-x
  22. Gold, B. P., Goodale, S. E., Zhao, C., et al. (2024). Functional MRI signals exhibit stronger covariation with peripheral autonomic measures as vigilance decreases. Imaging Neuroscience, 2, IMAG.a.00287. doi:10.1162/imag_a_00287
  23. Özbay, P. S., Chang, C., Picchioni, D., et al. (2019). Sympathetic activity contributes to the fMRI signal. Communications Biology, 2, 421. doi:10.1038/s42003-019-0659-0
  24. Epp, S. M., Castrillón, G., Yuan, B., et al. (2025). BOLD signal changes can oppose oxygen metabolism across the human cortex. Nature Neuroscience. doi:10.1038/s41593-025-02132-9
  25. Rohaut, B., Hermann, B., Kaufmann, B. C., et al. (2024). Multimodal assessment improves neuroprognosis performance in clinically unresponsive critical-care patients with brain injury. Nature Medicine, 30, 2482-2491. doi:10.1038/s41591-024-03019-1
  26. Amiri, M., Bødker Andersen, M., Jørgensen, S. H., et al. (2023). Multimodal prediction of residual consciousness in the intensive care unit: the CONNECT-ME study. Brain, 146(1), 50-64. doi:10.1093/brain/awac335
  27. Manasova, D., Belloli, L. M. L., Rosenfelder, M. J., et al. (2026). Multimodal multicentre investigation of diagnostic and prognostic markers in disorders of consciousness. Brain. doi:10.1093/brain/awaf412
  28. Boto, E., Holmes, N., Leggett, J., et al. (2018). Moving magnetoencephalography towards real-world applications with a wearable system. Nature, 555, 657-661. doi:10.1038/nature26147
  29. Rea, M., Holmes, N., Hill, R. M., et al. (2021). Precision magnetic field modelling and control for wearable magnetoencephalography. NeuroImage, 241, 118401. doi:10.1016/j.neuroimage.2021.118401
  30. Mellor, S. J., Tierney, T. M., O'Neill, G. C., et al. (2022). Magnetic field mapping and correction for moving OP-MEG. IEEE Transactions on Biomedical Engineering, 69(2), 528-536. doi:10.1109/TBME.2021.3100770
  31. Holmes, N., Leggett, J., Hill, R. M., et al. (2025). Wearable magnetoencephalography in a lightly shielded environment. IEEE Transactions on Biomedical Engineering, 72(2), 609-618. doi:10.1109/TBME.2024.3465654
  32. Rhodes, N., Rier, L., Boto, E., Hill, R. M., & Brookes, M. J. (2025). Source reconstruction without an MRI using optically pumped magnetometer-based magnetoencephalography. Imaging Neuroscience, 3, IMAG.a.8. doi:10.1162/IMAG.a.8
  33. Wu, T., Xiao, W., Peng, X., Wu, T., & Guo, H. (2025). Crosstalk reduction in optically pumped magnetometers arrays for biomagnetic measurement. Review of Scientific Instruments, 96(8), 085004. doi:10.1063/5.0273491
  34. Spedden, M. E., O'Neill, G. C., West, T. O., et al. (2025). Using wearable MEG to study the neural control of human stepping. Sensors, 25(13), 4160. doi:10.3390/s25134160