Ear-EEG / cEEGrid Evidence Matrix
Working synthesis from the detailed paper cards and local text extraction. Use this as a review scaffold; verify exact figures, tables, and page numbers in the PDFs before formal citation.
Cross-Paper Matrix
| # | Paper | Route / hardware | Task / evidence base | Main finding | Key limitation | Use for our system |
|---|---|---|---|---|---|---|
| 01 | Looney 2011 | Custom in-ear AgCl earplug | Alpha attenuation; single-subject scalp comparison | In-ear alpha increased after eye closure; temporal/mastoid correlations were strongest. | Extremely small sample; hardware details incomplete. | Minimal viability test: EC/EO alpha plus scalp reference. |
| 02 | Looney 2012 | Custom in-ear Ear-EEG | AAR, ASSR, visual P300, SSVEP concepts | Ear-EEG showed AAR/ASSR/P300 with lower amplitude but usable SNR. | Mostly proof-of-concept examples. | Validate time, frequency, and ERP markers, not one task only. |
| 03 | Kidmose 2012 | Truly in-ear earpiece, g.USBamp | ASSR, P1-N1-P2, MMN | Ear-EEG amplitudes were 10-20 dB lower, but SNR was comparable after averaging. | Sparse sample reporting; auditory-only. | Hearing applications should start with ASSR/AEP/MMN. |
| 04 | Kidmose 2013 | Ear-EEG vs scalp | ASSR, SSVEP, AEP, VEP | ASSR was most favorable for ear sites; transient/visual responses were weaker. | Ear coverage cannot replace full scalp sources. | Choose validation tasks by source location and orientation. |
| 05 | Mikkelsen 2015 | Ear-EEG method characterization | ASSR, alpha, MMN comparison | Ear electrodes resembled temporal scalp channels; ASSR SNR was close to TP9/TP10. | Some channels rejected; MMN harder. | Use ASSR/alpha as robust early benchmarks. |
| 06 | Bleichner 2015 | Miniaturized scalp/ear electrodes | Online visual P300 speller | Ear HC site captured P300 with lower amplitude but similar effect size windows. | Ear-only capability mixed with scalp sites in parts of setup. | Report whether performance is truly ear-only. |
| 07 | Debener 2015 | Original bilateral cEEGrid + SMARTING + phone | 7 h wear, EC/EO alpha, auditory oddball | Stable impedance, alpha in 9/10, P300 around 400 ms, LDA about 70% across sessions. | Smartphone timing latency; motion not stress-tested. | Core first replication: comfort, impedance, alpha, P300, cross-session classification. |
| 08 | Norton 2015 | Soft auricle/mastoid electronics | Long wear, alpha, SSVEP, P300 | Auricular soft electrodes supported 2-week alpha and high SSVEP spelling accuracy in small tests. | Small samples; not cEEGrid; hardware fabrication complexity. | Useful for conformal soft-electrode design ideas. |
| 09 | Bleichner 2016 | Bilateral cEEGrid vs 84-channel cap | Spatial auditory attention | cEEGrid median decoding 66% vs cap 70%; vertical long-distance pairs worked best. | Passive-BCI-level accuracy; real mobility untested. | For AAD, prioritize vertical bipolar pairs and behavior checks. |
| 10 | Mirkovic 2016 | Bilateral cEEGrid + cap comparison | Two-speaker envelope tracking | cEEGrid decoded attended speaker at 69.33% vs cap about 84.8%; spatial placement drove gap. | 30-60 s windows; cEEGrid below cap. | Honest AAD baseline and cap/ear layout comparison. |
| 11 | Goverdovsky 2016 | Generic viscoelastic in-ear earpiece | 8 h impedance, ASSR/VEP/SSVEP | Impedance stayed mostly <10 kOhm; ASSR close to mastoid/temporal, SSVEP weaker. | Detailed EEG mainly one subject; external reference/ground. | Generic earpiece route must handle jaw motion and occlusion. |
| 12 | Bleichner & Debener 2017 | cEEGrid / transparent EEG review | Method review + pilot use cases | cEEGrid offers more spatial information than in-ear but less than cap; best as research platform. | Review/pilot evidence, not clinical validation. | Define transparent EEG requirements beyond “wearable.” |
| 13 | Pacharra 2017 | cEEGrid vs 64-channel cap | Visual Simon task | cEEGrid captured P1/N1, P300, and posterior/temporal ERL; motor LRP was weak. | Lower occipital/central SNR; task-specific sensitivity. | Avoid motor LRP as first cEEGrid target. |
| 14 | Sterr 2018 | Behind-/around-ear cEEGrid sleep | Same-night cEEGrid vs PSG scoring | Sleep-stage kappa was moderate; 9/13 sleep parameters agreed well. | REM/NREM and AASM amplitude criteria issues; data losses. | Sleep proof-of-concept needs PSG and user-error safeguards. |
| 15 | Denk 2018 | Hearing device + right cEEGrid/in-concha | Hearing-device ERP switch | cEEGrid SNR exceeded local in-concha; N100/P300 detected to device setting changes. | ERP confounded by hardware noise onset. | Control acoustic/device artifacts in hearable EEG tests. |
| 16 | Mikkelsen 2019 | Bilateral cEEGrid sleep ML | Sleep/wake random forest vs PSG/actigraphy | Automatic cEEGrid beat actigraphy/manual cEEGrid and approached PSG-derived systems. | Healthy small cohort; WASO/REM latency weak. | Use PSG labels, LOSO validation, and derived cEEGrid channels. |
| 17 | Knierim/Reali 2021 | OpenBCI Cyton+Daisy + cEEGrid | Alpha, workload, ECG, flow | Low-cost setup captured Berger alpha, workload frequency effects, and ECG R-waves in 4/5 valid cases. | N=6 all male; motion and workload specificity limited. | Low-cost prototype route; document channel map and timing. |
| 18 | Knierim 2022 | OpenBCI-cEEGrid Adapter | HardwareX adapter, BOM, artefact demo, bruxism | Provides PCB/enclosure/BOM; impedance improved over minutes; bruxism F1=0.73 in held-out session. | Hardware report; bruxism single user; OpenBCI timing/channel limits. | Primary replication guide for adapter, enclosure, routing, and artefact reference. |
| 19 | Holtze 2022 | cEEGrid continuous speech | 36 participants, envelope tracking/ISC/entropy | Envelope tracking about 71-72%; individualized non-nested tuning overfit; ISC and entropy added evidence. | Reuses datasets; ASR benefit limited; calibration data small. | Use multi-metric AAD and nested validation. |
| 20 | Knierim 2023 | Smarting vs OpenBCI cEEGrid | Timing test, alpha/workload/P300, simultaneous ERP | OpenBCI can reproduce alpha/workload/P300 after lag/jitter correction; Smarting timing is cleaner. | Small samples; simultaneous comparison single subject. | Mandatory timing test before ERP/AAD with OpenBCI. |
| 21 | Van Den Broucke 2023 | Custom wireless 16 kHz cEEGrid hardware | ABR-oriented hardware vs BioSemi | 16 kHz, low-noise wearable design can record ABR-like waveforms from cEEGrid. | Single-subject evaluation; incomplete channel map. | ABR needs different front end than OpenBCI/Smarting. |
| 22 | Zhu 2024 | cEEGrid four-speaker AAD | 16 participants, SR and deep ASAD | Four-speaker ear-EEG SR reached 41.3% at 60 s; deep ASAD claimed >90% at 1 s. | Preprint; deep models risk trial shortcut learning. | Use 25% chance baseline and strict trial/subject-independent validation. |
| 23 | Geirnaert 2025 | Simultaneous scalp, around-ear, dry in-ear | 15 participants, two-speaker AAD | 60 s accuracy: scalp 83.44%, around-ear 67.22%, in-ear 61.11%; around-ear generalized better. | Preprint; young normal-hearing sample; wet/dry confound. | Best current route-comparison benchmark for AAD trade-offs. |
Practical Evidence Tiers
- Replication-ready foundations: Debener 2015, Bleichner 2016, Mirkovic 2016, Knierim 2021/2022/2023.
- Hardware design references: Looney 2011/2012 for in-ear, Debener 2015 for cEEGrid, Knierim 2022 for OpenBCI adapter, Van Den Broucke 2023 for high-sampling ABR.
- Application benchmarks: Kidmose/Mikkelsen for auditory evoked responses, Sterr/Mikkelsen for sleep, Holtze/Zhu/Geirnaert for AAD.
- Highest-risk claims: real-world motion robustness, fast AAD gain switching, clinical sleep diagnosis, deep AAD accuracy, and bruxism generalization.