A gated artifact management pipeline for low-density eyewear EEG.

A wearable EEG pipeline identified contaminated segments and left cleaner segments untouched. Results support more selective signal cleanup, but the work was performed offline and does not establish clinical accuracy or consumer readiness.

Costanzo Palmisciano A, Farabbi A, Latino F, Rossi M, Antonello N, Cerveri P, Mainardi L.
Physiological measurement · Publication date: 2026-09-29
Added to site: 2026-09-30
DOI: 10.1088/1361-6579/ae99ab

Electronic publication date verified in Europe PMC and Crossref. This source date is separate from the site's addition date.

Modalities

EEG, Wearable EEG

Clinical focus

Artifact management, Dry electrodes, Edge processing

Population

Adults contributing low-density dry-electrode eyewear EEG recordings and participants represented in an external ocular-artifact benchmark

Sample

29 participants in the eyewear EEG dataset; external benchmark from 24 participants across two studies

Intervention / method

A lightweight gate detected contaminated windows before a three-class artifact classifier and U-Net denoiser processed eye-blink, horizontal-eye-movement and facial-movement artifacts

Comparator

Leave-one-subject-out validation, an independent external benchmark and comparisons with two single-channel denoising models

Clinician summary

In 29 participants, the gate reached balanced accuracy 0.89; artifact classification reached 0.84 with median macro-F1 0.83. End-to-end median normalized RMSE was 0.19 with spectral cosine similarity 0.89. External gate accuracy was 0.82, with dataset-specific variation across later stages.

How to interpret the findings

Testing was offline, participant performance varied, and the four temporal channels with nose reference limited spatial information. Eye blinks and horizontal eye movements were often confused. Denoising targets partly relied on ASR and semi-synthetic contamination, the proprietary dataset was not publicly available, and embedded deployment remains future work.

Equipment and disclosures

A four-temporal-channel low-density dry-electrode EEG eyewear platform with nose referencing was used. The summary does not identify it as a clinical diagnostic device.

The work was carried out in the EssilorLuxottica Smart Eyewear Lab and supported by a European Union NextGenerationEU scholarship. Two authors were EssilorLuxottica employees; the other authors declared no conflict, and the article states that commercial interests did not influence design, analysis, interpretation or reporting.

Related evidence

Compare findings from the same clinical area, including mixed results.