AOR: adaptive orthogonal regression for depression feature selection with incomplete EEG channels

Published online October 26, 2026. A new analysis method maintained strong depression-classification performance when EEG channels were missing or noisy across two existing datasets. This is a retrospective computing result, not a clinical diagnostic test, and it does not show that EEG alone can diagnose depression in routine care.

Dong W, Gong Z, Xu X, Liu K, Wei F, Zhuo L.
Journal of Neural Engineering · Publication date: 2026-10-26
Added to site: 2026-10-27
DOI: 10.1088/1741-2552/aeb32f

Published online October 26, 2026 and assigned to the October 2026 issue. Added to Neurofeedback.com October 27, 2026.

Modalities

EEG, Machine learning

Clinical focus

Depression classification, Missing channels, Feature selection

Population

Participants represented in the MODMA and PRED-d003 EEG datasets

Sample

The checked abstract does not report participant denominators

Intervention / method

Adaptive orthogonal regression selected EEG features under simulated incomplete-channel conditions across 3-, 64- and 128-channel configurations

Comparator

Thirteen state-of-the-art feature-selection methods across different missing-data ratios

Clinician summary

The authors report average cross-subject accuracy of 92.25% and 95.12% across missing ratios on MODMA and PRED-d003, with EEG kurtosis receiving high feature weight. Performance was assessed across several channel configurations and against 13 feature-selection methods. Prospective external clinical validation, calibration, subgroup performance and decision consequences are not reported in the abstract.

How to interpret the findings

The study reanalyzed existing datasets and the checked abstract omits participant denominators, recruitment details, diagnostic-reference procedures, confidence intervals and independent prospective validation. Accuracy in curated datasets can overstate performance in new clinics. Feature weights are not a biological explanation, and the result does not establish clinical utility, fairness or safety.

Equipment and disclosures

The abstract reports 3-, 64- and 128-channel analyses but does not establish that more channels improve patient care or identify a superior commercial EEG system. No equipment is attributed to BrainMaster.

Crossref lists support from the National Natural Science Foundation of China and the Engineering Research Center of Autonomous Unmanned System Technology, Ministry of Education. Competing-interest information was not available in the checked abstract or retained Crossref record. No correction, retraction or expression-of-concern relation appeared in the retained records; this is a bounded check. AI-assisted source review.

Related evidence

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