TopoAdapter: a plug-and-play multi-hop topology adapter for MI-EEG decoding.

A very small software module modestly improved average EEG movement decoding across several benchmark tests. This is engineering evidence, not a clinical benefit or a prospective online BCI trial.

Cai Y, Guo Q, Lin X.
Journal of neuroscience methods · Publication date: 2026-09-18
Added to site: 2026-09-20
DOI: 10.1016/j.jneumeth.2026.110909

Electronic publication date verified in Europe PMC.

Modalities

EEG, BCI, Motor imagery

Clinical focus

EEG decoding, Electrode topology, Lightweight machine learning

Population

Existing motor-imagery and executed-movement EEG datasets

Sample

Three public EEG datasets plus a seven-backbone compatibility analysis on BCI Competition IV-2a

Intervention / method

A 70-parameter input adapter encoded fixed electrode-layout and multi-hop neighborhood information while preserving the downstream decoder architecture

Comparator

Baseline ATCNet runs, seven selected EEG backbones and the open-source Adaptive Channel Mixing Layer

Clinician summary

Mean accuracy gains were 1.17, 0.13 and 0.36 percentage points across three datasets. Two bootstrap intervals excluded zero, but no paired dataset test survived Holm correction; seven backbones improved on average in a separate compatibility analysis.

How to interpret the findings

Effect sizes were small and varied by dataset and participant. Corrected paired tests were not significant across the three primary datasets, the direct comparison with ACML was unresolved, and prospective online performance was not tested.

Equipment and disclosures

Computational validation used existing motor-imagery and executed-movement EEG datasets; acquisition hardware was not detailed in the accessible abstract.

Funding and competing-interest details were not available in the checked indexed record.

Related evidence

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