Using data from cue presentations results in grossly overestimating semantic BCI performance.

Area of Science:

• Neuroscience
• Cognitive Science
• Biomedical Engineering

Background:

• Neuroimaging enables semantic neural decoding, identifying concepts from brain activity.
• This has potential for brain-computer interfaces (BCIs) but faces implementation challenges.
• Existing electroencephalography (EEG)-based semantic decoding often uses cue-present data, impacting reliability.

Purpose of the Study:

• To investigate the impact of cue presentation on EEG-based semantic decoding accuracy.
• To differentiate semantic categories (animals vs. tools) during distinct cue and mental task periods.
• To assess the feasibility of semantic decoding without external cues.

Main Methods:

• An experiment separated cue presentation from mental task periods.
• Four distinct mental tasks were employed.
• State-of-the-art decoding analyses were applied to EEG data.
• Classification accuracy was assessed during cue and mental task phases.

Main Results:

• Significant mean classification accuracies up to 71.3% were achieved during cue presentation.
• Decoding accuracy did not reach significance during mental task periods, even with adapted analyses.
• Results suggest cue presentation significantly influences decoding performance.

Conclusions:

• Neural activity during cue presentation may inflate semantic decoding performance.
• Semantic decoding without external cues is more challenging than previously suggested.
• Findings necessitate re-evaluation of current methods for developing robust semantic BCI applications.