Decoding and Characterizing the Intracranial Representation of Semantic Information

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Brain-computer interfaces (BCIs) excel at decoding motor signals for speech.
• Decoding higher-level semantic representations from cortical activity remains less explored.
• Understanding semantic decoding is crucial for advancing BCIs and language neuroscience.

Purpose of the Study:

• To investigate the feasibility of decoding semantic information from human intracranial neural activity.
• To determine if conceptual category membership can be identified from brain signals.
• To explore semantic decoding as a complementary approach for future language BCIs.

Main Methods:

• Recorded intracranial neural activity using stereotactic electroencephalography (sEEG) in epilepsy patients.
• Extracted high-gamma power from local field potentials during language tasks.
• Utilized supervised machine learning with cross-validation for classification.

Main Results:

• Semantic category information was decoded significantly above chance levels.
• Mean classification accuracy reached 29.8% across 15 semantic categories (chance = 6.7%).
• High-gamma activity was found to contain extractable information about conceptual category membership.

Conclusions:

• Semantic information is accessible from intracranial population recordings.
• High-gamma power provides a viable neural signal for semantic decoding.
• Findings support the development of BCIs leveraging conceptual information and advance understanding of the human language network.