Time-frequency-spatial channel attention network for semantic decoding: an exploratory EEG study.

Area of Science:

• Neuroscience
• Cognitive Science
• Biomedical Engineering

Background:

• Semantic decoding is vital for understanding neural language processing.
• Brain-computer interface (BCI) technology offers new avenues for semantic decoding research.
• Limited Chinese EEG datasets hinder the study of semantic representations.

Purpose of the Study:

• To develop and validate a novel semantic task paradigm for decoding language comprehension and expression using scalp EEG.
• To investigate the neural representations of semantics in language perception, particularly in individuals with aphasia.
• To create an advanced deep learning model for enhanced semantic decoding from EEG data.

Main Methods:

• Designed a novel semantic task paradigm incorporating overt speech perception and silent speech imagery.
• Recruited 17 participants (aphasia patients and healthy controls) for EEG data collection.
• Developed and optimized a deep learning model, Time-Frequency-Spatial Channel Attention Network (TFSANet), for EEG feature extraction.

Main Results:

• TFSANet successfully decoded semantic information from EEG data for ten categories of four-word phrases.
• Decoding accuracy reached 60.73% for aphasia patients and 75.09% for healthy subjects.
• The model demonstrated improved ability in decoding semantically relevant EEG features through multidimensional extraction.

Conclusions:

• The proposed TFSANet model shows significant potential for decoding semantic information from EEG signals.
• The developed semantic task paradigm is effective for studying language processing in both healthy and clinical populations.
• This research contributes to advancing BCI technology for applications in language and communication research.