Decoding semantic information from human electrocorticographic (ECoG) signals

Area of Science:

• Neuroscience
• Cognitive Science
• Biomedical Engineering

Background:

• Understanding semantic information processing in the human brain is crucial for neuroscience.
• Developing brain-computer interface (BCI) systems requires decoding complex cognitive states from neural activity.

Purpose of the Study:

• To investigate the feasibility of decoding semantic information from human cortical activity.
• To identify brain regions involved in semantic processing.
• To evaluate machine learning algorithms for semantic decoding using electrocorticography (ECoG) data.

Main Methods:

• Recorded electrocorticographic (ECoG) signals from four subjects during language tasks.
• Analyzed high-gamma band (60-120 Hz) activation in specific brain regions like the left inferior frontal gyrus (LIFG) and posterior superior temporal gyrus (pSTG).
• Utilized Gaussian Naïve Bayes and Support Vector Machine classifiers to predict semantic categories from ECoG data.

Main Results:

• Observed robust high-gamma band activation in LIFG and pSTG, correlating with speech production and perception.
• Machine learning classifiers accurately predicted the semantic category of objects based on cortical activity.
• Demonstrated successful decoding of semantic information from ECoG signals across frontal, temporal, and parietal cortices.

Conclusions:

• Decoding semantic information from human cortical activity is feasible.
• These findings support the development of semantic-based brain-computer interface (BCI) systems.
• BCI systems could aid individuals with severe communication disorders by enabling thought and intention expression.