Deep learning on independent spatial EEG activity patterns delineates time windows relevant for response inhibition

Area of Science:

• Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• Inhibitory control is vital for executive functions and goal-directed behavior.
• Previous neurophysiological studies often relied on correlative data, limiting insights into predictive neural dynamics for response inhibition.
• The complex spatio-temporal nature of electroencephalography (EEG) data presents challenges in identifying predictive neural markers.

Purpose of the Study:

• To investigate whether independent spatial activity profiles derived from EEG data can predict response inhibition.
• To apply explainable artificial intelligence (AI) methods to EEG data from a Go/Nogo task to identify predictive neural signatures.
• To explore the functional significance of event-related potentials in inhibitory control.

Main Methods:

• Combined independent component analysis (ICA) with EEG-based deep learning.
• Utilized data from 255 participants performing a Go/Nogo task.
• Employed source localization analyses to identify brain regions associated with activity profiles.

Main Results:

• Identified four dissociable spatial activity profiles crucial for classifying Go and Nogo trials using deep learning.
• Neural activity between 300 and 550 ms post-stimulus was most informative across all identified profiles.
• Source localization linked these profiles to the pre-central gyrus (BA6), middle frontal gyrus (BA10), inferior frontal gyrus (BA46), and insular cortex (BA13).

Conclusions:

• Independent spatial activity profiles in EEG data are useful predictors of response inhibition.
• Specific spatio-temporal patterns, particularly between 300-550 ms, reflect underlying neural processes involved in inhibitory control.
• Findings contribute to understanding the functional significance of neural correlates of executive functions and response inhibition.