Decoding Hidden Cognitive States From Behavior and Physiology Using a Bayesian Approach.

Area of Science:

• Computational Neuroscience
• Cognitive Science
• Systems Neuroscience

Background:

• Measuring complex cognitive processes like learning and flexibility is challenging due to high-dimensional neural activity.
• Existing methods struggle with continuous, objective sampling of cognitive states.
• Dimensionality reduction is crucial for both research and clinical applications.

Discussion:

• A novel modeling framework defines cognitive processes as low-dimensional latent variables (cognitive states).
• This framework separates neural and behavioral data modeling into encoding and decoding steps.
• It utilizes state-space models for behavioral decoding and generalized linear models (GLMs) for neural encoding.

Key Insights:

• The framework successfully estimates cognitive states from neural data (LFP power) in human participants.
• Neural decoding achieved 95% confidence interval accuracy compared to behavioral decoding across 90% of trials.
• Incorporates LFP spectral power for robust cognitive state encoding and decoding.

Outlook:

• This approach captures the temporal dynamics of cognitive processes.
• Enables the development of closed-loop experiments for real-time cognitive state manipulation.
• Paves the way for novel therapeutic interventions targeting cognitive functions.