Integrating Choice Selection and Feedback During Decision-Making: A Role for Functional Connectivity

To learn from decisions, individuals form cognitive representations of the outcomes of their choices. Prediction error (PE) is a metric that represents the difference between a choice's expected and actual reward in its magnitude and valence. Previous work has demonstrated changes in frequency band-specific power and functional connectivity (FC) that correspond to changes in PE during decision-making tasks. However, few studies have compared the neural representations of choice across these encoding modalities (band power vs. FC). We address this gap in knowledge by analyzing the intracranial EEG (iEEG) data from 15 participants during a decision-making task, during which participants chose between a guaranteed small reward or a chance of receiving a larger reward, then received feedback about their actual reward values. We calculated average power and coherence (a metric of functional connectivity) during the second following feedback. To compare PE-relevant information across encoding modalities, we used power and coherence feature sets to classify the PE magnitude and valence for each trial using logistic regression. Additionally, we examined the regression coefficients to identify the frequency bands of power and coherence that carried the most PE-relevant information. For most participants, both feature sets resulted in above-chance classification of PE magnitude and valence. PE-relevant information was concentrated in the higher frequency bands in the power feature set, and to a lesser degree in the coherence feature set. Although the power feature set outperformed the coherence feature set using this simple classification approach, our results suggest that both modalities play roles in encoding PE.