EEG Based Inference of Causal Cortical Network Dynamics in Reward-Based Decision Making

Experimental investigation and monitoring of reward signaling implicated in addiction and neurological disorders has traditionally been limited to invasive measurement of deep-brain dopamine activity. Here we introduce a systematic methodology and algorithmic pipeline to quantify causal relationships between regions of interest (ROIs) in the cerebral cortex revealing reward-based signaling pathways involved in human decision making using only non-invasive scalp electroencephalography (EEG). The data is processed by extracting epochs around time-locked stimuli of interest and performing independent component analysis (ICA) on individual datasets to remove artifacts and identify cortical sources. The pipeline entails identifying ROI's with the Measure Projection Toolbox (MPT) through clustering of ICs, localizing current sources in these ROIs using Bayesian inference based constrained low resolution electromagnetic tomography (cLORETA), and computing causal relationships between ROI's using the Source Information Flow Toolbox (SIFT). The proposed methodology and pipeline are demonstrated on 64-channel scalp EEG signals recorded from healthy adults performing a reward-based decision making task conducted through a brain computer interface (BCI) framework. In comparison to a standard method for Group-ICA, our pipeline generates far more biologically plausible and consistent causal connections between ROIs.