Graduate Student Seminar Series – Mohammad Rezaei

Location: HS610 – 155 College St, Room 610

Presentation Title: Inferring cognitive state underlying conflict choices in verbal Stroop task Abstract: The subthalamic nucleus (STN) within the basal ganglia interacts with the medial prefrontal cortex (mPFC) to form a control loop, particularly active when the brain encounters contradictory information from different sensory systems or conflicting inputs from sensory data and pre-existing knowledge. Experimental studies have shown that significant increases in theta activity (2–8 Hz) in both the STN and mPFC, along with enhanced phase synchronization between these regions, are key features of conflict processing. While these neural markers highlight the critical role of the STN-mPFC circuitry in managing conflicts, a low-dimensional representation of the mPFC-STN interaction—referred to as a cognitive state—that links neural activities to behavioral outcomes, such as response time, remains to be fully understood. To address this, we propose the heterogeneous input discriminative-generative decoder (HI-DGD) model, designed to infer cognitive states underlying decision-making by integrating neural activities from the STN and mPFC with behavioral signals, specifically response times recorded from ten Parkinson’s disease (PD) patients performing a Stroop task. PD patients may experience conflict processing that differs quantitatively, and possibly qualitatively, from that of healthy individuals. Our results, derived from extensive synthetic and experimental data, demonstrate that the HI-DGD model can simultaneously process neural and behavioral data to estimate cognitive states in conflict and non-conflict trials more effectively than traditional methods. Moreover, the HI-DGD model identifies which neural features significantly contribute to conflict and non-conflict decisions, with the inferred features closely aligning with those reported in experimental studies. Importantly, the HI-DGD model’s ability to estimate cognitive states from single trials makes it highly suitable for use in closed-loop neuromodulation systems.

Supervisor Name: Milos Popovic

Year of Study: 5

Program of Study: PhD