JOINT MODELING FOR LEARNING DECISION-MAKING DYNAMICS IN BEHAVIORAL EXPERIMENTS.

Area of Science:

• Computational Psychiatry
• Cognitive Neuroscience
• Machine Learning in Healthcare

Background:

• Major Depressive Disorder (MDD) is a significant global health issue linked to reward processing and attention deficits.
• Existing models often fail to capture the dynamic and strategic nature of decision-making in MDD.

Purpose of the Study:

• To develop and validate a novel computational framework integrating Reinforcement Learning (RL), Drift-Diffusion Models (DDM), and Hidden Markov Models (HMM).
• To jointly analyze reward-based decision-making and response times, accounting for strategy switching in MDD.
• To investigate brain-behavior associations in decision-making processes.

Main Methods:

• Proposed a novel RL-DDM-HMM framework to model latent state switching in decision-making.
• Implemented the model using a generalized expectation-maximization (EM) algorithm with forward-backward procedures.
• Validated the framework through extensive numerical simulations and application to the EMBARC study dataset.

Main Results:

• The proposed RL-DDM-HMM framework demonstrated superior performance over competing methods in various conditions.
• MDD patients exhibited significantly lower engagement and longer response times compared to healthy controls.
• Neuroimaging data correlated with decision-making characteristics only in the engaged state, not the lapsed state.

Conclusions:

• The novel computational framework effectively models dynamic decision-making strategies and their alterations in MDD.
• MDD is characterized by reduced engagement and impaired reward-based decision-making, with distinct neural correlates.
• This approach offers a promising tool for understanding cognitive deficits in psychiatric disorders.