Computational mechanisms in ADHD
Wenting Wang, Peter Dayan, Tobias Kaufmann
Attention-Deficit/Hyperactivity Disorder (ADHD) is a common neurodevelopmental condition affecting children and adolescents, often accompanied by other psychiatric disorders. Despite its prevalence, ADHD is frequently under-diagnosed. Understanding the underlying mechanisms involves a complex interplay of genetic, environmental and neurobiological factors, impacting various cognitive functions through alterations in specific brain regions and circuits.
The Adolescent Brain Cognitive Development (ABCD) study serves as a starting point for this research. This extensive longitudinal study tracks brain development in 11,875 youths aged 9-10 across the United States over a decade into young adulthood. It includes three cognitive tasks, namely the Stop Signal Task (SST), Monetary Incentive Delay (MID), and Emotional n-back (EN-Back) These tasks effectively capture elemental aspects of ADHD behaviors such as execution, inhibition, impulsivity, motivation, reward processing, working memory, and emotion regulation.
In this research, we employ Partially Observable Markov Decision Process (POMDP) modeling to understand the cognitive processes of subjects during the aforementioned tasks. POMDP is a mathematical framework used for decision-making under uncertainty, where states are partially observable to agents, enabling the derivation of behavior policies that maximize overall rewards over time. For instance, in the SST, the objective is to maximize rewards by achieving pre-defined benefits (correct inhibition) while minimizing costs (delayed time and errors). The POMDP model allows us to predict subjects' inhibitory control ability, reaction time, and other hidden parameters such as prediction error, environment volatility, and learning rate.
Additionally, the computational parameters extracted by POMDP can be mapped onto neuroimaging data of brain regions or connectivity networks, enabling further investigation of the neural implementation of cognitive functions.
To enhance our understanding in ADHD, we propose a combination of theory-driven and data-driven methods. This approach not only allows us to test hypothetical mechanisms but also depicts behavioral and neural feature distributions across populations that might be missed with smaller samples. The iterative process also enables us to refine, validate, and improve the models' accuracy, explanatory power, and predictive capabilities.
Furthermore, large-scale longitudinal and cross-sectional studies tracking neurodevelopmental stages and deficits over time hold the promise of predicting ADHD diagnoses before the critical window for brain region development. These findings can significantly contribute to clinical diagnoses and enable personalized treatment and intervention approaches for ADHD and related mental disorders.