Latent brain state dynamics distinguish behavioral variability, impaired decision-making, and inattention

Attention Deficit Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental disorder affecting 5-10% of children, characterized by heterogeneous behaviors and symptoms. Intra-individual response variability (IIRV), measuring trial-to-trial response variance, is a consistent behavioral phenotype associated with attentional deficits and poor cognitive control. Abnormal fluctuations in latent brain states are hypothesized to contribute to unstable performance in ADHD, but this hypothesis requires rigorous testing. This study uses a Research Domain Criteria (RDOC) approach to examine IIRV, attention fluctuations, and decision-making processes as continuous distributions. A core set of brain regions, including the salience network (SN), frontoparietal network (FPN), and default mode network (DMN), are implicated in attention and cognitive control. Meta-analyses of fMRI studies in children with ADHD report abnormal activation in these networks during cognitive tasks. Aberrant intrinsic network interactions and weak task-modulated connectivity are associated with poorer attentional performance and more severe inattention symptoms. However, little is known about brain state dynamics during cognitive performance and their contribution to attention problems. This study uses a novel Bayesian switching linear dynamic systems (BSDS) model to investigate fluctuations of latent brain states and their relation to behavior. BSDS is an unsupervised Bayesian learning algorithm that determines hidden brain states and dynamic state transitions from observed data, avoiding limitations of existing methods. To minimize trial-by-trial behavioral adaptation, a simple choice response task was used. IIRV was indexed using standard deviation (std) and tau from an ex-Gaussian model, while latent cognitive processes were examined using a hierarchical drift-diffusion model (HDDM), estimating decision threshold, drift rate, and non-decision time. The study aimed to determine the relationship between engagement of a task-optimal dynamic brain state, behavioral variability, decision-making processes, and inattention, as well as the relationship between these factors and functional connectivity between SN, FPN, and DMN. Finally, the study aimed to determine if behavior and brain state dynamics could distinguish children with ADHD from typically developing (TD) children.

A substantial body of research has linked attentional deficits in ADHD to aberrant brain dynamics, primarily focusing on the salience network (SN), frontoparietal network (FPN), and default mode network (DMN). Meta-analyses consistently show abnormal activation within these networks in ADHD during attentionally demanding tasks. Aberrant intrinsic and task-modulated connectivity among these networks have also been linked to poorer attentional performance and more severe inattention. Previous studies have used resting-state fMRI to identify DMN abnormalities in ADHD, but research on dynamic brain states during cognitive tasks is limited. Increased intra-individual response variability (IIRV) is a robust behavioral marker of ADHD, associated with attention problems. While IIRV reflects attentional fluctuations, it doesn't capture underlying decision-making processes. Drift-diffusion models provide a framework to analyze latent decision-making components associated with IIRV, allowing for the investigation of how attention impacts evidence accumulation speed during decision-making. However, integrating these behavioral and neuroimaging findings into a comprehensive model of brain state dynamics in ADHD remains a challenge. This necessitates a more sophisticated approach than traditional fMRI analyses, capable of capturing the dynamic interplay between brain networks and behavior.

Fifty-two participants (29 children with ADHD, 23 TD children) completed a simple choice response task during ultrafast fMRI scanning (490 ms temporal resolution). The task involved making left or right button presses in response to corresponding arrows. Behavioral data were analyzed using several methods: Intra-individual response variability (IIRV) was quantified using standard deviation (std) and tau from an ex-Gaussian model fit to reaction times (RTs). A hierarchical drift-diffusion model (HDDM) was employed to estimate latent decision-making parameters: decision threshold (a), drift rate (v), and non-decision time (t). Neuroimaging data were analyzed using a novel Bayesian Switching Dynamic System (BSDS) model. BSDS is an unsupervised Bayesian learning algorithm that identified latent brain states from fMRI time series data extracted from regions of interest (ROIs) within the SN, FPN, and DMN. The occupancy rate (how often a state occurs) and mean lifetime (dwell time in a state) of each latent brain state were calculated. Correlations between behavioral measures (IIRV, HDDM parameters), clinical inattention scores, and brain state dynamics were examined using dimensional and categorical analyses. Specifically, the study investigated the relationships between brain state occupancy rates, mean lifetimes, IIRV, decision-making parameters, and inattention scores, both dimensionally and categorically (comparing ADHD and TD groups).

Demographic variables did not significantly differ between ADHD and TD groups. Dimensional analyses revealed a negative correlation between accuracy and inattention scores but no correlation between RT and attentional scores. Categorical analyses showed significantly lower accuracy in the ADHD group. Behavioral variability analyses indicated that ADHD children exhibited significantly larger RT std and tau compared to TD children. Dimensional analyses showed marginal correlations between RT std, RT tau, and inattention scores. HDDM analysis showed a significant negative correlation between drift rate (v) and RT tau, indicating that slower evidence accumulation was linked to increased response variability. Dimensionally, drift rate (v) negatively correlated with inattention scores, suggesting that slower evidence accumulation was associated with increased inattention. Categorically, ADHD children showed slower drift rates and longer non-decision times compared to TD children. BSDS analysis revealed four latent brain states (S1-S4). The occupancy rate of state S2 was positively correlated with inattention scores (although not significant after correction). Categorically, the occupancy rate of S1 was higher in TD children, while the occupancy rate of S2 was higher in ADHD children. The occupancy rate of S1 negatively correlated with RT std and RT tau, indicating that the occurrence of this state was associated with reduced response variability. Conversely, the occupancy rate of S2 positively correlated with RT std and RT tau, suggesting its involvement in increased response variability. These correlations were not found for mean lifetimes.

This study provides novel insights into the neural mechanisms underlying attention deficits in ADHD. The findings demonstrate that distinct latent brain states are associated with different aspects of attentional dysfunction: a task-optimal state (S1) is associated with reduced response variability and efficient evidence accumulation, while a non-optimal state (S2) is linked to increased response variability and inattention. The functional connectivity patterns between the SN, FPN, and DMN further elucidate the role of these networks in supporting or hindering attentional performance. The identification of dissociable brain states underscores the heterogeneity of ADHD and suggests that interventions targeting specific brain states could be more effective than those relying on a more general approach. The use of ultrafast fMRI and advanced computational modeling provided a more nuanced understanding of dynamic brain activity in ADHD.

This study successfully integrated behavioral and neuroimaging data using advanced computational methods to reveal distinct latent brain states associated with response variability, decision-making, and inattention in ADHD. The findings support the hypothesis that aberrant brain state dynamics contribute significantly to the core symptoms of ADHD and highlight the potential for developing targeted interventions based on these findings. Future research could investigate the longitudinal stability of these brain states, explore the influence of specific genetic or environmental factors on brain state dynamics, and test the effectiveness of interventions designed to modulate these states.

The relatively small sample size might limit the generalizability of the findings. The cross-sectional nature of the study prevents causal inferences about the relationship between brain states and behavioral outcomes. The choice response task, while simple, may not fully capture the complexity of attentional processes in real-world settings. The BSDS model, while innovative, involves assumptions about the underlying data structure that warrant further investigation.