Target-level convergence of directional prefrontal connectivity across Alzheimer's disease stages: an fNIRS study

Alzheimer's disease (AD) is increasingly recognized as involving network-level dysfunction rather than isolated regional impairment. Although functional and effective connectivity have been studied in dementia, the reorganization of directional information flow within prefrontal networks remains incompletely characterized, particularly with functional near-infrared spectroscopy (fNIRS).

This study investigated whether target-level convergence of directional effective connectivity within the prefrontal cortex differs across clinically defined cognitive groups, with emphasis on the mild cognitive impairment (MCI) vs. AD contrast.

fNIRS signals were recorded from 60 participants (healthy controls (HC), MCI, and AD) during resting state and cognitive paradigms, including working memory (2-back), semantic verbal fluency (SVFT), and Stroop tasks. Signals from 48 prefrontal channels were aggregated into eight anatomically defined regions. Directional interactions were quantified using transfer entropy (TE) with adaptive hyperparameter tuning. Group differences were assessed using linear mixed-effects models with covariate adjustment and target-wise false discovery rate correction. Robustness was evaluated through parameter sensitivity, covariate, and surrogate-normalization analyses.

The analysis revealed structured target-level convergence patterns of directional connectivity across clinically defined cognitive groups, with the most prominent alterations observed in the MCI vs. AD comparison. Compared with MCI, AD showed reduced directed information flow to the right frontopolar cortex (FPC-R) during the 2-back and SVFT tasks (β = -0.71 to -1.18, q < 0.05), indicating diminished convergence of inputs from distributed prefrontal regions. Conversely, convergence toward the left dorsolateral prefrontal cortex (DLPFC-L) increased across multiple paradigms (β = 0.56-0.85, q < 0.05). Additional task-specific alterations were observed during the Stroop paradigm, although some effects demonstrated reduced stability in sensitivity analyses. Exploratory ROC analysis suggested contrast-specific discriminative utility of selected convergence features, with moderate performance in MCI vs. AD (AUC ≈ 0.77).

These findings suggest that MCI and AD are associated with distinct patterns of directional information flow within prefrontal networks, characterized by reduced convergence toward FPC-R and increased directional influence toward DLPFC-L. Collectively, the results indicate that TE applied to task-based fNIRS can capture directional alterations in network connectivity, highlighting the potential of directional connectivity metrics for studying network-level changes associated with cognitive decline.