State Guided ICA of Functional Network Connectivity Reveals Temporal Signatures of Alzheimer's Disease

Elaheh Zendehrouh, Mohammad Se Sendi, Anees Abrol, Armin Iraji, Vince D. CalhounmedRxiv 2025 · DOI

Abstract

Identifying robust neuroimaging biomarkers for Alzheimer’s disease (AD) and mild cognitive impairment (MCI) is essential for early diagnosis and intervention. This study introduces a novel, fully automated, guided dynamic functional connectivity (dFNC) framework for extracting multiple dynamic measures to distinguish MCI/AD from cognitively normal (CN) individuals. Using resting-state fMRI data, subject-specific brain networks were extracted via spatially constrained independent component analysis (scICA). These components informed a dFNC analysis using a sliding-window approach. A state-constrained ICA (St-cICA) was then applied, utilizing five canonical brain states derived from the UK Biobank (UKBB) as biologically informed priors to guide individual-level decomposition in the OASIS-3 and ADNI datasets. By analyzing both primary and secondary connectivity states, we identified significant alterations in AD/MCI participants, characterized by increased dwell times and reduced neural flexibility. A classification model trained on these dynamic features achieved a mean AUC of approximately 0.85, highlighting the potential of guided dFNC as a biomarker framework for early-stage AD detection.

Introduction

Alzheimer’s Disease (AD) disrupts large-scale brain networks long before clinical symptoms like memory loss become apparent. Traditionally, neuroimaging research has relied on static Functional Network Connectivity (sFNC)—a single "snapshot" of brain correlations over the course of a scan. However, the human brain is inherently dynamic, constantly cycling through different states of engagement. While dynamic FNC (dFNC) methods exist, they often rely on group-level clustering that can obscure individual variability or fail to account for the noise inherent in clinical fMRI data. This study addresses these limitations by introducing a state-guided framework that preserves individual differences while ensuring biological interpretability through normative priors.

Methodology

Spatial and Temporal Constraints

The researchers implemented a two-stage automated framework to refine the extraction of connectivity patterns:

  1. Spatially Constrained ICA (scICA): This step identified 53 functional networks in each subject. By using a multi-objective optimization framework, the researchers ensured these networks aligned with known brain anatomy while preserving the unique spatial "flavor" of each participant.
  2. State-Constrained ICA (St-cICA): To ensure temporal consistency, the team first analyzed 1,000 healthy adults from the UK Biobank to identify five "canonical brain states"—normative templates of network interaction. These templates served as priors to guide the decomposition of dFNC data from the OASIS-3 and ADNI datasets.

Advanced Dynamic Metrics

A key innovation in this framework is the implementation of state-wise mean removal, which subtracts the stable background structure of a state to isolate "transient excursions"—the subtle, moment-to-moment fluctuations often lost in the noise. Furthermore, the study moved beyond the standard analysis of the "primary" (most dominant) state. By introducing metrics for the secondary state (the second-most expressed pattern at any given time), the researchers captured subdominant and transitional dynamics that provide a more holistic view of neural flexibility.

Key Findings

The "Stuck" Brain and Reduced Flexibility

The most significant finding was a marked reduction in neural flexibility among AD and MCI patients. These individuals exhibited significantly higher dwell times (remaining in a single state for longer periods) and a higher probability of self-transitions (looping back into the same state). In contrast, cognitively normal (CN) individuals displayed diverse and recurrent transitions between different brain states, reflecting a more adaptable and healthy neural system.

State-Specific Signatures

The analysis revealed specific connectivity patterns associated with disease progression:

  • State 1 (Sensory-Subcortical): Healthy controls showed stronger, more fluid connectivity between sensory and subcortical regions.
  • State 3 (Cognitive-Visual): AD/MCI patients spent significantly more time in this state, which is characterized by altered engagement of the Default Mode Network (DMN) and Cognitive Control (CC) networks.
  • Secondary Transitions: Analysis of subdominant states revealed that AD/MCI patients show reduced cross-state engagement even in their secondary network configurations, suggesting that neural rigidity in Alzheimer’s is a global phenomenon affecting multiple levels of the brain hierarchy.

Classification Performance

To validate the clinical utility of these temporal signatures, a classification model was trained on 6,960 dynamic features. The model achieved a mean Area Under the Curve (AUC) of 0.85 in distinguishing AD/MCI from healthy controls. The most informative features were concentrated in States 1, 2, and 3, particularly involving the DMN and CC networks, reinforcing their role as primary targets of neurodegeneration.

Discussion and Future Directions

The use of normative priors from the UK Biobank allowed the model to identify disease-specific deviations more effectively than purely data-driven methods. By showing that the Alzheimer’s brain becomes "stuck" in specific configurations, this research provides a clear temporal signature for the disease.

However, several challenges remain. The study utilized data from multiple sites with different scanner strengths (1.5T vs. 3T), introducing hardware-related noise. Additionally, the low signal-to-noise ratio of resting-state fMRI makes capturing subtle dynamic shifts difficult. Future work should focus on longitudinal data to determine if these temporal signatures can predict the specific timing of conversion from MCI to Alzheimer’s. Integrating these dFNC signatures with structural MRI or PET scans measuring amyloid and tau could further enhance diagnostic precision.

Summary updated

Abstract

Identifying robust neuroimaging biomarkers for Alzheimer's disease (AD) and mild cognitive impairment (MCI) is essential for early diagnosis and intervention. In this study, we introduce a novel, fully automated, guided dynamic functional connectivity (dFNC) framework for extracting multiple dynamic measures to distinguish MCI/AD from cognitively normal (CN) individuals. Resting-state fMRI data were used to extract subject-specific brain networks via spatially constrained independent component analysis (scICA), using a multi-objective optimization framework to ensure alignment with known functional networks while preserving individual variability. Using these components, dFNC was computed through a sliding-window approach. ICA was then applied to the concatenated dFNC matrices from the UK Biobank (UKBB) dataset to identify five canonical brain states, each representing a replicable, independent pattern of connectivity. These states served as biologically informed priors in a state-constrained ICA (St-cICA), which was applied to each subject in the combined OASIS-3 and ADNI datasets to guide individual-level decomposition and ensure interpretable connectivity states guided by state priors derived from the normative UKBB sample. St-cICA extracted subject-specific dFNC features and associated weighted timecourses. To characterize dFNC patterns, we computed metrics from the most strongly expressed (primary) state and introduced estimation of the second-most expressed (secondary) state at each time point, including dwell time, occupancy rate, and transition probabilities. Group comparisons using two-sample t-tests revealed widespread and significant alterations in AD/MCI compared to CN individuals. AD/MCI participants exhibited higher dwell times and increased self-transitions, indicating reduced neural flexibility and a tendency to remain in specific connectivity states. In contrast, CN individuals showed more diverse and recurrent transitions, reflecting greater adaptability. Secondary transitions revealed widespread selective switching in CN, whereas AD/MCI showed reduced cross-state engagement. A classification model trained on 6,960 dynamic features achieved strong performance in distinguishing AD/MCI from CN (mean AUC ≈ 0.85). These findings highlight the potential of guided dFNC as a biomarker framework for early-stage AD detection using resting-state fMRI.
BibTeX
@misc{zendehrouh2025state,
  author = {Elaheh Zendehrouh and Mohammad SE Sendi and Anees Abrol and Armin Iraji and Vince D. Calhoun},
  title = {{State Guided ICA of Functional Network Connectivity Reveals Temporal Signatures of Alzheimer's Disease}},
  year = {2025},
  doi = {10.1101/2025.09.23.25336175},
  howpublished = {medRxiv},
}
Figure