Functional Brain Age Acceleration from Dynamic and Static Connectivity Predicts Working Memory and Attention Deficits in Schizophrenia

Sabrina J. Edwards-Swart, Bradley T. Baker, Daniel H. Mathalon, Judith M Ford, Adrian Preda, Theo G.m. van Erp, Godfrey D. Pearlson, Jessica Anderson Turner, Vince D. Calhoun, Mohammad Se SendimedRxiv 2025 · DOI

TL;DR

Researchers have developed a "brain age" clock using dynamic functional connectivity (dFNC) that outperforms traditional static methods in predicting cognitive decline. By measuring the gap between a person's predicted brain age and their actual age, the study identifies a powerful new biomarker for the working memory and attention deficits central to schizophrenia.

What Problem Does This Paper Address?

Schizophrenia is often defined by its clinical symptoms (like hallucinations), but its most debilitating aspect is frequently cognitive impairment—specifically in attention and working memory. While we know these patients often exhibit "accelerated brain aging," most research has relied on structural MRI (anatomy) or static functional connectivity (the "average" communication between brain regions).

The problem? The brain is not static. It is a highly dynamic system that reconfigures itself in milliseconds. Static models "blur" these temporal fluctuations, potentially missing the very neural volatility that drives cognitive dysfunction. This paper asks: Can we use the dynamics of brain communication to better quantify brain aging and predict cognitive deficits in schizophrenia?

Methods

The authors employed a sophisticated deep-learning pipeline to compare two modalities:

  1. Static Functional Network Connectivity (sFNC): Measuring the Pearson correlation between brain regions across an entire fMRI scan.
  2. Dynamic Functional Network Connectivity (dFNC): Using a sliding-window approach to see how these correlations change over time.

To process this data, they used NeuroMark, an automated ICA pipeline, to extract 53 independent components (brain regions) categorized into seven functional networks.

  • The Models: For static data, they used a Graph Convolutional Network (GCN). For the dynamic, time-series data, they utilized a bidirectional Long Short-Term Memory (biLSTM) network, which is specifically designed to "remember" and process sequences of data.
  • The Metric: They calculated the Brain Age Gap (BAG)—the difference between the model's predicted age and the participant's chronological age.

Datasets

The study is notable for its massive scale, leveraging some of the largest fMRI datasets available:

  • Training/Validation: A total of 22,569 resting-state scans from 17,522 healthy adults across the UK Biobank (UKBB) and the Human Connectome Project (HCP/HCP-Aging).
  • Independent Clinical Testing: The FBIRN dataset, consisting of 153 participants (including individuals with schizophrenia and healthy controls) used to test the link between BAG and cognitive performance.

Key Insights and Ideas

  • Dynamics Matter: The "aha moment" here is that dFNC-based brain age gaps were more sensitive to cognitive deficits than static ones. This suggests that the brain's ability to flexibly reconfigure its networks is a more critical indicator of cognitive health than its "average" wiring.
  • Network-Specific Aging: Accelerated aging isn't uniform. The study found that "older-looking" subcortical and sensorimotor networks were the strongest predictors of attention deficits.
  • The Schizophrenia "Signature": While both healthy controls and patients show age-related decline, individuals with schizophrenia exhibit a significantly wider BAG. This "accelerated clock" correlates directly with how poorly they perform on memory and attention tasks.

Key Results

  • High Accuracy: Both models were remarkably good at predicting age in healthy individuals (sFNC: $r=0.8755$; dFNC: $r=0.8675$).
  • Cognitive Correlation: Wide-brain BAGs showed strong negative associations with attention vigilance (dFNC: $r=–0.2715$, $p=0.0016$) and working memory (dFNC: $r=–0.2508$, $p=0.0064$).
  • Sub-network Sensitivity: Sub-network BAGs (subBAGs) within the subcortical, sensorimotor, and default mode networks (DMN) were robust predictors of deficits, with dFNC-derived subBAGs showing the most significant effects ($p < 0.01$).

Limitations and Future Work

  • The "Young Adult" Gap: The models performed less accurately for individuals under 38 years old, leading the authors to restrict their clinical analysis to older participants. Future work needs more diverse training data to capture the full adult lifespan.
  • Window Size: The dFNC analysis relies on a specific "window size" (40 TRs). The authors acknowledge that different window sizes might capture different temporal scales of brain dynamics.
  • Environmental Factors: The current models do not account for lifestyle factors like diet, exercise, or socioeconomic status, which are known to influence both brain aging and cognitive performance.
  • Clinical Translation: While a powerful biomarker, the next step is determining if these dFNC BAGs can be used to track the success of specific treatments or cognitive remediation therapies.

Summary updated

Abstract

Our findings establish dFNC-based BAGs as a sensitive and clinically relevant biomarker of cognitive impairment in schizophrenia, outperforming sFNC and highlighting the translational potential of dynamic connectivity for precision diagnosis and treatment.
BibTeX
@misc{edwardsswart2025functional,
  author = {Sabrina J. Edwards-Swart and Bradley T. Baker and Daniel H. Mathalon and Judith M Ford and Adrian Preda and Theo G.M. van Erp and Godfrey D. Pearlson and Jessica Anderson Turner and Vince D. Calhoun and Mohammad SE Sendi},
  title = {{Functional Brain Age Acceleration from Dynamic and Static Connectivity Predicts Working Memory and Attention Deficits in Schizophrenia}},
  year = {2025},
  doi = {10.1101/2025.07.21.25331956},
  howpublished = {medRxiv},
}
Figure