NeuroMark dFNC: Transferable Dynamic-State Templates for Subject-Specific and Frequency-Resolved Chronnectomics

The research team has developed NeuroMark dFNC, a method that leverages transferable dynamic-state templates to facilitate subject-specific and frequency-resolved chronnectomics. This technique extends the NeuroMark principle of prior-guided network estimation to dynamic connectivity, enabling the estimation of corresponding subject-specific states using reproducible reference states as priors.

The reference-state templates were generated from the UK Biobank dataset using windowed functional network connectivity and time-frequency moments (TFMs). These templates were evaluated for reproducibility using split-sample approaches, including k-means, independent component analysis (ICA), and principal component analysis (PCA) template-generation methods.

Once the templates were established, constrained ICA was employed to estimate subject-specific state patterns and continuous loadings within independent FBIRN schizophrenia and ADNI Alzheimer's disease cohorts. The ICA loadings were calibrated by their contribution to reconstructing the original feature matrix. The analysis then calculated fractional occupancy and conditional mean dwell time from the state with the largest absolute calibrated loading at each observation point.

By utilizing predefined templates, the need for dataset-specific state discovery and post-hoc state matching was eliminated, streamlining the analysis process.

In both clinical cohorts, the NeuroMark dFNC approach identified state-dependent, distributed edgewise differences and variations in selected temporal summaries. The frequency-resolved findings were nominal and varied across states and moment-frequency bins, treated as exploratory in nature. Overall, NeuroMark dFNC offers a common state space for cross-cohort comparisons, while preserving subject-specific connectivity patterns and continuous state expression.