Dynamic FC Patterns Analysis Methods

Date: January 31, 2024 10:04 AM

  1. Sliding Window Correlation Analysis:
    • One of the most common approaches, where the time series data is segmented into overlapping or non-overlapping windows.
    • Functional connectivity is calculated separately for each window using measures like Pearson correlation or partial correlation.
    • Provides insights into the temporal dynamics of connectivity patterns.
  2. Dynamic Conditional Correlation Models:
    • Extends traditional correlation analysis by allowing the correlation coefficient to vary over time.
    • Models the temporal evolution of connectivity using dynamic linear models or time-varying autoregressive models.
    • Captures changes in connectivity strength and directionality across different time points.
  3. Dynamic Graphical Models:
    • Represent brain networks as graphs where nodes correspond to brain regions and edges represent functional connections.
    • Employ dynamic graphical models such as dynamic Bayesian networks or dynamic graphical LASSO to estimate time-varying connectivity patterns.
    • Allows for the identification of evolving network structures and connectivity dynamics.
  4. Hidden Markov Models (HMM):
    • Model the brain’s dynamic connectivity states as a sequence of hidden states governed by transition probabilities.
    • Estimate the most likely sequence of hidden states given the observed fMRI data using the Viterbi algorithm or forward-backward algorithm.
    • Provides a framework for characterizing discrete brain states and transitions between them.
  5. Dynamic Connectivity Regression:
    • Incorporates external variables or experimental conditions as regressors to investigate how they modulate dynamic connectivity patterns.
    • Fit a regression model to estimate the relationship between covariates and time-varying connectivity metrics.
    • Allows for the investigation of task-evoked or stimulus-dependent changes in functional connectivity.
  6. Time-Frequency Analysis:
    • Decompose the fMRI time series data into different frequency bands using techniques like wavelet transform or Hilbert-Huang transform.
    • Estimate dynamic connectivity within each frequency band to capture oscillatory dynamics of brain networks.
    • Reveals frequency-specific changes in connectivity patterns over time.
  7. Functional Network Dynamics Analysis:
    • Characterize the dynamics of functional brain networks using graph-theoretical measures such as modularity, network flexibility, or resilience.
    • Investigate how network properties evolve over time and their relationship to cognitive or behavioral states.
    • Provides insights into the reconfiguration of functional brain networks during different tasks or mental states.
  8. Deep Learning Approaches:
    • Utilize deep learning architectures such as recurrent neural networks (RNNs) or convolutional neural networks (CNNs) to model the temporal dynamics of functional connectivity directly from fMRI data.
    • Train neural networks to predict future connectivity patterns based on past observations, enabling the capture of complex temporal dependencies.