The Time-varying Dynamic Network Model for Extracting the Dyanmic Resting State Functional Connectivity

This is repo is to implement a time-varying dynamic network (TVDN) method to extract the resting state functional connectivity (RSFC) from both functional magnetic resonance (fMRI) and the magnetoencephalography (MEG) imaging.

If you encounter any problems when using the repo, be free to contact us ([email protected])

Installation

• Enviroment:
  • Python 3
  • R

git clone https://github.com/JINhuaqing/TVDN.git

For dependencies in Python, run

pip install -r requirements.txt

signal pakcage in R is also needed

install.packages("signal")

Examples

You may find demos for MEG and fMRI data in demo folder.

# import the class for detection
from pyTVDN import TVDNDetect
from pathlib import Path
import numpy as np


# Construnct the detection object
Detection = TVDNDetect(Ymat=Ymat, saveDir="../results", dataType="MEG", fName="subj2", r=8, kappa=2.95, freq=60)
# Ymat, d x n maitrx, where d is the number of sequences
# r: the number of rank used for detection
#    if r is decimal, the rank is the number of eigen values such that account for 100r% of the variance.
#    if r is None, r=0.8
# saveDir, the path to save the results. If not specified, the results will not be saved
# dataType, "MEG" or "fMRI". Different dataTypes have different default parameters. You may leave it blank
# All other parameters have default values, but you can still specify here.
# For the meaning of the other parameters, you can refer to the source code


# When n is large, the detection would take a while
# To reduce the computaion burden, we provide a screening step 
# to obain the candidate point set
# Screening is optional
Detection.Screening(wh=10)
# wh: the screening window size

# Run detection
Detection()

# You can tune kappa, the parameters for MBIC penalty term kappa by the reconstructed errors
# However, it does not always work very well
kappas = np.linspace(2.5, 3, 100)
Detection.TuningKappa(kappas)

# You can specify the number of change points you want via provide the argument `numChg`, then the `UpdateEcpts` will update the current estimated change point set accordingly
Detection.UpdateEcpts(numChg=12)
# If You don't specify the number of change points you want, then the `UpdateEcpts` will update the current estimated change point set based on optimal kappa values by TuningKappas function
Detection.UpdateEcpts()

# Plot the detection results
Detection.PlotEcpts(saveFigPath="detectionResults.jpg")
# save figure if you specify the `saveFigPath`
# You can specify the GT parameter to draw the ground truth for comparison

# Plot the reconstruncted Ymat.
Detection.PlotRecCurve(idxs=[43, 45, 59], saveFigPath="recCurve.jpg")
# idxs: The indices of sequences to plot

# Plot the eigen values curve
Detection.PlotEigenCurve()
# save figure if you specify the `saveFigPath`

# print the results
print(Detection)