iCAPs Toolbox

Posted by admin at 12:30 PM on Aug 18, 2018

Share:


Repository

The iCAPs Toolbox is released and maintained on the c4science.ch git infrastructure:

https://c4science.ch/source/iCAPs/

A nightly build (zip) can also be downloaded from here (approx. 2GB)

About innovation-driven Co-Activation Patterns (iCAPs)

Dynamics of resting-state functional magnetic resonance imaging (fMRI) provide a new window onto the organizational principles of brain function. Using state-of-the-art sparsity-pursuing deconvolution, termed "Total Activation" (TA), we extract innovation-driven co-activation patterns (iCAPs) from resting-state fMRI. The iCAPs' maps are spatially overlapping and their sustained-activity signals temporally overlapping. Decomposing resting-state fMRI using iCAPs reveals the rich spatiotemporal structure of functional components that dynamically assemble known resting-state networks. The temporal overlap between iCAPs is substantial and is consistent with their behavioral profiles. In contrast to conventional connectivity analysis, which suggests a negative correlation between fluctuations in the default-mode network (DMN) and task-positive networks, iCAPs show evidence for two DMN-related subnetworks consisting the posterior cingulate cortex that differentially interact with the attention network. This methodology demonstrates how fMRI resting state carries complex interactions between large-scale functional networks, a property that can be approached by decomposing the data into spatially and temporally overlapping building blocks using iCAPs.

References

Using iCAPs to reveal functional networks
Total activation deconvolution
Transient-informed regression
  • D. Zöller, T.A. Bolton, F. I. Karahanoglu, S. Eliez, M. Schaer, D. Van De Ville,  "Robust Recovery of Temporal Overlap Between Network Activity Using Transient-Informed Spatiotemporal Regression", IEEE Transactions on Medical Imaging, in press.
More background about dynamic functional connectivity


Legacy code

The following source code was released earlier, but it is not maintained anymore.

Sparsity-driven deconvolution of fMRI timecourses

The Total Activation toolbox is implemented by F. Isik Karahanoglu. The main contribution of TA is to combine generalized total-variation regularization (in time) with structured sparsity (in space) for recovering fMRI activity-inducing signals without specification of the paradigm. The Matlab code can be downloaded here.

Generalized total variation

Total Activation is based on the 'generalized total variation' framework that has been proposed in the IEEE Transactions on Signal Processing (2011) paper. The Matlab code allows to regenerate Figure 2.

The following software should be available on your system: (necessary tools are included in the package)

  • Before installing TA Toolbox, check that Matlab is correctly installed
  • Download the zip
  • Extract the archive
  • Add the toolbox to Matlab path

Here we demonstrate the use of TA toolbox. We show how to make the basic settings for running TA algorithm. A simple synthetic dataset (Neuroimage paper) is included in the toolbox (test_data/phantom1db).

1. Data Preprocessing

Generally the functional data is preprocessed (realignment, smoothing, etc.) before TA. Normalization of the functional data into a common space (MNI/Talairach) is not necessary, however, the atlas and the functional data have to be matched. We use atlas wrapper to convert the structural atlas to functional space (from IBASPM and Jonas Richiardi preprocessing pipeline). Atlas should be a volume whose intensity consists of postive numbers inside the atlas and zero otherwise.

2. Running TA

The algorithm is run through a main (main.m) function in the toolbox. First, the toolbox should be added into the matlab path. Then, to start the process user needs to input the data path s and parameters (MyInputs.m or MyInputsTest.m for synthetic data).


The user should enter the path to the data, atlas and the results and then enter the type of analysis he/she wants to perform. TA has various temporal and spatial regularization options:

  • METHOD_TEMP: "S", for spike-like activity-inducing signals, "B" for block-like activity-inducing signals and "W" for smooth activity-inducing signals
  • METHOD_SPAT: "NO" for no spatial smoothing, "STRSPR" for structured-sparsity, "Tik" for l_2-norm smoothing)
  • DETRENDING: "normalize" does only z-normalization to each time course, "dct" performs first Discrete Cosine Transform for low frequency drifts and then performs z-normalization. There are also some parameters tuned by the user such as type of hemodynamic response function (HRF), repetition time (TR) etc.Then, the algorithm is run by executing main.m.

    3. Results

    The results are saved in "path_results" folder defined in "MyInputs.m".
  • The whole workspace is saved into a mat-file. The format of the file is "OUTS_METHODTEMP_LAMBDACOEF_METHODSPATIAL_SPATIALLAMBDACOEF_NAMEOFDATA_DATE.mat"
  • 4 NIFTI-files are saved:
    "TCN_.nii" = NORMALIZED-DETRENDED DATA
    "TC_OUT_.nii" = RECOVERED ACTIVITY-RELATED SIGNALS
    "TC_D_OUT_.nii" = RECOVERED ACTIVITY-INDUCING SIGNALS
    "TC_D2_OUT_.nii" = RECOVERED INNOVATION SIGNALSThe results are 4D datasets and can be visualized with an appropriate data visualization tool (e.g., AFNI).
    Here, we show TA analysis results for one subject performing 10 visual stimuli (Neuroimage paper).