Non redundant connectivity (NRC) maps
Non-Redundant Connectivity (NRC) maps are summary maps of functional connectivity levels in the brain, estimated from functional Magnetic Resonance (fMRI) series. They are similar to weighted Global Brain Connectivity (GBC) maps (Cole et al. 2010), which are based on averages of correlations with each voxel. However, NRC is based on estimates of multiple correlations instead of bivariate correlations. To avoid dimensionality problems it applies the method of Supervised Principal Components (Bair et al. 2006).
A full description of the method is available in:
Non redundant functional brain connectivity in schizophrenia.
R. Salvador, R. Landín-Romero, M. Anguera, E. J. Canales-Rodríguez, J. Radua, A. Guerrero-Pedraza, S. Sarró, T. Maristany, P.J. McKenna, E. Pomarol-Clotet.
Brain Imaging Behav. 2017, 11:552-564.
Downloading the nrcon program
To run a NRC analysis you can use nrcon, a 64 bit linux executable that can be downloaded from the Download button located at the bottom of this page. The program is contained in the nrcon.zip file. Once nrcon.zip has been downloaded, it should be decompressed in a user chosen directory.
Before running nrcon a linux computer terminal should be opened on the user directory which contains the executable. Apart from the NRC image, the weighted GBC and the Degree Centrality (DC) images will be also calculated by nrcon. Specifically, the nrcon call should be followed by a single command line argument, which is the name of a plain text file containing the following information:
- Name of preprocessed and normalized 4D fMRI dataset. It should be a real image (of floats) in compressed NIFTI format (.nii.gz extension).
- Name of an integer Mask image (.nii.gz extension) with non zero values for the voxels to be considered in the analysis. It should have the same geometry as the 4D fMRI dataset.
- Correlation value or quantile value to threshold matrix of correlations (only for NRC and DC).
- Binary variable to indicate if the previous value is a correlation (0) or a quantile (1).
- Binary variable to indicate if we should consider the absolute values of correlations (only for GBC and DC). 0 = NO, 1 = YES.
- Amount of random subsampling to speed up NRC calculations (not applied in GBC or DC). 1 = no subsample, 2 = take (the equivalent of ) one every two voxels (in all three directions), 3 = take one in three, ...
- Fraction (value between 0 and 1) of variance to be kept from the components in the NRC analysis.
- Prefix to be added to the output images.
An example of the content of the input file could be:
/sda/nrc/fMRI_filtered.nii.gz /sda/nrc/mask1.nii.gz 0.6 1 1 3 0.95 prova
Output: It generates six files prefix_nrc.nii.gz (NRC map), prefix_gbc.nii.gz (GBC map), prefix_neigen.nii.gz (map of number of components used), prefix_std.nii.gz (map of voxel amplitudes), prefix_degree.nii.gz (DC map) and a log.txt file.
- nrcon is very computationally demanding. We are currently running it on images with a resolution lowered to a 4x4x4 mm voxel. In addition, there is the (highly recommended) option of random sub-sampling.
- It is assumed that the input images have been adequately preprocessed, correcting as much as possible for the different sources of noise. It is also assumed that the images have been time filtered to keep the patterns in the frequencies of interest (e.g. 0.01 – 0.1 Hz).
- Make sure that all non zero voxels in the mask contain data in the fMRI series. When running a group analysis make sure that the mask is valid for all individuals. The mask should also discard white matter areas.
- Running nrcon from its own directory will generate the output files in the same directory (e.g. type ./nrcon input.txt to run it in the same directory, where input.txt would be the text file with the input parameters). However, you can also call nrcon from any other directory by giving its full path. Thus, if nrcon is in the /media/sda/nrcon directory you can run it by calling /media/sda/nrcon/nrcon and the output will be created in the directory where you are (but don't forget to specify the input parameter file with the appropriate path as well).
For further help or to request the source code you can email me (Raymond Salvador): rsalvador at fidmag dot com
Cole MW, Pathak S, Schneider W (2010): Identifying the brain's most globally connected regions. NeuroImage. 49:3132-3148.
Bair E, Hastie T, Debashis P, Tibshirani R (2006): Prediction by Supervised Principal Components. Journal of the American Statistical Association. 101:119-137.