ANACOM2: HOW DOES IT WORK?


In Summary

AnaCOM2 is based on a previously published methods (Kinkingnéhun et al. 2007) and aims at establishing structure–function relationships. AnaCOM2 is using FMRIB software library (FSL) to interact with R software package.
Typically, the software compare neuropsychological scores between lesioned patients and controls to determine which brain area critically affect the performance.


!!! IMPORTANT !!!

You need to have R installed if you want to run this module.
To install R please visit : https://www.r-project.org/ click on « CRAN » at the top left, choose your preferred mirror and download the R version corresponding to your system.


*Please don’t use values lower or equal to zero in scores and use commas (not semicolon) in your csv files.



RESULTS


clusters.csv: For each given cluster of voxels, nb_disco(nb_spared if you compare only spared patients and controls) is the number of patients whose disconnectome involves that cluster of voxels. (kw_)pval is the pvalue of the test, (kw_)stat is the value of the statistical test (H for Kruskal-Wallis, U for Mann-Whitney, T for T-test, D for Kolmogorov-Smirnov) and (kw_)holm is the column containing the pvalues corrected for multiple comparison by the Bonferroni-Holm algorithm.


With Kruskal-Wallis option :

Without











In the example above, the uncorrected pvalues are lower than 0.05 but the BH correction, at some point, is higher than 0.05, hence these clusters did not survive BH correction for multiple comparison.

Note that optimising the overlap of lesions (and the number of voxels per clusters will reduce the total number of comparison and decrease the severity of the correction.


warnings.csv: Statistical assumptions are not always respected in every cluster. warning.cvs will report the cluster for which statistical assumptions were violated. For instance, the wilcoxon test will not be able to compute an exact value when a patient has the same score as the published normative value.


patients_info.csv: contains the name of patient’s files and the scores of patients who have disconnections within clusters.


clusters_holm.nii.gz is a nifti map in the MNI152 indicating Bonferroni-Holm (BH) corrected pvalues after the post_hoc test.


clusters.nii.gz is a nifti map in the MNI152 indicating uncorrected pvalues (If you chose the Kruskal Wallis test, the values are only for clusters that passed the Kruskal Wallis test).


kruskal_holm_clusters.nii.gz  is a nifti map in the MNI152 indicating Bonferroni-Holm corrected pvalues after the Kruskal-Wallis test.


kruskal_clusters.nii.gz is a nifti map in the MNI152 indicating the pvalue in the Kruskal-Wallis test before Bonferroni-Holm correction.


When activating the ‘keep temporary files’ option, the following files will be saved in anacomTemporaryFiles :


the folder anacomClustersDir contains all binary « cluster » images.

maskedMeanValMap.nii.gz is a nifti map in the MNI152 indicating the average score.

maskedStd.nii.gz  is a nifti map in the MNI152 indicating standard deviations.


PAPER METHOD SECTION (feel free to edit or copy and paste)


Disconnectome-symtoms mapping was assessed using AnaCOM2 as part of BCBtoolkit (Foulon et al. 2018). AnaCOM2 is a cluster-based lesion approach allowing to identify the brain lesions locations that are associated with a given deficit, i.e. the regions that are critical for a given function. Compared to standard VLSM (Bates et al. 2003), AnaCOM2 regroup voxels with the same distribution of neuropsychological scores into clusters of voxels. Additionally, AnaCOM2 performs comparisons between patients and controls as a first step in order to avoid drastic reduction of statistical power when two or more non-overlapping areas are responsible for patients reduced performance (Kinkingnéhun et al. 2007). AnaCOM2 resulted in a statistical map revealing for each cluster the significance of a deficit of patients at a given task, compared to controls. P-values are Bonferroni-Holm corrected for multiple comparisons.