|
Notes:
Source code (C/C++) for command line tool; has been compiled and tested with WindowsXP and Linux. Also included is WindowsXP executable and simple GUI for creating "input" files and starting command line tool.
Changes:
Changes in valmap, version 2.0
1. Added permutation cluster analysis
Input files now require you to specify a cluster threshold, and write out nifti files of clusters above the threshold (e.g., the largest cluster of voxels above p=0.01 will be labeled 1, the second largest cluster of voxels above p=0.01 will be labeled 2, etc.). If running permutations, valmap will also keep track of the maximum cluster at each permutation, so that permutation cluster thresholds can also be determined. Values at all permutations (as well as cutoff values for permutation multiple comparisons correction) are written to the output file.
2. Creates uncorrected p-value files, based on a permutation distribution (i.e., not the t or F distribution).
A statistician colleague was concerned that the statistical maps (t or F maps) were not following a t or F distribution, and requested this modification. So, when running permutations, a p-value map is also created, based on the distribution of statistics at each voxel. A p-value map is written, where the p-values are -1*ln(p) (so smaller p-values have larger intensities in the file). NOTE: the lowest p-value attainable is limited by the number of permutation (e.g., with 1000 permutations, the smallest p-value that can be calculated is p=0.001). If a t-value from the true model is larger than the maximum t-value from any permutation, then the p-value is set to 9/(num_perm*10) (i.e., a value <1/num_perm)...for 1000 permutations the largest value in the p-value file will be 7.01 (-1*ln(0.0009)).
3. Major bug fixed with spatially varying independent variable.
I found a bug that occurred when a spatially varying independent variable was used, and the value at a voxel was zero across all observations
|
