Your company here — click to reach over 10,000 unique daily visitors

tend-epireg - Man Page

register diffusion-weighted echo-planar images


tend epireg -i <dwi0 dwi1 ...> -g <grads> [-r <reference>] [-nv] [-p] [-bw <x,y blur>] [-t <DWI thresh>] [-ncc] [-f <fit frac>] [-k <kernel>] [-s <start #>] [-o <output/prefix>]


Register diffusion-weighted echo-planar images. This registration corrects the shear, scale, and translate along the phase encoding direction (assumed to be the Y (second) axis of the image) caused by eddy currents from the diffusion-encoding gradients with echo-planar imaging. The method is based on calculating moments of segmented images, where the segmentation is a simple procedure based on blurring (optional), thresholding and connected component analysis. The registered DWIs are resampled with the chosen kernel, with the separate DWIs stacked along axis 0.


-i <dwi0 dwi1 ...>

all the diffusion-weighted images (DWIs), as separate 3D nrrds, OR: one 4D nrrd of all DWIs stacked along axis 0 (1 or more nrrds)

-g <grads>

array of gradient directions, in the same order as the associated DWIs were given to “-i”, OR-g kvp” signifies that gradient directions should be read from the key/value pairs of the DWI (string)

-r <reference>

which of the DW volumes (zero-based numbering) should be used as the standard, to which all other images are transformed. Using -1 (the default) means that 9 intrinsic parameters governing the relationship between the gradient direction and the resulting distortion are estimated and fitted, ensuring good registration with the non-diffusion-weighted T2 image (which is never explicitly used in registration). Otherwise, by picking a specific DWI, no distortion parameter estimation is done. (int); default: “-1


turn OFF verbose mode, and have no idea what stage processing is at.


save out intermediate steps of processing

-bw <x,y blur>

standard deviations in X and Y directions of gaussian filter used to blur the DWIs prior to doing segmentation. This blurring does not effect the final resampling of registered DWIs. Use “0.0 0.0” to say “no blurring” (2 floats); default: “1.0 2.0

-t <DWI thresh>

Threshold value to use on DWIs, to do initial separation of brain and non-brain. By default, the threshold is determined automatically by histogram analysis. (float)


do NOT do connected component (CC) analysis, after thresholding and before moment calculation. Doing CC analysis usually gives better results because it converts the thresholding output into something much closer to a real segmentation

-f <fit frac>

(only meaningful with “-r -1”) When doing linear fitting of the intrinsic distortion parameters, it is good to ignore the slices for which the segmentation was poor. A heuristic is used to rank the slices according to segmentation quality. This option controls how many of the (best) slices contribute to the fitting. Use “0” to disable distortion parameter fitting. (float) default: “0.70

-k <kernel>

kernel for resampling DWIs along the phase-encoding direction during final registration stage; default: “cubic:0,0.5

-s <start #>

first number to use in numbered sequence of output files. (int); default: “1

-o <output/prefix #>

For separate 3D DWI volume inputs: prefix for output filenames; will save out one (registered) DWI for each input DWI, using the same type as the input. OR: For single 4D DWI input: output file name. (string); default: “-

See Also


Referenced By


April 2021