Output of utils.density_map() using tck file is different than MRTrix

[davidrs06]

Description

I generated a tractogram using MRTrix, and saved it in tck format. I also generated a tract density image using MRTrix's tckmap -template DWI.nii.gz tractogram.tck tdi.nii.gz where DWI.nii.gz is the input diffusion volume.

Issue

When I load the tractogram using nibabel and compute the tdi image using dipy, I get a map that has quite less fiber count per voxel (between a few to more than 200 for some regions, with a mean around 50 streamlines less). Is this expected ? I imagine MRTrix and Dipy might compute differently how voxels are "visited" or not by the streamlines, but I wondered if someone knew about another explanation, and which map is more accurate ?

Code

Here is the code I use to compute the map using dipy:

import nibabel as nib
from dipy.tracking import utils

DWI_img = nib.load('DWI.nii.gz')
tractogram = nib.streamlines.load('tractogram.tck')
tdi = utils.density_map(tractogram.streamlines,DWI_img.affine,DWI_img.shape[:3])
nib.save(nib.Nifti1Image(tdi,DWI_img.affine),'/tmp/tdi_dipy.nii.gz')

I'm using dipy v1.0.0, python 3.7.4 on a mac (Darwin-17.5.0-x86_64-i386-64bit)

Thanks in advance for any feedback !

Cheers,
David

[Garyfallidis]

Hi David (@davidrs06) ! As far as I know the method implemented in DIPY was not made to replicate the behavior of the mrtrix command. The algorithms used are possibly different. I think you should look inside the code to understand how things are calculated. There is another function that you could look into called track_counts. @jchoude has contributed an extension of this command that also accounts for the case were the streamlines may have points with variable step size.

[davidrs06]

Hi Ele :). Indeed, it's not so much of an issue, I was more wondering if the difference was intended or not. Thanks for the answer, I'll have a look at the code.
Take care !
David