Tracking effect of using upsampled DWI data

Hello experts,

I noted unexpected tracking results from using upsampled DWI data.

Raw DWI: 104 x 104 x 54 x 67; 2.3 isotropic
Upsampled DWI: 256 x 256 x120 x 67; 0.9 isotropic (using mrresize, set dimension size, select linear interpolation)
CSD and tensor estimated from the upsampled DWI; rigid registration of b0 with T1-based brain mask.

Comparing corticospinal tracking between using Raw DWI (Raw_CST) vs upsampled DWI (Upsampled_CST):

I kept all default parameters the same, and using the same manually defined ROIs: seed PLIC, include CP and pons, exclude midline and retrolenticular IC):

I noted the Upsampled_CST result misses a lot of cortical/subcortical projections (when visually comparing to the Raw_CST)

Given the use of upsampled DWI reduces the FOD interpolation errors, I would think it should produce “better” results than using raw DWI…

Many thanks

Joseph

Hi Joseph,
When you say ‘default parameters’, do you mean you fixed the step size and angle etc to be the same for both resolutions?

FYI, this paper comes to mind. Not sure if you have seen it. You may want to try cubic interpolation when up-sampling.
Cheers,
Dave

Hi, Dave

Fixed step size and angle etc for both resolutions (default step size: 0.5voxel size…etc).

Re-upsample the DWI data now using the cubic (default) interpolation. ( I just noted that’s what u recommended in the FBA section)

Will post an update soon

Thanks for the prompt reply,

Joseph

I think the biggest effect here is that of step size, as per your previous question. Upsampling means the nominal voxel size provided to tckgen is half what it was, so you’ll be using half the step size on essentially the same data. This will reduce the dispersion during tracking, and probably makes it less likely to branch, etc. This might explain the algorithm’s reduced ability to follow all the hard to track subcortical projections.

I had a brief look into the effect of upsampling on tractography in my 2012 MRtrix paper, and found it made very little difference - when using the same step size (the default in MRtrix 0.2 was a hard 0.2mm, independent of the voxel size). So I reckon you should try setting the step size to the same as would have been used with the original data. You may also need to modify the angle constraint to match (should be 45° by default in iFOD2). That would be a fairer comparison if you’re trying to evaluate the effects of upsampling alone.

BTW, all this relates to the use of probabilistic algorithms. With deterministic algorithms, you may well find that upsampling reduces interpolation errors. Although I’m not entirely convinced that the results will necessarily be noticeably better in practice…

Hello again,

I have tried the following (combining both Dave and Donald’s advice):

  1. Re-upsample DWI data using cubic interpolation.
  2. cubic interpolation + upscale by a factor of 2 ( rather than setting voxel dimensions)
  3. Using the upsampled DWI data from 1 and 2, repeat tracking - but setting the step size and angle the same as it was for the original data (step 1.15, angle 45 degree).

Results:

  1. Cubic interpolation alone still result in poor cortical/subcortical projections of the track
  2. Cubic interpolation + upscale by factor of 2: results in more cortical/subcortical projections than in 1, but still nowhere close to using the raw DWI data.
  3. Using the original step size and angle settings combining with 1: no visible improvement
  4. Using the original step size and angle settings combining with 2: appears to give the best result, in terms of resembling those reconstructed using the raw DWI data, although most of the projections missing are of these from the crossing fibre region (i.e. the lateral projections of the corticospinal tract).

Joseph