Hi Ophélie,

a classical preprocessing for T1-weighted images should be bias (inhomogeneity) correction and could be, additionally, a noise reduction.

A bias correction is rather important, as the intensity differences for similar tissue types induce complications during registrations. Noise reduction depends on data quality. If your T1s are measured without too many fancy settings decreasing signal-to-noise, it won't be necessary. However, noisy images should be processed.

Bias correction is included in FSL's FAST, where the option -B outputs the bias-field corrected image, but you need BET first. If you are using BET with option -B, bias-field correction is already included in the skull-stripped output, but not in the original image still containing the skull (if needed for other purposes).

ANTs (Tustison&Avants) offers standalone tools for exactly these purposes:

N4BiasFieldCorrection and DenoiseImage

Hope this helps!

Best,

Martin

Do you need really the registration?

Normally in the acquisition the geometry is similar between DTI and T1. When you open both files in itksnap at same time and pass the cross about the images if you noticed displacement in any axis you need a coregistration between both sequences.

In order to use DWI in T1 space you need

FDT diffusion from FSL.

1. Eddy current correction

2. Bedpostx

3. Probractkx from ROIs or masks (this is the tractography). The next step is only if you want...

4. For registration tool in FTD-diffusion you need T1 with BET and a file called nodif_brain from bedpostx folder. Sometimes the files looks like nodif_brain_mask, change it. Is the the mask from DWI.

5. In folder of bedpostx there is s subfolder xfm with transforms

6. Use flirt and the tool apply to transforms in utils in order to transform.

7. The result are the paths of tractography in T1 space.

regards!

Hello Ophélie Martinie ,

Yes you should correct your T1 images before doing anything with your DWI. Here are my recommendations, please keep in mind that this varies from person to person but the below mentioned steps I have seen most common in diffusion imaging papers.

1. Fixing the orientation i.e. Moving into standard space or center your brain to (0,0,0). You can use FSL's fslorient2std command to re-orient.

2. Denoising

3. Gibbs ringing correction

4. Bias field correction

5. Skull stripping

In case if your T1 data is not in isotropic voxel you may also need to resample it too.