T1-to-DTI Registeration error with FLIRT

,
1

Hello experts,

I want to linearly register T1 to DTI with FLIRT, but no matter what rigid-body or affine registeration, there is always a bad alignment, especially in the prefrontal lobe and the posterior part of brain.I tried several ways to solve this error, but I failed again and again. I quite need your help!!!

Here is my code:

  1. flirt -in brain.nii.gz -ref mean_b0.nii.gz -out t12b0.nii.gz -omat t12b0.mat -dof 6 -cost mutualinfo

*brain.nii.gz is preprocessed by Freesurfer, it is a brain-extracted, bias-field-corrected T1 scan.
*mean_b0.nii.gz is extracted from preprocessed dti data by Mrtrix3, including noise removal, eddy correction, and bias field correction. I only have b-1000 and one PA direction.

**The reason why I do this step is for final registeration of atlas and dti, which is achieved via T1. **
Detailed information is as follow:

  1. Register T1 to MNI152 template linearly and nonlearly with FLIRT and FNIRT, then I acquire T1-to-MNI152 warping transformation.
  2. Compute the MNI152-to-T1 warping transformation by inverting the T1-to-MNI152 warping transformation, using FSL’s invwarp command.
  3. Register the atlas from MNI152 space to T1 space using FSL’s applywarp command, by applying the non-linear MNI152-toT1 warping transformation.
  4. Register the atlas from T1 space to diffusion space using FSL’s applywarp command.
    applywarp --ref=mean_b0.nii.gz --in=atlas --warp=warps_mni2t1.nii.gz --postmat=t12b0.mat --out=atlas_in_dti_space.nii.gz --interp=nn

By visualizing the output data, I found the image quality of the atlas registered to T1 (output at step 4) was good, but the step 5 showed bad output as step 1. So I guess the error is originated from the step 1, wrong t1-to-b0 matrix leads to wrong atlas-to-b0 registeration.

Here are the outputs of step1 and step5
output1: mean_b0.nii.gz (grey color) is overlaid by t12b0.nii.gz (red color), it seems two images are not aligned well.
image

output5: mean_b0.nii.gz (grey color) is overlaid by atals.nii.gz (colorful image)
image

Thank you in advance for any insights!

Best,
Regard

2

FLIRT is a linear registration tool, and your DWI image still has a lot of nonlinear geometric artifacts. I would make sure you provide diffusion images with sufficient information to allow removing these using Eddy/TOPUP. MRtrix will run this pipeline for you, but the solution will only be as good as your input data. Ensure that your data has a whole-sphere sampling (e.g. the default Siemens gradients are only half-sphere). You will want to make sure your input data has images with oppoosite phase encoding polarity (blip up, blip down), so that TOPUP can reduce the EPI distortions. You may find this FSL course slide deck relevant.

If you are unable to specify data that provides good inputs for TOPUP/Eddy, you may need to consider non-linear solutions for warping the atlas to your distorted images.