Dear community,
I have performed B0 fieldmap based correction for EPI images in a previous step and I would like to assess the quality of my correction. Therefore, I want to rigidly register an uncorrected and a corrected b0 volume to a T1w image. For this, I have ran the following commands for both b0 images:
# initiali registration
flirt -ref ${vrefbrain} -in ${vepi} -dof 6 -omat ${vout}_init.mat -pedir -2 -v
# registration with or without BBR cost function
flirt -ref ${vrefhead} -in ${vepi} -dof 6 -init ${vout}_init.mat -omat ${vout}.mat -pedir -2 -v
# apply the transformations
applywarp -i ${vepi} -r ${vrefhead} -o ${vout} --premat=${vout}.mat --interp=spline -v
where
${vrefhead} - whole-head T1w
${vrefbrain} - masked T1w
${vepi} - corrected/uncorrected b0
${vout} - my registered b0
I have three questions:
2. When performing BBR-based registration with 6 DOF as done in the epi_reg I get a very good output. When performing plain 6 DOF registration with corratio as the default cost function I get misalignment slice-wise with respect to the Z axis. Is the BBR registration still a rigid registration? The alignment looks very good even when performing it on the uncorrected b0, in most of the slices.
1. Why is the output of this registration not aligned properly when it comes to the Z axis?
3. Can the cost of these rigid registration functions be considered a quality check for the similarity between the reference and the input? If so, what is the range of the cost functions and what should be the ideal value for perfect resemblance? I am suspecting close to 1 means good similarity.
Thank you for your help!
|