Unringing? Bias field correction --HCP Diffusion data preprocessing

Dear Mrtrix experts,
I’m working the diffusion MRI data from the HCP projects. I’m wondering whether you have a most updated pipeline for the minimally preprocessed data. It will be great to have your official suggestions.

If the pipeline has not been developed yet, do you recommend the Unringing and Bias field correction following the minimal preprocessing pipeline from HCP?

Thanks.

Best,
Jinglei

Hi @jingleil,

I’m not one of the developers, so you might want to double check this. I wouldn’t unring the data if using the minimal processing data, this step should be done before the distortion and eddy current corrections.

Regarding the bias field correction (using dwibiascorrect), this step is completelly optional. The bias field correction will be perform later using mtnormalise.

You could just use the minimally preprocessed data to calculate the response function (with the new dhollander algorithm), then calculate the FODs (note that if you want to do group studies, you need to use the average response function), after that mtnormalise and finally tckgen. Note that if you want to run ACT, you have new algorithms to generate the 5TT, gif and hsvs). Finally, you could use SIFT2. I hope this helps.

Best regards,

Manuel

1 Like

Hi Manuel
This is absolutely very helpful. Thank you for the suggestions. I will follow your advice unless the developers have different opinions.

Best,
Jinglei

Hello @jingleil,

Please see the thread below for my pipeline (for generating a connectome). I am in the MRtrix development team, but those are only my suggestions:

Please ask me if you have any questions about the pipeline. Also notice that it was developed for an older version of MRtrix (not the one just released), and I didn’t test for compataility.

Best,
Oren

1 Like

Hi Jinglei,

If the pipeline has not been developed yet, do you recommend the Unringing and Bias field correction following the minimal preprocessing pipeline from HCP?

Unringing should absolutely not be done on minimally processed data. As soon as any sort of image interpolation step is performed (as occurs for e.g. motion / eddy current / susceptibility field correction), the mathematical theory behind the manifestation of the effect and its estimation and correction no longer applies.

Bias field correction can be done on minimally processed data using either method, and generally mtnormalise is recommended over dwibiascorrect. mtnormalise can’t do quite as good a job on HCP data as it can on conventional data because of the gradient non-linearity effects present in the data, but it will actually try to somewhat correct for these.

Rob

1 Like

Hi Oren
Thanks for sharing your pipeline! It is very helpful.

Best,
Jinglei

Hi Rob
This closed my question and you also answered the other question of me posted elsewhere.
Thank you.

Best,
Jinglei

Hey @jingleil,

Just a small warning on this topic: while it might appear mtnormalise compensates a bit for gradient non-linearity effects, it’s not the appropriate way to deal with these, or a solution to it generally speaking. If you use the minimally preprocessed HCP data, I’d advise to practice caution when performing experiments that rely on absolute intensities being comparable across/between different brain regions or tracts within subjects. This even more so when using large numbers of subjects: due to the high quality and large numbers of subjects available in the HCP dataset, there is a realistic chance of pulling out even small effects. I still recommend to run mtnormalise regardless on the HCP data, at least for the other intensity inhomogeneities in this dataset. Just don’t expect it to correct the effects of gradient non-linearities.

Cheers & take care,
Thijs