Hi all,

I have a large dataset (n>100) on which I want to apply DTIPrep for faster QC. Multi-shell diffusion data were acquired with the following scheme: b-values of 0,700,1000 and 2000 s/mm2 at 7,20,32 and 60 directions respectively. I used DWIConvert and the FSLtoNrrd flag to convert my data into the right format for DTIPrep. The protocol includes 2mm translation and 0.5 degrees rotation thresholds.

1) I was wondering if anyone has experience using DTIPrep on multi-shell data? I noticed that during at the diffusion bcheck step, the bvalue is set at 2000 - does this mean that DTIprep assumes all gradients were acquired with a single bvalue? And to what extend is this an issue?

2) I applied DTIPrep on a subset of subjects and after looking at some of the reports I noticed this error during the bvalue check step:
"DWMRI_gradient_0000 mismatch!". This error is present in most of the gradients (regardless of bvalue) in almost all subjects. To what extend is this problematic for further outlier detection steps?

What I also noticed was that in some subjects DTIPrep systematically only evaluated included gradients up to a certain number (59 in our dataset) without going through the entire dataset (119 gradients). Has anyone experienced this issue before or any other possible explanations?


Many thanks in advance for the feedback and don't hesitate to ask me for more information!

Best,
Maria

Hi Maria,

1) What version of DTIPrep and DWIConvert are you using? Older versions tend to not preserve all the b-values. I use https://github.com/pnlbwh/conversion/tree/master/conversion to do my FSLtoNRRD conversions.

To my knowledge, the diffusion bcheck step is not too problematic. I usually disable it.

2) You may want to make sure your bval and bvec files are okay (DTIPrep may expect them in row or column format, and make sure they have the same number of elements). I have not come across that error yet.

3) Does that 59 index correspond to where your bval changes from 1000 to 2000 by any chance? Since image intensity changes with b-value, DTIPrep may be interpreting those high b-val scans as interlace-wise outliers. If so, you can change the protocol to make the interlace-wise check thresholds for gradients much more conservative, or disable the check altogether.

Does this makes sense?

Best,
Steven

Hi Maria 

1. Multishell: yes DTIPrep can deal with multishell, just make sure the data correctly converted to NRRD. See also: https://www.na-mic.org/wiki/NAMIC_Wiki:DTI:Nrrd_format

we apply DTIprep to multi-shell all the time (it's most common way to apply it). While it adjust threshold to be more forgiving for higher shells (due to the lower SNR), it's still the case that higher shell data is more likely (erroneously) removed. Thus, inspect your initial data after DTIprep and see whether you want to relax the threshold to be less strict.

2. best disable the diffusion check, just as Steven suggested

Best
Martin

Dear Steven and Martin,

Thank you both for the suggestions! I disabled the diffusion bcheck and it seems that the issue with stopping after a certain gradient is solved. It was indeed the part where the b2000 shell was starting. 

Many thanks again,
Maria