I have a question related to DWI data registration and FOD-based information that we want to incorporate into our subsequent analysis steps.
We are working with a data set of patients and we have two sessions for each patient with DTI measures - at two different time-points following a stroke. We want to look at white matter differences within subjects between the two time-points.
We, therefore, have two separate data sets for each time measurement (time point session 1 and time point session 2) after stroke. We have already completed the independent initial pre-processing steps (denoising, motion correction, edi current-correction). However, after that we want to estimate a response function for each subject twice - we want to compare the FOD results between sessions on a voxel-by-voxel basis. We need some advice on the pipeline to use for this purpose: shall we do the analysis first with the two data sets and then register the FOD images? Or should we perform the DWI registration after the pre-processing and then estimate the FOD images?
Our concern is that we are comparing data acquired at two-time points that we cannot merge into a single set, so we are not sure whether we can perform this if the data is not aligned in advance, and whether this would mean that we will be using two different response functions for the two sessions for each subject?
Your advice on how to solve this would be much appreciated.
I’d suggest you use a single (set of) response function(s) for your analysis. Think of it as a common basis that allows you to directly compare the FOD coefficients. If you have reason to believe that the stroke adversely affects single fibre voxels (see output of
dwi2response -voxels) in one of the time-points you might want to use the other time-point only, otherwise I’d use all time-points and subjects to calculate a (set of) average response functions (see docs). If there were genuine difference in the response function between the time-points, these would translate to differences in the FOD coefficients if you used one set of response functions. This way you can focus your analysis on the FODs only. If you used two sets of response functions you’d need to related FODs and response functions to each other making the interpretation of the results very hard.
Registration is typically performed and the resulting transformation applied to the FOD or FOD-derived images, not the DWI images as non-rigid reorientation requires a tissue-specific model. So the order is, 1. preprocess your data (as you have done), 2. estimate a group-specific (set of) response function(s), 3. use these to get the ODFs, 4. register the ODFs.
If you want to perform your analysis in the space after registration or in subject-native space and propagate the results to the joint space or a mixture of the two depends on the analysis you want to perform. You’ll need to think about spatial correspondence but also about how (nonlinear) transformations affect the properties of your data. Here is some food for thought.
Hi fellow mrtrixers,
Just a follow-up clarification on this point. We have a similar longitudinal design with multi-shell dwi data collected at two timepoints (and soon adding a third timepoint). After reading the documentation further and some of the posts on the forum, I understand that estimating a single set of response functions and using these for all further steps is the way to go. Just to clarify, does that mean we have to calculate the group averaged response functions using data from all subjects and all timepoints together?
Yes, a single set of response functions is usually what you want. You don’t have to use all data to calculate these. All of the points in my previous response hold true for more than two time points. Personally, I’d use all time points and subjects but you can use a representative subset for computational, data collection or data property reasons. You can use your first two time points to calculate the average response function and also use that for the following time points.
If you wanted to investigate if your set of response functions is equally representative of the tissue type they are attributed to, you could plot the coefficients of the response functions (for instance normalised to the b0 signal which is typically the first element in the response function files) for each datum against time or by group or other variable of interest and check for systematic differences. If there are large differences, that might hint at systematic problems in the response function estimation which you could investigate by inspecting dwi2response’s voxel selection masks (which is good practice anyway). If there were genuine large differences in response function values between groups, while your analysis might still not be negatively affected by this, it could change the interpretation of the resulting ODFs.
Thank you very much for the detailed explanation!