Longitudinal analysis in FSL-MRS: Preprocessing and QC for MEGA-PRESS Glu

Data Processing
1

Hi everyone,

I am moving from a cross-sectional to a longitudinal analysis of Glutamate (Glu) using a MEGA-PRESS sequence.

I have already processed my T1 and T2 timepoints separately using a standard pipeline (fsl_mrs_preproc). Before I proceed with the longitudinal modeling, I have a few questions regarding the best workflow in FSL-MRS:

  1. Functionality: Does FSL-MRS have specific tools for longitudinal coregistration or voxel shifting/reconstruction to ensure the same tissue is sampled across sessions?
  2. Preprocessing: Should I re-run my T1 and T2 scans through a specific longitudinal batch, or is it standard to use the results from the independent cross-sectional runs?
  3. Quality Control: What are the best-practice QC measures for longitudinal changes (e.g. CSF correction or tissue fraction comparison or other) ?

I want to ensure my results reflect biological changes in Glu rather than preprocessing inconsistencies between sessions.

Thank you so much,

Arsenii

2

Hi @Arsenic ,

Sorry that I didn’t spot this for a few days.

There’s a couple of points t tackle here. First if you are using MEGA-edited data I’d expect you to want to use fsl_mrs_preproc_edit which is aware of the alignment of the editing dimension.

Regarding voxel placement, I’m afraid that for SVS data, there is nothing you can do post-acquisition to align voxels. Unlike most imaging data, where data is recorded in k space, and the image space grid can be arbitrarily shifted, SVS relies entirely on intersecting slice selection. The three slice positions are selected at acquisition-time and cannot be changed after, you will then record signal only from tissue in the volume that experienced all three slices.

In terms of checking whether you did acquire from the same location, you can use svs_segment to generate a voxel mask in T1 space. You can then register subject timepoint T1w structural images (and then the voxel mask) to each other or a standard space, e.g. MNI, and compare overlap.

For preprocessing there is no longitudinal treatment so use independently. I don’t think anyone currently has methods to preprocess data together.

For question 3, are you thinking of spectral and fit quality or voxel overlap? For the latter, people might use tissue composition (including csf fraction) as a regressor for statistical analysis. Regarding spectral quality, linewidth would be the first point of call, then SNR of a prominent peak, e.g. NAA.

3

Thanks for the suggestion. I actually used fsl_mrs_preproc for this dataset. As I mentioned in my previous threads (Error in FSL-MRS Quantification: "Specified metabolite isn't in the list of basis spectra" and MEGA-PRESS Edited Sequence Subtraction in FSL-MRS).
Thank you for your reply. I actually already ran the segmentation, so I will use that data for statistical analysis and QC