hello, I am comparing metabolites in groups of young versus old adults. They of course have significantly different GM ratios within the 3x3x3 voxel.
Would it be recommended to draw concentration values from TissueCorrWaterScaled Voxel table in Osprey in this particular situation? Drawing from the tCr table vs. the tissue corrected table gives very different results. I do worry that I should be correcting for individual tcr levels
thank you!
Hi @sdw,
This is a perennially complicated issue for MRS scientific research. In my opinion, it heavily depends on what groups of younger and older adults you are studying. For example, is the older cohort a clinical population with known water or tCr changes attributed to their diagnosis? This could help you choose one of the two signal references. However, age-related changes in brain tissue water properties in even healthy aging adults are a well-established finding, so water still can be problematic when you have large age ranges unless you take care to account for this factor if using the TissueCorrWaterScaled value. However, you probably should be using the TissueAlphaCorrWaterScaled value for the following reason:
Another important consideration is the intrinsic differences in metabolites between GM and WM, which we describe in our paper. While the paper centers on intrinsic GABA differences between GM and WM, the mathematics applies to any MRS-measured metabolite concentration value. So, using the suitable correction parameters for each metabolite plays an important consideration in your reported results.
Related is the approach from Harris et al., which will normalize GM and WM fractions to the mean GM+WM fraction. However, in my opinion, it’s still uncertain what to do when you have two groups with significantly different GM/WM values. Some have argued that choosing one group to normalize values to is more appropriate, but I’m unsure of its statistical validity. In previous publications, my colleagues and I have decided to normalize each of the young/old groups’ metabolite values to the mean GM/WM values of their groups, which, to me, feels more correct.
An alternative and valid approach is to add your GM and WM segmentation fractions (normalized by one minus the fraction of CSF - which is appropriate for most metabolites except maybe Lac, which can be found higher in CSF [depending on your voxel placement, e.g., the ventricales]) for each volunteer into a multiple linear regression model, which would account for intrinsic metabolite-specific GM/WM ratio differences across individuals.
This is just one of several approaches presented in the literature, but one I feel is appropriate for you. That is unless you have T1/T2 water maps of each volunteer or availability to a template map that fits your younger and older adults (see, e.g., https://doi.org/10.1101/2024.09.27.615424).
Mark
dear @mmikkel,
Thanks so much for your help!!
I took the approach you suggested of including the GM/WM segmentation fractions into our linear regression model that uses creatine concentration ratios from Osprey, but am having trouble interpreting why this approach differs so drastically from a model that directly uses the TissCorrWaterScaled values? Is it obvious to you why this would happen – is this because of tissue water scaling issues you mentioned? We did not collect T1/T2 water maps + I’m uncertain how to incorporate the template map you mentioned into the Osprey workflow.
My models include covariates for sex, time of day, and age. Our age range is 18-85 and all adults are healthy and do not have any cognitive impairments and the location was occipital ctx (n=87 unique subjects, 120 scans).
original model 1: tCr ratio concentrations, without accounting for tissue fractions. Significant age effects for NAA, lactate, myo-inositol, and glutamate.
Model 2 (following your advice): tCr ratios + additional gray and white matter fractions as covariates (note: I did not normalize by 1–CSF.) After tissue correction, lactate was no longer significant. There was a significant age effect for myo-inositol, glutamate, and NAA.
Model 3: Used values from the **TissueCorrWaterScaled output. In this model, only lactate was significant — which confused me, as it was not significant in Model 2. Is it correct that I should not include tissue fractions as covariates in this model since I am using the values from the tissue and water scaled tables?
Why would Model 2 (creatine ratios + tissue fraction covariates) differ so much from Model 3 (TissCorrWaterScaled)??
Hi @sdw,
Try:
Mark
I’m sorry, I only have a TissueAlphaCorrWaterScaled for the GABA edited spectra and can only pull values for GABA+ from this table. Am I misunderstanding you or did you mean to pull only gaba+ from that table specifically
Ah, I see. I wasn’t aware that it only produces those values for GABA+.
Could you please email me (mam4041@med.cornell.edu) a copy of the code/script you are using to run your Gannet analysis?