Symmetric population template for fixel-based analysis

Hi all,

I’m doing a fixel-based analysis on healthy controls and patients with tumours that are on either the right side or left side of the brainstem and vary in size and compression of the brainstem. I want to negate the risk of poor subject-template registration if, for example, a subject has a large left-sided tumour compressing the brainstem, but the template better reflects the brain with a small right-sided tumour that doesn’t touch the brainstem. My aim is to compare the left and right sides of the brain between the group with left-sided tumours and healthy controls, right-sided tumours and healthy controls, and left-sided tumours vs right-sided tumours. Right now, I am using a symmetric template with a subset (randomly selected) of the left-sided tumour patients’ original FOD and flipped FODs, the right-sided tumour patients’ original and flipped FODs, and the healthy controls’ original and flipped FODs. I then use a hemisphere mask derived from the symmetric template to compare the fixel-based metrics in each subjects’ flipped image to their original image. I was wondering if this is the best approach in template creation to account for the wide variability in tumour size, side, and degree of brainstem compression in my patient cohort?

Thanks!

1 Like

Hi @PMT,

As far as template generation goes, I think production of a symmetric template is probably the way to go. You would want to have a reasonable distribution of not only tumour hemisphere, but also tumour locations and sizes as well so that the overall shape of the brain stem is a reasonable “mean” representative of your cohort. (Or you could just use all of them if you have the computational time…)

You only briefly mentioned what you’re doing with such data subsequent to the template generation:

I then use a hemisphere mask derived from the symmetric template to compare the fixel-based metrics in each subjects’ flipped image to their original image.

A basic comparison like this probably doesn’t actually necessitate a hemisphere mask, you could do that subtraction over the whole brain and just see negated values in the reflection through the mid-sagittal plane. If however you are intending on doing any higher-order analysis on inter-hemispheric differences across many subjects, the preferable approach is to explicitly calculate a fixel data file quantifying the inter-hemisphere difference for each subject, and it is that image on which you then perform your analysis.

Rob