I have a small question regarding possible biases in population templates used for fixel-based analysis.
I have created a population template for FBA using all my subjects (2 groups, patients and controls). Now I am trying to test different sub-groups within my subjects, and I was wondering if I would have to create a separate population template for each sub-group test separately or would simply excluding the unneeded subjects at the statistical analysis step be enough? I’m sure that doing so would introduce a bias of some kind, but I was wondering if that bias would be negligible or not. Thanks!
One way to think about this would be:
Can you make a prediction a priori regarding how the templates would look different between generating from the whole population versus the sub-population?
If the subset of the population that you are assessing clearly differs from the rest of the population, then you would expect that generating a template just from that specific sub-population would be a closer fit to those data, since it would not be “dragged” by the other dissimilar members of the whole population. Conversely, if the sub-population (from the perspective of the image data) are essentially a random subset of the whole population, such that generating a new template from that subset would basically look identical to the original template (maybe with lower SNR due to using less subjects), then committing a whole load of CPU clock cycles to produce a new template probably won’t yield any benefit.
That makes a lot of sense. Thank you so much for your input.