I’m helping a radiologist doing a machine learning study.
We have a dataset of brain MRIs taken from different subjects (and sometimes with different scanners too), i would like to register these MRIs to a common template before feeding them to a neural network. I’m currently using FSL for doing pre processing.
Should i use a two step registration or a single step registration ? Linear or non-linear ? And how many DOFs ?
Is it possible to use a MRI from the dataset as the template (instead of the MNI152 standard one) for registration, since the purpose is training a neural network ?
Thanks so much