Hello.

If you have access of T1 for your manually segmented images, then you can compute the transformation between T1 and FLAIR and apply it to the manually segmented binary (or other) image. This should be done individually for each subject.
Or another option is to compute the average T1 and FLAIR images for your data -- and use this one transformation map for all images.
Keep in mind that any warping of the spaces "after" segmentation is done is going to induce some errors on detection boundaries.

I would suggest comparing things in T1 space instead of FLAIR mainly because FLAIR is more sensitive to registration/warping errors than T1. See Spm8/Spm12 manual for further suggestions.