In longitudinal neuroimaging studies, it’s common to have T1-weighted structural scans acquired at multiple time points—sometimes even on different scanners or with different acquisition protocols. While FSL’s SIENA is a widely used tool for estimating percentage brain volume change (PBVC), things can get tricky when your images aren’t perfectly matched.
That’s when BET (the skull-stripping step SIENA uses) sometimes goes rogue—chopping off bits of the frontal or temporal lobes in one scan, or leaving in neck and jaw artifacts in another. This mismatch can really throw off PBVC estimates.
Some reasons it happens:
· Scanner hardware/software changes between visits
· Different head positions or fields of view
· Contrast/SNR changes from protocol tweaks
· Non-brain structures confusing BET.
A few fixes I’ve seen or tried:
· Tweaking BET’s -f and -g parameters for each time point.
· Cropping with robustfov before skull-stripping.
· Running bias field correction (FAST, N4) first.
· Using external brain extraction tools (SPM, FreeSurfer, ANTs) and passing SIENA the stripped brains directly.
· Manual mask touch-ups when all else fails.
Would be interested to hear other memberss’ experience with this issue, and how they usually deal with it.