fast marching question

I am attempting to remove part of the background of an image as it causes issues with the registration pipeline (i think). I am using a combination of FastMarchingUpwindGradientImageFilter and ScalarChanAndVeseDenseLevelSetImageFilter in SimpleITK.

I use seed points each of the image corners execute the fast marching filter on a smoothed image (anisotropic diffusion + gaussian + normalization) and it returns an image containing what appears to be outliers, where the peak value is on the order of 94 standard deviations larger than mean, (mean and median are close together)

when this is passed to the levelset filter the outliers essentially wash out everything and I get a blank image in return.

I fixed this after a lot of tinkering by thresholding the image to anything less than 6 standard deviations away from the mean.

can someone give me some insight as to why fast marching does that? I am concerned its either a bug or a lack of understanding on my part. I couldn’t find a flag or variable that indicated something useful.

I also used the FastMarchingImageFilter and got the same kind issue.