I have a question regd. just the data pre-processing part. I am looking to train a model for brain tumor segmentation using nnunet. I am using all the preprocessing steps listed here: fslreorient2std, hd-bet, flirt, fslmaths. The issue is, the ground truth tumor segmentations have also been created on the raw images (not pre-processed images). So, any transform I apply to the raw images to pre-process them (all the listed steps) would need to be applied to the tumor segmentation images as well. Do you have any advice/pointers/code on the corresponding set of preprocessing commands, which also save the transformations in each step, and then apply them on the segmentation images?
Hi,
not really because I did not do that myself. Thoughts:
You should be able to apply fslreorient2std on the segmentation as well
hd-bet cannot be used on the segmentation, so you use it on the corresponding image and save the mask (there is an option for that). Then you apply the mask to the segmentation manually (if needed at all)
flirt needs to be applied to the images. You can then apply the resulting transformation matrix to the segmentation as well. Follow this code: https://github.com/NeuroAI-HD/HD-GLIO#this-is-how-to-do-it-with-fsl6. IMPORTANT change the interpolation to nearest neighbor for the segmentations!!!
Hi Fabian,
Thank you for this excellent tool!
I have a question regd. just the data pre-processing part. I am looking to train a model for brain tumor segmentation using nnunet. I am using all the preprocessing steps listed here: fslreorient2std, hd-bet, flirt, fslmaths. The issue is, the ground truth tumor segmentations have also been created on the raw images (not pre-processed images). So, any transform I apply to the raw images to pre-process them (all the listed steps) would need to be applied to the tumor segmentation images as well. Do you have any advice/pointers/code on the corresponding set of preprocessing commands, which also save the transformations in each step, and then apply them on the segmentation images?
Thank you! Best