Cannot reproduce results on FeTA using your slicing strategy. about nesvor

[daviddmc]

Yes, res_s is the in-plane resolution of the simulated images, so it should be greater than the resolution of the source volume and the reconstructed volume..

[daviddmc]

Hi @Edenzzzz, did you try to reconstruct using the default parameters, like this?

nesvor reconstruct \
    --input-stacks /incoming/stack-1.nii.gz ... /incoming/stack-N.nii.gz \
    --thicknesses <thick-1> ... <thick-N> \
    --output-volume /outgoing/volume.nii.gz

The segmentation model (MONAIfbs) is used to remove the maternal tissue in the images and the model was trained on data with maternal tissue. Therefore, If you apply it to an image without background. It might remove some of the brain.

Did you check the simulated stack data? Could you share the simulated files with me?

[Edenzzzz]

feta_stack3_gap=3.nii.gz
feta_stack2_gap=3.nii.gz
feta_stack1_gap=3.nii.gz

Thanks for your reply. Here are my stack files sliced from FeTA and I'm looking into the options you mentioned.

[daviddmc]

Hi, you may try to save each stack as follows.

from nesvor.image import Stack

Stack(
    slices, 
    transformation=transform, 
    resolution_x=res_s, 
    resolution_y=res_s,
    thickness=s_thick,
    gap=gap,
).get_volume().save('stack.nii')

You may find more information about the Stack class here.

[Edenzzzz]

Hi, you may try to save each stack as follows.

from nesvor.image import Stack

Stack(
    slices, 
    transformation=transform, 
    resolution_x=res_s, 
    resolution_y=res_s,
    thickness=s_thick,
    gap=gap,
).get_volume().save('stack.nii')

You may find more information about the Stack class here.

Thanks for the reply! The result now looks closer to the input.
However just to be sure, based on my code is res_s always 1.5 or the resolution for the source volume (0.5 mm for FeTA) and res is the desired output stack resolution (0.8 according to the paper)? I've found only setting res_s to 1.5 produces reasonable results.

This repo is well-structured and promising, but I think users who are not expert in MRI processing may benefit from a bit more instruction on processing stacks and specifying resolutions. Thanks for your contribution!