There is a promble about segmentation on unprocessed HCP (320×320×256) ?

[FreshmanMa]

why is 27? not 36

[fepegar]

Sorry, I need more context to help you.

[FreshmanMa]

Sorry, I need more context to help you. Thanks， I'm using deepgif to segement on preprocessed HCP 1200 MRI.nii.gz data?

[FreshmanMa]

[FreshmanMa]

So，deepgif is not suitable for preprocessed HCP 1200 MRI.nii.gz？

[FreshmanMa]

So，deepgif is not suitable for preprocessed HCP 1200 MRI.nii.gz？

Is suitable for this kind of mri data？

[fepegar]

It should be. Have you tried?

[FreshmanMa]

It should be. Have you tried? Yeah，I had tried with deepgif. It doesn't work on my mri.nii.gz? I have no idea about how to do. I feel a little sad, can you give me some advice or provide other method to segmentation about my data？

[fepegar]

Can you please clearly explain what you tried, what you expected and what happened? And share a minimal working example I can reproduce?

[FreshmanMa]

Can you please clearly explain what you tried, what you expected and what happened? And share a minimal working example I can reproduce?

OK，Thanks, I wanna use HighRes3DNet to do segmentation task on my 3d mri.nii.gz after skull-stripped, and get the picture like the following img which is included in the thesis, "MRI Super-Resolution with GAN and 3D Multi-Level DenseNet: Smaller, Faster, and Better". But, in this paper, their data is original data, is not preprocessed. I provide one data, 531536.nii.gz data. I use deefgif 531536.nii.gz to run, you can have try.

531536.nii.gz

[fepegar]

Here's a reproducible example using your data: https://colab.research.google.com/drive/12Z5QnWlGuShMnKlnM_HWmNdHNdmxoK2r?usp=sharing. It doesn't seem to generate good results. This is probably because the network wasn't trained on skull-stripped images.

I recommend using more modern and robust tools such as SynthSeg.

[FreshmanMa]

Ok, thanks very much, I'll try the SynthSeg! It's nice of you, thanks again!

[fepegar]

You're welcome!