search

cilix-ai / on-the-fly-guidance

[MICCAI 2024] On-the-Fly Guidance Training for Medical Image Registration. Pre-print available in link below.
https://arxiv.org/abs/2308.15216
Creative Commons Attribution 4.0 International
29 stars 3 forks source link

Question about Datasets #2

Closed VGANGV closed 1 year ago

VGANGV commented 1 year ago

Hi! Thanks for your work! You did subject-to-atlas experiments on both LBPA40 and IXI datasets, and I noticed that the DataSet you designed for them is inconsistent, at datasets.py#Line28-29, x means atlas, y means moving image (which is the same as TransMorph's implementation), but at datasets.py#Line158-159 it is reversed, and you still used the same strategy for training (x to y for both). Is there any reason for this design?

yuelinxin commented 1 year ago

Hi, if I'm not mistaken, the IXI dataset is patient to atlas, and the LPBA40 dataset should be atlas to patient. But this part of the code is handled by @Cycyes , you can wait for his confirmation.

VGANGV commented 1 year ago

Thank you for your reply! It is confusing because I didn't read in the paper that the two datasets are for different tasks. I will wait for his/her reply.

Cycyes commented 1 year ago

Thank you for bringing this to my attention! The inconsistency you observed in the dataset setup is intentional, as the code was designed to handle both 'atlas-to-patient' and 'patient-to-atlas' tasks to verify its generality. I apologize for any confusion, and I'll work on adding clearer documentation or comments in the code to explain this design choice.

VGANGV commented 1 year ago

Thank you, that makes sense, I will close this issue. 👍