search

xmed-lab / FSDiffReg

MICCAI 2023: FSDiffReg: Feature-wise and Score-wise Diffusion-guided Unsupervised Deformable Image Registration for Cardiac Images
24 stars 7 forks source link

Data preprocessing #6

Open Miraclerice opened 8 months ago

Miraclerice commented 8 months ago

Hi, if it's convenient for you, can you provide the preprocessing code of your dataset? @Eason-Qin

Eason-Qin commented 2 weeks ago

Thank you for your interest in our work! I can provide the details of the preprocessing for you, and it should be fast to implement one:

To improve training efficiency, the training dataset is preprocessed by cropping and resampling before putting it to training pipeline. The preprocessing steps are consistent with DiffuseMorph. Specifically, for resampling, the SimpleITK BSpline Resampler is used to resample the image to (1.5, 1.5, 3.15) Voxel Spacing. For cropping, SimpleITK is used to center crop to (128, 128), and the processing of the z-axis is already written in the dataloader. The order is to resample first and then crop. The validation is performed using patients 090-100, and the same training/validation split is used for all baselines and our method.

Reference: Kim, B., Han, I. and Ye, J.C., 2022, October. DiffuseMorph: unsupervised deformable image registration using diffusion model. In European Conference on Computer Vision (pp. 347-364). Cham: Springer Nature Switzerland.