search

microsoft / Recursive-Cascaded-Networks

[ICCV 2019] Recursive Cascaded Networks for Unsupervised Medical Image Registration
https://arxiv.org/abs/1907.12353
MIT License
361 stars 87 forks source link

Question on cropping the 3D medical image to match dataset size to (128, 128, 128) #36

Open yoonguusong opened 3 years ago

yoonguusong commented 3 years ago

I have a question to match different 3D medical dataset into same dataset size without losing or tilting medical information. I am trying to match medical image using nib.Nifti1Image(image_resized, affine=None), but this seems changing medical data information.

zsyzzsoft commented 3 years ago

Our preprocessing stage also resizes all image crops to the same resolution. The spacing information will be changed, but the resulting deformation field can be transformed back.

yoonguusong commented 3 years ago

I have another questions as well.

  1. If you see the demo.py, there is "img_fixed, reader_fixed = preprocess_dcm(args.fixed)" what preprocessing will be done if you run this code?

  2. which data format did you applied? nii? mnc? dcm?

I'm trying to follow your method but hard to understand it. Thank you for your kindness

zsyzzsoft commented 3 years ago

The demo assumes the inputs being of DICOM series. The preprocessing stage first finds a rough bounding box of the liver area (using a threshold-based algorithm), crop the area accordingly and resize to 128^3, and then normalizes the intensity.

zhixiaoqiang-haha commented 3 years ago

我想请问一下 当我用您的网络去训练一组别的数据集 他是192192192的 怎么进行更改呢