mattiaspaul / deedsbcv Goto Github PK

Hi Mattias,

I was going through your tutorial for image preprocessing for training a deep registration model. I have a dataset of 131 liver images from Medical Segmentation Decathlon with a fixed size of (432, 320, 240). I am using LinearBCV for the affine pre-alignment but a single registration takes around 40 seconds on an AMD Ryzen 3950x processor. So, for 131 images that's 17030 registrations and it would take around 8 days. Do you have any suggestions for what to do for pre-alignment when dealing with this size of dataset. Thank you.

Regards,
Roman

hi,
I got a problem when I ttried deedsbcv. hope anyone can help.
I just tried deedsBCV in MRI data, and the data is all right. I checked already.
the following is what I tried:
I tried the rigid method, which came out the same.
I also changed the input and output image in case these two phrases had few differences.
but the result remained the same.
-G -L -Q parameters also changed as the github recommended.

here is the output:

Hi author, thank you for opensource your work. I have a question about the demo.

I have used the default setting to run the kidney data but the result was not very well. The moved image is not overlapped with the fixed image. Do I need to run the code several times? If you don't mind, could you please share the parameter used for the kidney data. Thanks.

when I wanted to download the dataset to run your exsample , I found that "Beyond the Cranial Vault" MICCAI workshop dataset has been modified, is not the dataset for abdomen. Could you give your dataset for me? I really want to learn you code and run it.Thank you very much!

Hi,

When I implement the second case,I have a problem：

Why we should flip label image around y axis ? Because I find that the label of pancreas correspond to the abdominal cavity image,but when flip label image around y axis,they are not corresponding.

and:
When I not flip the label image,the result is as the following figure(use 3dslicer to observe the result):

the seg1_res.nii.gz and nonlinear_1_3_deformed_seg.nii.gz overlap.

When I flip the label image,the result is as the following figure:

we can find that there have a displacement between the seg1_res.nii.gz and nonlinear_1_3_deformed_seg.nii.gz,the displacement make the two images cannot overlap.

Looking forward to your answer！

Thanks!

Hi, I'm using deedsBCV to do registration on two binary mask. Below is pipeline I used:

1. Use linearBCV to do liner registration
2. Use deedsBCV to get ddf.
   Now I have a moving binary mask and several coordinates. How can I get the warped coordinates?

Hi, I'm trying to do CT to MR image registration. But it throw an error when I was following the "Step 4" as:

I have no idea the reason of this error. Could you help me with this?

I want to read the displacement field by python. How to read the field using Python?

Dear Author:
I have noticed that you have published the 2D mind version. Do you have 2D examples of rigid registration? Because I want to make a comparison with your method. If you are willing to share it, it would save a lot of trouble for reproductivity. This will also guarantee that such a good method gets a fair comparison.

Hi there.

I want to do registration by deeds, but my image are stored in PNG format. I convert the PNG files into Nifti and implement this argument: ./linearBCV ... , but error happens, it says: Segment error (core dumped).

Can you give me some insight about this question?

Thanks.

Dear Author:
Thanks a lot for making your distinguished research public available. I am wondering have you tried your methods on the brain CT and MRI image multimodel registration?

Dear developers

Thanks for your efforts. It is a really useful tool.
In our case, we have two images (reference image and deform image), we want to registry the deform image to reference image.
When we use the default setting, the deform image will be registered to reference image, even we can say it is too similar to the reference including intensity values. Do you have any idea to keep the range of intensity values of deform image, so that the details can be kept. Looking forward to your reply.