Serious Issue Found in Your Dataset
I highly appreciate your outstanding dataset and was eager to apply it to my work. However, I have discovered a critical issue regarding the image preprocessing quality.
In your preprocessing code using CT-Rate data, I have observed that you explicitly adjust the spacing to (x, y, z) = (1,1,3).
However, I noticed that the depth dimension in "train_7157/train_7157a/train_7157_a_1.nii.gz" exceeds 800, which indicates a significant inconsistency in data preprocessing. The depth axis is rather stretched.
In the CT-Rate dataset, the depth of the corresponding case is 408, with a z-axis spacing of 1. Therefore, the depth dimension should have decreased rather than increased. Could you clarify what might have caused this issue?
This is not an isolated case, as similar issues seem to exist in other instances as well.
I would appreciate it if you could review this matter and verify the dataset integrity.
Hi. Thanks for you interest in our work!
For each raw image in CT-RATE, we read its z-spacing from the meta information released by the authors, and re-space it accordingly:
https://huggingface.co./datasets/ibrahimhamamci/CT-RATE/blob/main/dataset/metadata/train_metadata.csv
https://huggingface.co./datasets/ibrahimhamamci/CT-RATE/blob/main/dataset/metadata/validation_metadata.csv
However, there are some mistaken z-spacing numbers in these csv files released by the author in their first commit. Although these numbers are later fixed in the second update, this cause the re-spaced images we released are not always 3mm on z-axis. You can check the history version of these csv files to see which images are corrected and get their correct z-spacing. We will also fix this problem in a future update.
Thank you for your response.
Has your work been processed based on the first commit of CT-RATE?
If so, could this be the reason for the issue I mentioned?
I hope this can be improved as soon as possible :)
Yes. We have processed the images base on the first commit of CT-RATE.
Take valid_1_a_1.nii.gz for example, the z-space is reported as 0.75mm in the first commit, so we downsample z-axis by 4 times. However, the correct z-space is 1.5mm, as updated in the second commit, so the processed image is actually 6mm instead of 3mm. You can do such simple calculation to fix this issue.