junma11 / abdomenct-1k Goto Github PK

In masks of the Whole Dataset, I found that two cases (Case_00219.nii.gz, Case_00656.nii.gz) miss spleen annotations. Why?

Dear AbdomenCT-1K Team,

Thank you for making this dataset publicly available to the medical imaging research community—I was wondering:

Are there any subject overlaps between the training (Tr) / testing (Ts) sets of fully (F) supervised and semi (S) -supervised datasets? More specifically, are there overlaps between the following pairs of datasets? TrF–TrS, TrF-TsS, TrS–TsF, TsF–TsS

Thank you,

Sean I Young

This seems like an incredible contribution to the medical imaging community :)
Do you have any estimates of when it will be released?

First of all, thank you for the AbdomenCT-1K dataset, which really solved the problem of partition generalization I encountered.However, when I downloaded the data, I found that the Mask data did not correspond to the Image data, for example, the Image data was 1~1000, but the maximum file of the Mask data was 1062, and the total number of mask was 1000. How to correspond to these data?

Hello. Thank you for your work.

Unfortunately, the Google Drive link to the 773 cases with pseudo tumor labels is not working, could I get the link for that?

In the Subtask 2 of AbdomenCT-1K Fully Supervised Learning Benchmark, it said that the training set is adapted from MSD Pancreas (281 cases) , LiTS (40 cases) , and KiTS (40 cases), where the cases are from different phases. However, after checking the data set, I found that the training set consists of MSD Pancreas (281 cases) , MSD Spleen (9 cases), LiTS (27 cases) , and KiTS (44 cases).

你好，JumMa， 十分感谢你提供的数据集。但是此处我遇到了一个问题，导致我Image数据集下载出现了问题。
我通过dropbox下载每次下载到最后就报错了，尝试了很多次, 对于0-1000的数据都出现了这个问题。这可能和数据集过大有关， 请问能不能国内百度网盘的下载地址了？多谢了。

Great job and thanks for the kindness to share the code and baseline models!
We have a dataset consisting of multi-phase contrast enhanced abdominal CT scans, and would like to segment the livers and spleens. The model trained for subtask2, which was trained with data from multi-phases, multi-vendor and multi-centers, should perform well for our dataset.

The models in BaiduNetDisk were named 'Task311_OrganSparse5', 'Task312_OrganSparse30', 'Task313_OrganSparse15', what are their differences? According to their pkl files, the training files seem to be the 41 MSD spleen cases ('Task314_SpleenSparse15' ). Thus they don't seem to be the mdoels trained for subtask2.

Could you make the model trained subtask2 public? Thank you very much.

Hi, Jun, it is mentioned that both Dice and NSD are important in medical segmentation, did you try to employ NSD as the segmentation loss?, and how about the performance in segmentation using the NSD as the segmentation loss?
Thanks, waiting for your reply~~~

hi，thank you for your work on Abdominal organ dataset.
When I download the dataset of 50 cases with 12 annotated organs from Dropbox，I find Image only has 48(3-50) niis while Mask has 50(1-50) niis.

Maybe you can upload the missing nii to the dropbox，hope this won't bother you.

Thanks for this great work.

I want to ask about the pseudo tumor label. Does it contains any tumor such as colon cancer and stomach cancer, or just the cancer in the four organs?

And I am also wondering if the dataset have overlap with the dataset in Flare-2022 and 2023?

Thank you very much!

Thanks for this great contribution.
It is noticed that 773 cases with pseudo tumor labels without groudn truth. Could you tell me where I can find ground truth of pseudo tumor labels (i.e. the tumor label of Case_00001-00773) for noisy label research.

The dataset looks awesome but the download speed from zenodo seems to be too slow ......
The share link from Baidu Yun is invalid now, could you please update it? Thanks a lot.

Hi, I have several questions for the data.

1. In the link in the https://forms.gle/XDrxSgoCXs3jzn8U7, there are downloading link for the data. But the data only contains the image data but not includes the labels. So if I want to use a certain benchmark, should I download the data in the grand challenge page but not the whole dataset indicated in the readme?
2. I find the data in this grand challenge pape (https://zenodo.org/records/5903037#.YfDOl-pBybh) contains the labels. However, I find the issue #7 mentions that the data is not correct and you updated it in the baidu net disk (but the baidu net disk is broken). The date in the grand challenge page is Jul 27, 2021 (earlier that the date of the issue). So does the data of subtask2 on the grand challenge page correctly correspond to the setting in the paper?

First of all thanks for your effort in making this incredible contribution to medical analysts! However I cannot find a way to download the 773 cases with pseudo tumor labels without Baidu account (I'm in Austria, with an Austrian mobile number) .. is there a Dropbox/MEGA/alternative link available?

Thanks!

Are metadata available for the different recordings somewhere? Patient age/sex/height/weight/diagnoses/ethnicity, anything? Recording hospital, scanner device, settings?

Hi,

Thanks for this incredible collection of abdominal data.
I was wondering what kind of preprocessing is used in experiments in paper (normalization, resolution or image size, if patchwise then input size). This information will greatly help everyone to relate their results to the one in the paper.

Thanks,
Sukesh

直接用来推理不可以吗？
报错：
starting prediction...
inference done. Now waiting for the segmentation export to finish...
WARNING! Cannot run postprocessing because the postprocessing file is missing. Make sure to run consolidate_folds in the output folder of the model first!
The folder you need to run this in is /home/xhw205/nnUNetFrame/DATASET/nnUNet_trained_models/nnUNet/3d_fullres/Task003_L iver/nnUNetTrainerV2__nnUNetPlansv2.1

Hello, I'd like to ask you a question. I feel that some data ct sequences are not just abdomen. Is there an automatic method to extract abdominal ct sequences from all ct sequences?

Hi, Jun, perfect and selfless work, and admire you.
I hope that the paper will be published as soon as possible!
Can I ask you two questions?
Q1: in “4.2 Semi-supervised organ segmentation benchmark”, before this step, all four organs’ GTs have been labeled. I want to know, in this SSL step, the input GT is only one organ, or all four labeled organs during training?
Q2: also in section 4.2 about SSL, you employed model is nnUNet, so the input batch size is still 2?
Thanks, waiting for your reply, and wish you many scientific achievements.

I was wondering if the ct volumes of one of the AbdomenCT1K datasets you have provided are labeled to indicate which dataset they are from: LiTS, KiTS, MSD Spleen, MSD Pancreas, NIH Pancreas, etc. For example, I was wondering which dataset the ct volume named Case_00413_0000.nii.gz that I downloaded from your GitHub came from.

Thanks a lot for this awesome work which release a large dataset for the Abdomen organ segmentation task. And in the paper, you mention that these cases are from 12 different centers. Could you please release the center information or the correspondence between the sample and the original datasets.

Do you mind sharing the gist to create the "3D rotating view" segmentation gif you use in your readme ? It's pretty cool!
Thanks in advance.

Could you perhaps share with us which subjects (1-1062) in the dataset come from which dataset (NIH, KiTS, LiTS, MSD, etc.)? Thank you in advance!