Did you use five-fold cross-validation during the evaluation in the WORD paper? You only mention that you use the default settings and 5F-CV is part of the default settings, but I want to be sure.
Hello, thank you for your valuable work. I noticed in your paper that there are samples of rectal cancer mentioned. Are these samples not individually labeled for the cancer site? Could you disclose which samples contain rectal cancer? Thank you very much!
Hi
I'm really in need of a neural network which has been trained on your dataset, but as result of lack of appropriate hardware resource its impossible for me to do it. Is it possible for you to share few models that you have trained in your paper?
Thank you for the great dataset! I read the paper and was impressed by your scribble-supervision approach. However, I was unable to find the implementation in this repo. It was stated in the paper that the implementation was public. Could you be so kind as to direct me to the scribble-supervision code?
The relevant parts for me would especially be these:
The implementation of nnU-Net V2 with ResUNet 2D backbone
The implementation of the pCE + L_ent + L_ivm losses
I would also be grateful if the trained weights for the scribble-supervision model could be made public.
I've tried training both 2D nnUNetv2 and 3D nnUNetv2 using the WORD dataset. But I found that the segmentation results of Head of femur were not as good as illustrated in the paper. The result of 3D nnUNetv2 were:
target
DSC
HD95
Head of femur(L)
0.545±0.114
196.485±22.436
Head of femur(R)
0.471±0.138
173.113±35.954
If you have met this situation, I wonder if you can offer me an explanation or guideline. Thank you😀
Hello, thank you for your valuable work. I want to use your data set to train the segmentation model. I want to use threshold interception in the preprocessing process. I wonder if you know how much the most suitable threshold should be selected?
(Alaa )
Primary Affiliation: (King Fahd University of Petroleum & Minerals (KFUPM))
Job Title: (Master Student in AI major)
Affiliation Email: ([email protected],)
How to use: (for academic project under computer vision course.)
Hi
In the another issue I asked for pre-trained networks, that you generously shared pre-trained nnUNet3DV2 model with me. Is it possible for you to also share a 2D pre-trained model?
Thank you for your amazing work! I am currently trying to reproduce your scribble experiments. The WORD dataset includes a labelsTr and labelsVal folder for training with the full segmentation, but only a scribblesTr folder for the scribble segmentation. Does that mean that you validated on the full segmentation (labelsVal) during training? Or did you make a separate train/val split for the scribbleTr cases?
I have no problem with either variant. I just need to know in order to reproduce your results :)