Questions Regarding Data

1. Is the data provided in the competition pre-processed? Has deformation registration already been performed?

2. Why does the mask contour appear significantly expanded compared to the CT contour? During the testing phase, will the HU outside the CT body contour also be compared?

For 1, I think those images were pre-processed by the provided preprocessing codes:

Dear Jing,

Yes, the data is preprocessed as explained in the previous answer, and as you can find described at https://arxiv.org/abs/2502.17609.

Further, we expanded the mask to ensure that the model may match dimensions of the body contours and to make sure that also air is correctly generated. To avoid that very large air/blank region are considered, only some centimeters around the body are considered.

I hope this addresses your questions.

Warm Regards,

Matteo

Dear organizers,

Just to be sure, is the github code for preprocessing the one really used for the data provided ? Because the code there explicity says that the outputs are nii.gz files which is different from the arxiv and the challenge data description.

Best regards

The conversion to .mha was applied to all the data afterwards. The data processing pipeline in the code is indeed the one used to prepare the dataset—there’s no doubt about that.

Do I need to process the data of the current.MHA file again through the preprocessed code provided by you, and then proceed to the next step of training?

The data provided has been preprocessed with the provided code. The code has been provided for transparency.

It is up to you as participant to the challenge to assess and decide whether the data needs further pre-processing to achive better (ideally the best) performance.

In the data preprocessing code, stage2 sets the CT valid HU range at -1024,3072. Is this recommended for all body parts?

Hi,

Data is available for all body sites and from all centers, so you have all the information needed to answer this question. The processing is consistent across the training, validation, and test sets.

As organizers, we will not provide guidance on how to best approach the task — this is up to you as a participant.

However, we are happy to assist if any rules are unclear or if you encounter any issues with the data, submission system, or evaluation.

Good luck with developing the best solution!

Warm regards, Matteo