Modifying Preprocessing

[hyunjung00]

Hello! In the preprocessing step, I want a fixed patch size among different datasets and also I don't want to crop the original dataset. Do you know how I can do this? Thank you : )

[sten2lu]

Hi @hyunjung00, I am not sure whether I understand you correctly. Generally, nnU-Net creates plans with specific patch sizes for training and specific resolutions for training. During preprocessing, the images are cropped (for specific dataset e.g. BraTS) and resampled to the resolution specified in the plans.

Do you want to share the preprocessing across datasets, or do you want to share the nnUnetPlans across experiments?

Best regards,

Carsten

[hyunjung00]

Dear @sten2lu,

I want to share the preprocessing across datasets. What I want to do is to apply triplet loss for different modalities of MRI. I preprocessed 2 different datasets (e.g. Dataset id: 111 for modality 1 and Dataset id: 222 for modality 2) and these two have different patch sizes. I want to make them the same so the model can be the same.

Best regards, Hyunjung

[sten2lu]

Hi @hyunjung00,

Ok, so I assume that you also want to use the same U-Net architecture of nnU-Net on both datasets. For this, you will need to share the plan files across datasets. Also, the patch_size should be a parameter given in the plans file but is used during the preprocessing. Therefore, I assume that sharing the plan files should solve your issue. A description of how plans can be shared/moved across datasets -- here is a description in the Documentation.

For a detailed description of the plans file and the network topologies, I refer to the Documentation.

I hope this helps you solve your issue. Also, sorry for the delay. I am currently on vacation since last week.

Best regards, Carsten