Loading BraTS-20 dataset for Tumor Segmentation

[faizan1234567]

I have a BraTS-20 dataset for brain tumor segmentation. I want to load this dataset for training. I have gone through this tutorial, however, in this tutorial author uses DecathlonDataset for loading the BraTS-16 and BraTS-17 datasets. I need to load BraTS-20, I have searched through the issues on this forum but I have not found the required tutorial. I want to load the BraTS-20 dataset using data loaders. Anyone, please help me to sort this out.

[Nic-Ma]

Hi @faizan1234567 ,

Thanks for your interest.
I think you can refer to the dataset and dataloader in this tutorial:
https://github.com/Project-MONAI/tutorials/blob/master/3d_segmentation/spleen_segmentation_3d.ipynb

Thanks.

[WenqingZong]

Hello, I'm doing transfer learning so wondering if it's possible to only load one modality from BraTS dataset? Thanks in advance!

[Nic-Ma]

Hi @dongyang0122 ,

Could you please help share some best practices about this question?

Thanks in advance.

[dongyang0122]

Hi @WenqingZong, you can customize a simple transform via Lambdad to select input channels of images. For example,
Lambdad(keys='image', func=lambda x: x[0:1, :, :]).
Once you select the channel, you may need EnsureChannelFirstd to make sure the input data is 4-D before batching.

[WenqingZong]

Hi @dongyang0122 , thank you for your reply, I got another question on the dataset. I know BraTS contains 4 modalities, (T1, T1ce, T2, FLAIR), and I currently have a batch of shape (BatchSize, Modality, Height, Width, Depth), I'd like to know which modality index corresponds to which actual modality?

[WenqingZong]

It's important as my transfer learning especially focuses on the direction from T1ce to T2.