How datasets are organized using multimodal data to use MONAI's dataset classes

[LReion]

Please use MONAI's Discussions tab
How datasets are organized using multimodal data to use MONAI's dataset classes

I have 3D data for T1, T2, how should I organize these two 3D data as two channel input models, and how the dataset organizes the 3D data of these two channels

[KumoLiu]

Hi @LReion, you could use ConcatItemsd to concat these two 3d data.

MONAI/monai/transforms/utility/dictionary.py

Line 1016 in 25c9c39

class ConcatItemsd(MapTransform):

Hope it can help you, thanks!

[LReion]

I wanted to organize two 3D Ntifi file as two channel as a input, I need use MONAI dataset to load two 3d Ntifi file as two channels, but not use additional ways.

[KumoLiu]

I think you should have a compose transform containing maybe LoadImage and some other transforms that will send to Dataset, right?

[LReion]

Can LoadImage stitch keys like Label1 and Label2 in json files to stitch images?

[KumoLiu]

You could use LoadImaged.

transforms = Compose([
    LoadImaged(keys=["label1", "label2"], ensure_channel_first=True, image_only=True),
    ConcatItemsd(keys=["label1", "label2"], name="cat_label")
)

[LReion]

Thank for your answer.