Add metadata from other keys to MetaTensor.meta

[tom-osika]

I'm using a dataloader with the LoadImaged transform (with image_only=True) to read in MetaTensor objects. The data dictionary yielded by the data loader looks like the following, for example:

{'image': tensor([[[ 5,  2,  9],
          [61, 39,  4],
          [56, 32, 13]]]),
 'other_meta': {'meta1': 'val1', 'meta2': 'val2'}}

Where the tensor is a MetaTensor that contains other metadata such as the affine, etc. Is there any way to combine the other_meta dictionary with the metadata in MetaTensor.meta using transforms? There are no overlapping keys between the two dictionaries.

Thanks!

[KumoLiu]

Hi @tao558, you could try to use ToMetaTensord.

MONAI/monai/transforms/meta_utility/dictionary.py

Lines 89 to 94 in 1a55ba5

	class ToMetaTensord(MapTransform, InvertibleTransform):
	"""
	Dictionary-based transform to convert a dictionary to MetaTensor.
	If input is `{"a": torch.Tensor, "a_meta_dict": dict, "b": ...}`, then output will
	have the form `{"a": MetaTensor, "b": MetaTensor}`.

Hope it can help you, thanks!

[tom-osika]

Hi @KumoLiu, thank you for your response.

I tried using ToMetaTensord initially. If I set image_only=False when loading the images, I'll get something like the following each iteration before the ToMetaTensord:

{'image': tensor([[[ 5,  2,  9],
          [61, 39,  4],
          [56, 32, 13]]]),
 'image_meta_dict': {'affine': torch.Tensor(...), ...} # from reading the image with LoadImaged
 'other_meta': {'meta1': 'val1', 'meta2': 'val2'}} # My metadata

With ToMetaTensord, I can put the image_meta_dict into the MetaTensor, but not both meta dictionaries. Am I missing something?

Thanks again.

[KumoLiu]

Hi @tao558, you could combine these two meta info like this:

def combine_meta(data):
    data["image_meta_dict"].update(data["other_meta"])
    return data

out = Lambda(func=lambda x: combine_meta(x))(data)

Hope it can help you, thanks!