SaveImage with Compose

[vikashg]

Hi all,
I want to do some preprocessing and save the preprocessed images but if I write a transformation chain as

transforms = Compose([LoadImage(), EnsureChannelFirst(), Resize(spatial_size = (nx, ny, nz)), 
SaveImage(output_dir = './ResizedImage', output_ext='.nii.gz')])

If I save this image I see that the image saved has its original dimensions. I wonder if there is something that I am doing wrong.

[KumoLiu]

Hi @vikashg, yes SaveImage will resample the image by default, you could set resample=False then it will not resample the image based on the spatial_shape (and original_affine).

MONAI/monai/transforms/io/array.py

Lines 320 to 321 in 66422f8

	resample: whether to resample image (if needed) before saving the data array,
	based on the `spatial_shape` (and `original_affine`) from metadata.

Hope it can help you, thanks!

[vikashg]

@KumoLiu I think I spoke too soon. Can you try one more time with a slice number other than 108 and see if it works? Because the image I shared with you already has 108 slices. So, it was a bad test case.

Thanks
Vikash

[KumoLiu]

Hi @vikashg, I just take a look at the original data, the shape is (512, 512, 83). I would like to say you may check the data after transform chain, it didn't replace the data that was originally with key “image”.

from monai.transforms import LoadImage

data = LoadImage(image_only=True)(data_path)
print(data.shape)

Thanks!

[vikashg]

So @KumoLiu. Here is the code (I just added 1 line before the transformation to get the shape of the input)

from monai.transforms import LoadImaged, EnsureChannelFirstd, Compose, Resized, SaveImaged

data_path = "/workspace/app/002D_raw.nii.gz"

data = {"image": data_path}

input1 = LoadImaged(keys=['image'])(data)
print(input1['image'].shape)

transform_chain = Compose([
    LoadImaged(keys=["image"]),
    EnsureChannelFirstd(keys=["image"]),
    Resized(keys=["image"], spatial_size=(-1, -1, 48)),
    SaveImaged(keys=["image"], output_dir='.', output_postfix='resized', output_ext = '.nii.gz', resample=True)
            ])
out_resized = transform_chain(data)


out = LoadImaged(keys=["image"])({"image": "./002D_raw/002D_raw_resized.nii.gz"})
print(out['image'].shape)  # --> torch.Size([512, 512, 108])

and here is the output

torch.Size([512, 512, 83])
2023-02-23 02:55:58,101 INFO image_writer.py:194 - writing: 002D_raw/002D_raw_resized.nii.gz
torch.Size([512, 512, 83])

So the data has the same size after resampling

[KumoLiu]

Hi @vikashg, l noticed that you still using resample=True. If you don't want data resample to the original shape, you may need set it to False.

[vikashg]

Aah ok. I will give it a go tomorrow morning thanks