Nifti Inversion for ResizeWithPadOrCrop shows error

[Rubaida001]

Hi there,

I am currently working on harmonizing T1w MRI images. For the preprocessing step, I utilized the ResizeWithPadOrCrop function to resize the data. Following the harmonization process, I tried to convert the inferred image into subject space. I have attempted to use both the inverse and InvertD functions for this purpose, but unfortunately, none worked. I have checked multiple tutorials but none of the scripts run the InvertD function on previously saved datasets. Could you please help me to solve this problem?

The general steps I am following are as follows:

• Preprocess the data (resize the data into (192,192,192))
• Save the resized images and pass them to the harmonisation model
• Get the harmonized data and run the inversion function to get the original size

Here is the code with InvertD function for forward transform:

load_transformer =  LoadImaged(keys=["image"], image_only=False, ensure_channel_first=True)
    load_image = load_transformer({"image": image_path})
   
 # Define the forward transform pipeline
    preprocess_transform = Compose([
        ResizeWithPadOrCropd(keys=["image"], spatial_size=(192, 192, 192)),
        ToTensord(keys=["image"])
    ])

# Apply the forward transform
resized_img = preprocess_transform(load_image)

#Save Image    
img_saved = SaveImage(output_dir=output_dir, resample=False, output_ext="nii", separate_folder=False)
img_saved(resized_img["image"])

#Save transformed function and loaded image
with open('loaded_image.pkl', 'wb') as f:
    pickle.dump(load_image,f)	
with open('preprocess_transform.pkl', 'wb') as f:
    pickle.dump(preprocess_transform,f)	`

Inverse transform:

#Load pre-process transform function and original image with metadata
with open('loaded_image.pkl', 'rb') as f:
      batch_data = pickle.load(f)	
 
 with open('preprocess_transform.pkl', 'rb') as f:
      preprocess_transform = pickle.load(f)	
  
#Load the new harmonized image and append it to the original image dictionary
  load_transformer =  LoadImaged(keys=["pred"], image_only=False, ensure_channel_first=True)
  pred_image = load_transformer({"pred":'n4_PSYD_0106_S001_B_T1w_Skl_MNI_scaled_harm.nii'})
  print(pred_image.keys())
  batch_data["pred"] = pred_image["pred"]

# Define the inverse transform pipeline
inverse_transform = Compose([
    Invertd(
    keys="pred",
    transform=preprocess_transform,
    orig_keys="image",
    meta_keys="image_meta_dict",
    allow_missing_keys=True
         )
])
# Apply the inverse transform
inverted_data = inverse_transform(batch_data)

But I am getting following error:

`Traceback (most recent call last):
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/transform.py", line 141, in apply_transform
return _apply_transform(transform, data, unpack_items, lazy, overrides, log_stats)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/transform.py", line 98, in _apply_transform
return transform(data, lazy=lazy) if isinstance(transform, LazyTrait) else transform(data)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/utility/dictionary.py", line 464, in inverse
self.pop_transform(d, key)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/inverse.py", line 338, in pop_transform
return self.get_most_recent_transform(data, key, check, pop=True)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/inverse.py", line 320, in get_most_recent_transform
self.check_transforms_match(all_transforms[-1])
IndexError: list index out of range

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/transform.py", line 141, in apply_transform
return _apply_transform(transform, data, unpack_items, lazy, overrides, log_stats)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/transform.py", line 98, in _apply_transform
return transform(data, lazy=lazy) if isinstance(transform, LazyTrait) else transform(data)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/post/dictionary.py", line 715, in call
inverted = self.transform.inverse(input_dict)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/compose.py", line 364, in inverse
data = apply_transform(
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/transform.py", line 171, in apply_transform
raise RuntimeError(f"applying transform {transform}") from e
RuntimeError: applying transform <bound method ToTensord.inverse of <monai.transforms.utility.dictionary.ToTensord object at 0x7fe456d95a00>>

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "/data/project/PSY_SHARED/HARMONISATION_PROCESSING/CycleGAN/Python/cyclegan/preprocessing/inverse-1.py", line 97, in
inverse_transform(output_dir)
File "/data/project/PSY_SHARED/HARMONISATION_PROCESSING/CycleGAN/Python/cyclegan/preprocessing/inverse-1.py", line 76, in inverse_transform
inverted_data = inverse_transform(batch_data)
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/compose.py", line 335, in call
result = execute_compose(
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/compose.py", line 111, in execute_compose
data = apply_transform(
File "/home/k2031007/.local/lib/python3.9/site-packages/monai/transforms/transform.py", line 171, in apply_transform
raise RuntimeError(f"applying transform {transform}") from e
RuntimeError: applying transform <monai.transforms.post.dictionary.Invertd object at 0x7fe26c743af0>
`

[KumoLiu]

Hi @Rubaida001,
Looks like you want to apply invert on a new data without any information, perhaps you can try include the meta information from existing data. Something like:

batch_data["pred"] = MetaTensor(pred_image["pred"], meta=exsiting_data.meta)

You can refer to this tutorial for more details about how to using invert: https://github.com/Project-MONAI/tutorials/blob/main/3d_segmentation/spleen_segmentation_3d.ipynb

Hope it helps, thanks.

[Rubaida001]

Thank you so much for the suggestion. I changed the code and included meta information from ** resized_img["image_meta_dict"]** (output of forward transforms). But the error remained same.

batch_data["pred"] = MetaTensor(pred_image["pred"], meta=resized_img["image_meta_dict"])

Just for the clarification:

• Original MRI data size: 182, 218, 182
• After forward transforms: 192, 192, 192 (Run harmonization model on those data)
• Now I want to go back to the original space: 182, 218, 182

[Rubaida001]

Hi,
I followed the tutorial and given more details regarding the predicted data. The updated inverse transform is the following:

load_transformer =  LoadImaged(keys=["pred"], image_only=False, ensure_channel_first=True)
 pred_image = load_transformer({"pred":predicted.nii})
    
 batch_data["pred"] = pred_image["pred"] 
 batch_data["pred_meta_dict"] = pred_image["pred_meta_dict"]    

 # Define the inverse transform pipeline
 inverse_transform = Compose([
     Invertd(
     keys="pred",
     transform=preprocess_transform,
     orig_keys="image",
     meta_keys="pred_meta_dict",
     orig_meta_keys="image_meta_dict",
     nearest_interp=False,
     to_tensor=True,
     allow_missing_keys=True)])
 inverted_data = inverse_transform(batch_data)

However, I am still getting the same error (get_most_recent_transform self.check_transforms_match(all_transforms[-1]) IndexError: list index out of range).

Any suggestions would be appreciated. Thank you.

[KumoLiu]

Please check whether the resized_img["image_meta_dict"] has applied_operations by using resized_img["image_meta_dict"]["applied_operations"], By the way, since 0.9.1 we have introduced MetaTensor, you can direct use resized_img.applied_operations when data is a MetaTensor.
https://github.com/Project-MONAI/MONAI/wiki/MetaTensor-guide

[Rubaida001]

resized_img is in MetaTensor format; however, there are no applied_operations in the keys. I am using LoadImaged to read the NIfti files. I am not sure if I should do it in a different way.

[KumoLiu]

If there is no applied_operations in the data, then you can't apply invert. You need just ensure the preprocess_transform have been applied on resized_img, then you can use invert.
I would suggest you play with this notebook to see what will be saved in the applied_operations in the images.
https://github.com/Project-MONAI/tutorials/blob/main/3d_segmentation/spleen_segmentation_3d.ipynb