CropForeground not working

[oscarramirezs]

Hello,

I'm working on a project segmenting two bone from CT the images are 512x512. (The number of images can vary).
My issue is that I need to create a nifti output for each bone without any background.
However, when I do this, the coordinates of each bone are shifted compared to the original.

this is the output of my model

            val_labels = [post_label(i) for i in decollate_batch(val_labels)]
                                      
            for i, c in enumerate(val_outputs):
                label = ["bone1", "bone2"]
                   
                bone1 = []
                bone2 = []                                  
              
                bone1.append(c[1])
                bone2.append(c[2])

                bone1 = [np.stack(scapula, axis=0).astype(np.float16)]
                result= np.array(bone1)

                cropper = CropForeground(select_fn=lambda x: x > 1, channel_indices=None, margin=0, return_coords=True)
                result_=cropper(result)


                metadata=val_data["image_meta_dict"]  
                
                saver_h_2 = NiftiSaver(output_dir= "val_cut_test/", output_postfix='cut', output_ext=".nii")
                saver_h_2.save_batch(result_[0], val_data["image_meta_dict"])       

can someone help me?

thanks

Oscar

[KumoLiu]

Hi @oscarramirezs, thanks for your interest here.

If you want to crop both image and label, you could use CropForegroundd and set keys=["image", "bone1"], source_key="bone1".

MONAI/monai/transforms/croppad/dictionary.py

Lines 709 to 719 in 2daabf9

	class CropForegroundd(Cropd):
	"""
	Dictionary-based version :py:class:`monai.transforms.CropForeground`.
	Crop only the foreground object of the expected images.
	The typical usage is to help training and evaluation if the valid part is small in the whole medical image.
	The valid part can be determined by any field in the data with `source_key`, for example:
	- Select values > 0 in image field as the foreground and crop on all fields specified by `keys`.
	- Select label = 3 in label field as the foreground to crop on all fields specified by `keys`.
	- Select label > 0 in the third channel of a One-Hot label field as the foreground to crop all `keys` fields.
	Users can define arbitrary function to select expected foreground from the whole source image or specified
	channels. And it can also add margin to every dim of the bounding box of foreground object.

If you are still having issues, could you please provide a small piece of code and data so that I can reproduce your issue?
Thanks!

[KumoLiu]

Hi @oscarramirezs, may I know your MONAI version? Could you please try with MONAI v1.2? You can check the updated affine by using result_.affine. We have updated to MetaTensor after v0.9.
You can find more info here. https://github.com/Project-MONAI/MONAI/wiki/MetaTensor-guide

Hope it helps, thanks!

[oscarramirezs]

Version de MONAI: 1.0.0.
my script look fine?

[KumoLiu]

Since you didn't crop the image, affine in val_data["image_meta_dict"] will not update, you need to use the updated meta information after CropForeground.

[oscarramirezs]

"I'm sorry, it's not working. I'm a bit new to this. Can I ask you the las question?"

Do you have a simple example code that reads a NIfTI, applies the crop function, and saves the new NIfTI, updating the affine matrix? Here's my code, but it's not working becouse, The CropForegroundd transformation does not return metadata in the form of a dictionary with "image_meta_dict" key. Instead, it returns a tuple containing the cropped image and the bounding box coordinates

import os
import glob
import numpy as np
from monai.transforms import Compose, CropForegroundd
from monai.data import NiftiSaver
import nibabel as nib

Define the image directory

image_dir = r'C:\Users\Nifti1\val_cut_test\2EA840D22B5D_mimics_'

Search for NIfTI image files

cut_images = sorted(glob.glob(os.path.join(image_dir, "*.nii")))

Define the transformation

post_test = Compose([CropForegroundd(keys=["image"], source_key="image")])

Initialize the NIfTI saver

saver = NiftiSaver(output_dir=image_dir, output_postfix='_cropped', output_ext=".nii")

nifti_img = nib.load(image_path)
img_data = nifti_img.get_fdata()

transformed_data = post_test({"image": img_data})["image"]

Convert to a NumPy array of appropriate data type

transformed_data_np = transformed_data.detach().cpu().numpy().astype(np.float32)

Save the transformed image

output_image_path = os.path.join(image_dir, os.path.basename(image_path).replace(".nii", "_cropped.nii"))
nib.save(nib.Nifti1Image(transformed_data_np, transformed_data.affine ), output_image_path)

print("Transformation and saving completed.")

[KumoLiu]

Hi @oscarramirezs, if you are using MONAI v1.0, I think you can directly use Metatensor.
Here is a simple example.

import monai.transforms as mt
data = {"img": img_path, "lbl": lbl_path}

trans = mt.Compose([
    mt.LoadImaged(keys=['img', 'lbl'], ensure_channel_first=True),
    mt.CropForegroundd(keys="lbl", source_key="lbl"),
])

out = trans(data)
print(out['lbl'].affine) # check whether the affine has been updated

mt.SaveImaged(keys='lbl', resample=False, output_dir='/workspace/')(out)

For more end-to-end examples, you can refer to the tutorial repo.
Hope it helps, thanks!