Save CenterSpatialCropd transform as nii

[carlpe]

I want to crop a nifti image from original size to a given size in the middle of the volume using CenterSpatialCropd transform and save it using SaveImaged.

Original image:
1,512,512,512

Here I apply the transforms and save.
This is the output:

It seems like the volume has been correctly cropped from the center as I want, but the borders look strange and "stretched out".

What can I do to fix it so that it look like this:

Version:

MONAI version: 1.0.1
Numpy version: 1.19.2
Pytorch version: 1.9.1
MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False
MONAI rev id: 8271a193229fe4437026185e218d5b06f7c8ce69
MONAI __file__: C:\Users\carlk\anaconda3\envs\monai\lib\site-packages\monai\__init__.py

Optional dependencies:
Pytorch Ignite version: NOT INSTALLED or UNKNOWN VERSION.
Nibabel version: 3.2.1
scikit-image version: 0.18.3
Tensorboard version: 2.6.0
gdown version: 4.0.2
TorchVision version: 0.10.1
tqdm version: 4.62.3
lmdb version: 1.2.1
psutil version: 5.8.0
pandas version: 1.3.3
einops version: 0.3.2
transformers version: 4.11.3

Code:

from monai.utils import first
from monai.config import print_config
from monai.transforms import (
    EnsureChannelFirstd,
    Compose,
    LoadImaged,
    SaveImaged,
    CenterSpatialCropd,
    Spacingd,
)
from monai.data import (
    CacheDataset,
    DataLoader,
    Dataset,
    list_data_collate,
    decollate_batch,
)
import torch
import numpy as np
import os
import glob

print_config()
torch.cuda.get_device_name(0)

device = "cuda:0"

root_dir = r"D:\Documents\monai"
path_to_target = r"D:\Documents\temp"

org_img = sorted(glob.glob(os.path.join(path_to_target, ".", "aaaaa.nii")))
data_dicts = [{"image": image_name} for image_name in zip(org_img)]
org_files = data_dicts[:1]

print(len(org_files))
print("------------")
print("last img:" + str(org_files))
print("------------")

#

nii_transforms = Compose(
    [
        LoadImaged(keys=["image"]),
        EnsureChannelFirstd(keys=["image"]),
        Spacingd(
            keys=["image"],
            pixdim=(1.0, 1.0, 1.0),
            mode=["nearest"],
            padding_mode="border",
        ),
    ]
)

check_ds = Dataset(data=org_files, transform=nii_transforms)
check_loader = DataLoader(check_ds, batch_size=1)
check_data = first(check_loader)
image = check_data["image"][0]
print(f"image shape: {image.shape}")

nii_ds = CacheDataset(
    data=org_files, transform=nii_transforms, cache_rate=1.0, num_workers=0
)
nii_loader = DataLoader(nii_ds, batch_size=1, shuffle=True, num_workers=0)

post_transforms = Compose(
    [
        CenterSpatialCropd(keys=["image"], roi_size=(320, 320, 320)),
        SaveImaged(
            keys="image",
            output_dir="./out",
            separate_folder=False,
            output_ext=".nii.gz",
            output_postfix="cropped",
            resample=True,
            padding_mode="border",
        ),
    ]
)

device = torch.device("cuda:0")

with torch.no_grad():
    for data_dicts in nii_loader:
        inputs = data_dicts["image"].to(device)
        data_dicts = [post_transforms(i) for i in decollate_batch(data_dicts)]

Thank you!

[wyli]

I think you can set resample=False in SaveImaged

[carlpe]

It worked @wyli

:) Thank you!