SaveImaged Saving NIFTI files does not work correctly

[bxdgit]

Describe the bug
I use SaveImaged to save the file, which changes the spacing information of the file。image meta spacing(0.1540, 0.1540, 1.0000)，But the spacing written in the file is (0.1540, 1.0000, 1.0000)

To Reproduce
The following code can reproduce the problem

import os
import argparse
import numpy as np
from monai.config import KeysCollection
from monai.data import Dataset, DataLoader, decollate_batch
from monai.inferers import sliding_window_inference
from monai.networks.nets import UNet
from monai.transforms import (
    MapTransform,
    Compose,
    Activationsd,
    AsDiscreted,
    Invertd,
    AddChanneld,
    LoadImaged,
    LoadImage,
    NormalizeIntensityd,
    Orientationd,
    Spacingd,
    EnsureChannelFirstd,
    SaveImaged,
    SqueezeDimd,
    ScaleIntensityd,
    CropForegroundd,
    EnsureTyped,
    ToDeviced
)
import torch


class Permute(MapTransform):
    def __init__(self, keys: KeysCollection, allow_missing_keys: bool = False) -> None:
        super().__init__(keys, allow_missing_keys)
    def __call__(self, data):
        d = dict(data)
        # print(d["pred"].meta)
        # print(d["image"].meta)
        # d["pred"].meta = d["image"].meta
        for key in self.keys:
            if key == "pred":
                print(d[key].shape)
                print(type(d[key]))
                d[key] = d[key].permute(1,2,0)
                print(d[key].shape)
                print(d[key].meta["spacing"])
               
        return d

def run_inference(model_file:str, input, output_dir=None,  VAL_AMP=False, print_log=True):
    if output_dir is None and os.path.isdir(input):
        output_dir = input
    # threshold at -1024
    def threshold_at_one(x):
        return x > -1024

    pre_transforms = Compose(
        [
            LoadImaged(keys=["image"]),
            SqueezeDimd(keys=['image'], dim=-1),
            EnsureChannelFirstd(keys=["image"]),
            # CropForegroundd(keys=["image"],source_key="image"),
            CropForegroundd(keys=["image"],source_key="image",select_fn=threshold_at_one,margin=0),
            EnsureTyped(keys=["image"])
        ]
    )

    post_transforms = Compose([
        EnsureTyped(keys="pred"),
        Activationsd(keys="pred", sigmoid=True),
        AsDiscreted(keys="pred", threshold=0.5),
        Permute(keys="pred"),
        Invertd(
            keys="pred",
            transform=pre_transforms,
            orig_keys="image",
            nearest_interp=False,
            to_tensor=True,
        ),

    
        SaveImaged(
            keys="pred",
            output_dir=output_dir,
            output_postfix="label",
            output_dtype=np.uint8,
            resample=False,
            separate_folder=False,
            print_log=print_log,
        ),
    ])

    if os.path.isdir(input):
        images = [
            {
                "image": os.path.join(input, fname),
            }
            for fname in os.listdir(input)
            if fname.endswith(".nii.gz")
        ]
    else:
        images=[{"image":input}]
  
    dataset = Dataset(data=images, transform=pre_transforms)
    data_loader = DataLoader(dataset, batch_size=1, shuffle=False, num_workers=0)

    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

    model = UNet(
            2,
            in_channels=1,
            out_channels=1,
            channels=(16, 32, 64, 128, 256),
            strides=(2, 2, 2, 2),
            num_res_units=2,
        ).to(device)
    # model.load_state_dict(
    #     torch.load(model_file))
    model.eval()
    # define inference method
    def inference(input):

        def _compute(input):
            return sliding_window_inference(
                inputs=input,
                roi_size=(192, 192),
                sw_batch_size=4,
                predictor=model,
            )

        if VAL_AMP:
            with torch.cuda.amp.autocast():
                return _compute(input)
        else:
            return _compute(input)

    with torch.no_grad():
        for data in data_loader:
            image = data["image"].to(device)
            data["pred"] = inference(image)
            # post_transforms(data)
            _ = [post_transforms(i) for i in decollate_batch(data)]


def main():
    run_inference("model file","inputfile", "output",print_log=True)

if __name__ == "__main__":
    main()

The input parameter model file can be null. here is my input file
img949.zip

Expected behavior
I expect the spacing information of the saved file to be the same as the image meta information.
As you can see from the screenshot below, it changes the spacing information

Screenshots

Environment

Ensuring you use the relevant python executable, please paste the output of:

================================
Printing MONAI config...
================================
MONAI version: 1.1.0
Numpy version: 1.21.5
Pytorch version: 1.11.0+cu113
MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False
MONAI rev id: a2ec3752f54bfc3b40e7952234fbeb5452ed63e3
MONAI __file__: C:\Users\Bxd\AppData\Local\Programs\Python\Python39\lib\site-packages\monai\__init__.py

Optional dependencies:
Pytorch Ignite version: 0.4.10
Nibabel version: 3.2.2
scikit-image version: 0.19.2
Pillow version: 9.1.0
Tensorboard version: 2.8.0
gdown version: NOT INSTALLED or UNKNOWN VERSION.
TorchVision version: 0.12.0+cpu
tqdm version: 4.64.0
lmdb version: NOT INSTALLED or UNKNOWN VERSION.
psutil version: 5.9.0
pandas version: 1.2.4
einops version: 0.5.0
transformers version: NOT INSTALLED or UNKNOWN VERSION.
mlflow version: NOT INSTALLED or UNKNOWN VERSION.
pynrrd version: 0.4.3

For details about installing the optional dependencies, please visit:
    https://docs.monai.io/en/latest/installation.html#installing-the-recommended-dependencies


================================
Printing system config...
================================
System: Windows
Win32 version: ('10', '10.0.19044', 'SP0', 'Multiprocessor Free')
Win32 edition: Professional
Platform: Windows-10-10.0.19044-SP0
Processor: Intel64 Family 6 Model 167 Stepping 1, GenuineIntel
Machine: AMD64
Python version: 3.9.7
Process name: python.exe
Command: ['C:\\Users\\Bxd\\AppData\\Local\\Programs\\Python\\Python39\\python.exe', '-c', 'import monai; monai.config.print_debug_info()']
Open files: [popenfile(path='C:\\Program Files\\WindowsApps\\Microsoft.LanguageExperiencePackzh-CN_19041.57.180.0_neutral__8wekyb3d8bbwe\\Windows\\System32\\zh-CN\\KernelBase.dll.mui', fd=-1), popenfile(path='C:\\Program Files\\WindowsApps\\Microsoft.LanguageExperiencePackzh-CN_19041.57.180.0_neutral__8wekyb3d8bbwe\\Windows\\System32\\zh-CN\\kernel32.dll.mui', fd=-1)]
Num physical CPUs: 8
Num logical CPUs: 16
Num usable CPUs: 16
CPU usage (%): [6.2, 4.5, 10.2, 1.9, 5.1, 18.5, 16.6, 51.0, 5.1, 3.2, 4.5, 5.1, 3.2, 4.5, 2.5, 14.6]
CPU freq. (MHz): 3600
Load avg. in last 1, 5, 15 mins (%): [0.0, 0.0, 0.0]
Disk usage (%): 73.2
Avg. sensor temp. (Celsius): UNKNOWN for given OS
Total physical memory (GB): 127.8
Available memory (GB): 108.9
Used memory (GB): 18.9

================================
Printing GPU config...
================================
Num GPUs: 1
Has CUDA: True
CUDA version: 11.3
cuDNN enabled: True
cuDNN version: 8200
Current device: 0
Library compiled for CUDA architectures: ['sm_37', 'sm_50', 'sm_60', 'sm_61', 'sm_70', 'sm_75', 'sm_80', 'sm_86', 'compute_37']
GPU 0 Name: NVIDIA GeForce RTX 3090 Ti
GPU 0 Is integrated: False
GPU 0 Is multi GPU board: False
GPU 0 Multi processor count: 84
GPU 0 Total memory (GB): 24.0
GPU 0 CUDA capability (maj.min): 8.6

Additional context
Add any other context about the problem here.

[KumoLiu]

Hi @bxdgit, seem your Permute caused this issue. May I know why you permute the prediction? The 1 is channel dim I think. Could you please try to remove Permute(keys="pred") in post_transforms and try to save it again? Thanks!

[bxdgit]

Hi @bxdgit, seem your Permute caused this issue. May I know why you permute the prediction? The 1 is channel dim I think. Could you please try to remove Permute(keys="pred") in post_transforms and try to save it again? Thanks!

Glad to get your reply，I remove Permute(keys="pred") in `post_transforms, but the result is still the same

[KumoLiu]

Hi @bxdgit, oh I got it. An additional issue is that SqueezeDimd(keys=['image'], dim=-1), in the pre_transforms, it changes your input spacing. You could remove this line and update EnsureChannelFirstd with the specified channel dim.

EnsureChannelFirstd(keys=["image"], channel_dim=-1),

Hope it can help you, thanks! Just move it to the discussion.

[bxdgit]

hi @KumoLiu ，Thank you, the problem is solved！