Spacingd does not work properly together with Transposed

[TuanDTr]

Describe the bug
Spacingd results in a different size if it is used together with Transposed. This is the compose of transform that I want to apply on the image:

transforms = Compose([
    LoadImaged(keys=("image")),
    Transposed(keys=("image"), indices=None), # change to (z, y, x) dim
    EnsureChannelFirstd(keys=("image")),
    Spacingd(keys=("image"), pixdim=(0.7, 0.7, 0.7)),
])

If I remove Transposed from the list of transforms, I get the correct target shape and spacing. Please see details in below sections.

To Reproduce

from monai.transforms import LoadImaged, Transposed, Spacingd, Compose, EnsureChannelFirstd

image_path = "Medical_Decathlon_06_Lung_Segmentation/imagesTr/lung_003.nii.gz"

# This image has the shape of (512, 512, 288) and spacing of (0.9375, 0.9375, 1.2456597089767456)

transforms_with_transpose = Compose([
    LoadImaged(keys=("image")),
    Transposed(keys=("image"), indices=None),
    EnsureChannelFirstd(keys=("image")),
    Spacingd(keys=("image"), pixdim=(0.7, 0.7, 0.7)),
])

transforms_without_transpose = Compose([
    LoadImaged(keys=("image")),
    EnsureChannelFirstd(keys=("image")),
    Spacingd(keys=("image"), pixdim=(0.7, 0.7, 0.7)),
])

datum_dict = {"image": image_path}
datum_dict_with_transpose = transforms_with_transpose(datum_dict)
datum_dict_without_transpose = transforms_without_transpose(datum_dict)

print("Image size after transforming with transpose: ", datum_dict_with_transpose["image"].shape)
# returns: Image size after transforming with transpose:  torch.Size([1, 385, 685, 910])
print("Image size after transforming without transpose: ", datum_dict_without_transpose["image"].shape)
# returns: Image size after transforming without transpose:  torch.Size([1, 685, 685, 512])

Expected behavior
The transforms with Spacingd and Transposed should result in the shape of (1, 512, 685, 685)

Environment

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

python -c 'import monai; monai.config.print_debug_info()'

================================
Printing MONAI config...
================================
MONAI version: 1.1.0
Numpy version: 1.23.4
Pytorch version: 1.12.1
MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False
MONAI rev id: a2ec3752f54bfc3b40e7952234fbeb5452ed63e3
MONAI __file__: /home/tuan_truong_bayer_com/miniconda3/envs/seg_clone/lib/python3.9/site-packages/monai/__init__.py

Optional dependencies:
Pytorch Ignite version: 0.4.8
Nibabel version: 4.0.2
scikit-image version: 0.19.3
Pillow version: 9.3.0
Tensorboard version: 2.10.1
gdown version: 4.5.3
TorchVision version: 0.13.1
tqdm version: 4.64.1
lmdb version: 1.3.0
psutil version: 5.9.3
pandas version: 1.4.4
einops version: 0.5.0
transformers version: 4.24.0
mlflow version: 1.30.0
pynrrd version: 1.0.0

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: Linux
Linux version: Debian GNU/Linux 10 (buster)
Platform: Linux-4.19.0-12-cloud-amd64-x86_64-with-glibc2.28
Processor: 
Machine: x86_64
Python version: 3.9.13
Process name: python
Command: ['python', '-c', 'import monai; monai.config.print_debug_info()']
Open files: []
Num physical CPUs: 8
Num logical CPUs: 16
Num usable CPUs: 16
CPU usage (%): [5.1, 4.8, 4.8, 4.5, 99.7, 4.8, 4.8, 4.8, 4.1, 4.5, 5.1, 5.1, 4.5, 4.5, 4.8, 5.1]
CPU freq. (MHz): 2000
Load avg. in last 1, 5, 15 mins (%): [6.8, 5.8, 4.8]
Disk usage (%): 55.0
Avg. sensor temp. (Celsius): UNKNOWN for given OS
Total physical memory (GB): 59.0
Available memory (GB): 55.8
Used memory (GB): 2.5

================================
Printing GPU config...
================================
Num GPUs: 1
Has CUDA: True
CUDA version: 11.6
cuDNN enabled: True
cuDNN version: 8302
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: Tesla T4
GPU 0 Is integrated: False
GPU 0 Is multi GPU board: False
GPU 0 Multi processor count: 40
GPU 0 Total memory (GB): 14.8
GPU 0 CUDA capability (maj.min): 7.5

[KumoLiu]

Hi @beantr96, thanks for pointing this out. The reason is that in Transposed it didn't update the affine. You could try to use "Orientationd" to change the orientation in medical images.

transforms = Compose([
    LoadImaged(keys=("image")),
    EnsureChannelFirstd(keys=("image")),
    Orientationd(keys=('image'), axcodes="LPI"), # completely opposite axcodes
    Spacingd(keys=("image"), pixdim=(0.7, 0.7, 0.7)),
])

Do we need to add the support for the update of affine in Transposed? cc @wyli @Nic-Ma

[TuanDTr]

Hi @KumoLiu, many thanks for the quick reply. As far as I am aware of, Orientationd does not really change the shape of the tensor. For implementation reason in our pipeline, we want the tensor shape to be (D, W, H) instead of (H, W, D) by default with Monai's LoadImaged. We once tried Monai's ITK reader with the reverse_indexing flag but we observed the inconsistent behavior (but this is another story though). So we thought of using the Transposed as an option. Is there any alternative solution that I can use as a workaround to this problem?

[wyli]

I think you should use Orientationd in this case.

for this input:

>>> import torch
>>> from monai.data import MetaTensor
>>> x = MetaTensor(torch.zeros(1, 30, 40, 50))

you can get the image's current orientation using nibabel:

>>> import nibabel as nib
>>> nib.aff2axcodes(x.affine)
('R', 'A', 'S')

then set the output orientation to SAR or try other orientations:

>>> from monai.transforms import Orientation
>>> Orientation("SAR")(x).shape
torch.Size([1, 50, 40, 30])
>>> Orientation("RAS")(x).shape
torch.Size([1, 30, 40, 50])
>>> Orientation("RSA")(x).shape
torch.Size([1, 30, 50, 40])