Flipd transform not working while being used in Post transform on the segmented image

[ArpanGyawali]

I am working in SkullStripping app using segmentation model and the predicted segmented image is flipped as shown in image

and i get this warning while loading segmentation in ITKSnap

So to reorient the segmentation, i tried to used Flipd transform in post processing as shown in code below but there was no change in the output. I also triend Orientd but no changes.

def post_process(self, pre_transforms: Compose, out_dir: str = "./prediction_output") -> Compose:
        """Composes transforms for postprocessing the prediction results."""

        pred_key = self._pred_dataset_key
        return Compose(
            [
                EnsureTyped(keys=pred_key),
                Activationsd(keys=pred_key, softmax=True),
                AsDiscreted(keys=pred_key, argmax=True),
                KeepLargestConnectedComponentd(keys=pred_key, applied_labels=1),
                Flipd(keys=[pred_key], spatial_axis=2),
                SaveImaged(
                    keys=pred_key,
                    output_dir=out_dir,
                    output_postfix="seg",
                    output_dtype=uint8,
                ),
            ]
        )

[KumoLiu]

Hi @ArpanGyawali, I suggest using Orientationd instead of Flipd here.
You can refer to the "Reorientation to a designated axes codes" section in this tutorial https://github.com/Project-MONAI/tutorials/blob/main/modules/3d_image_transforms.ipynb.
Thanks!

[ArpanGyawali]

Thank you @KumoLiu
I had already tried orientationd too and there was no difference in the prediction

def post_process(self, pre_transforms: Compose, out_dir: str = "./prediction_output") -> Compose:
        """Composes transforms for postprocessing the prediction results."""

        pred_key = self._pred_dataset_key
        return Compose(
            [
                EnsureTyped(keys=pred_key),
                Activationsd(keys=pred_key, softmax=True),
                AsDiscreted(keys=pred_key, argmax=True),
                KeepLargestConnectedComponentd(keys=pred_key, applied_labels=1),
                # Flipd(keys=pred_key, spatial_axis),
                Orientationd(keys=pred_key, axcodes='RPI', labels=(('R', 'L'), ('P', 'A'), ('I', 'S'))),
                SaveImaged(
                    keys=pred_key,
                    output_dir=out_dir,
                    output_postfix="seg",
                    output_dtype=uint8,
                ),
            ]
        )

this was what i did.

[KumoLiu]

Hi @ArpanGyawali, SaveImage resample image based on the spatial_shape (and original_affine) from metadata by default. Could you please try to set resample=False and try again?

MONAI/monai/transforms/io/array.py

Lines 320 to 321 in f519699

	resample: whether to resample image (if needed) before saving the data array,
	based on the `spatial_shape` (and `original_affine`) from metadata.

Thanks!

[ArpanGyawali]

Thank you @KumoLiu for your reply
Yes but without resampling, the size of the image and the segmentation differ.

What i did was added resample=False on SaveImaged from above mentioned code.

[ArpanGyawali]

@KumoLiu

Converted Image object metadata:
SeriesInstanceUID: 1.3.12.2.1107.5.1.4.73615.30000018083022163054400025684, type <class 'str'>
SeriesDate: 20230101, type <class 'str'>
SeriesTime: 000000, type <class 'str'>
Modality: CT, type <class 'str'>
SeriesDescription: DE AXIAL THIN SOFT  F_0.6, type <class 'str'>
PatientPosition: HFS, type <class 'str'>
SeriesNumber: 8, type <class 'int'>
row_pixel_spacing: 0.576171875, type <class 'float'>
col_pixel_spacing: 0.576171875, type <class 'float'>
depth_pixel_spacing: 0.5, type <class 'float'>
row_direction_cosine: [1.0, 0.0, 0.0], type <class 'list'>
col_direction_cosine: [0.0, 1.0, 0.0], type <class 'list'>
depth_direction_cosine: [0.0, 0.0, 1.0], type <class 'list'>
dicom_affine_transform: [[ 5.76171875e-01  0.00000000e+00  0.00000000e+00 -1.47211914e+02]
 [ 0.00000000e+00  5.76171875e-01  0.00000000e+00 -3.41211914e+02]
 [ 0.00000000e+00  0.00000000e+00  5.00000000e-01  9.77500000e+02]
 [ 0.00000000e+00  0.00000000e+00  0.00000000e+00  1.00000000e+00]], type <class 'numpy.ndarray'>
nifti_affine_transform: [[-5.76171875e-01 -0.00000000e+00 -0.00000000e+00  1.47211914e+02]
 [-0.00000000e+00 -5.76171875e-01 -0.00000000e+00  3.41211914e+02]
 [ 0.00000000e+00  0.00000000e+00  5.00000000e-01  9.77500000e+02]
 [ 0.00000000e+00  0.00000000e+00  0.00000000e+00  1.00000000e+00]], type <class 'numpy.ndarray'>
StudyInstanceUID: 1.2.840.113845.11.1000000001927670467.20180831120329.5375294, type <class 'str'>
StudyID: 118CTA1, type <class 'str'>
StudyDate: 20230101, type <class 'str'>
StudyTime: 000000, type <class 'str'>
StudyDescription: CT Angio COW (+/- Brain) - (First 12 Months), type <class 'str'>
AccessionNumber: 2023R0000000-1, type <class 'str'>
selection_name: (.*?), type <class 'str'>
2023-08-07 19:12:12,433 INFO image_writer.py:197 - writing: /home/gsharma/SkullStripDeploy/OutSeg/prediction_output/1.3.12.2.1107.5.1.4.73615/1.3.12.2.1107.5.1.4.73615_seg.nii.gz
Output Seg image numpy array shaped: (371, 295, 295)
Output Seg image pixel max value: 1
Output Seg image pixel min value: 0

Here are the logs while the operations happen

[KumoLiu]

Could you please try to use Invertd to invert the previously applied transforms if a different shape is generated from that?