[FEATURE] Question about Mapping ROI and Center Modality in brainles_preprocessing

[min9kwak]

Hi,

First of all, thank you for developing the brainles_preprocessing package. It has been incredibly helpful.

I have a question regarding the usage of the package. I would like to apply the ROI information, which corresponds to t1 and t2 modalities, to the space mapped by brainles_preprocessing. Is it possible to achieve this using the package? The ROI is also in .nii.gz format.

Additionally, in the tutorial, the center modality is specified as t1c. Is there a specific reason for this choice? Is it acceptable to use a different modality as the center?

Thank you!

[neuronflow]

Dear @min9kwak, thanks for your interest in our software. Can you elaborate on the ROIs? Are these masks coming with the T1 and T2 files? What do you want to do with them?

We chose T1c as "center" modality in the tutorial, as many hospitals record the native T1 files in a lower resolution. However, you can choose whatever modality you want to serve as "center". The moving images will be co-registered to this center modality.

[min9kwak]

Dear @neuronflow , thanks for your answer. I meant ROIs as the binary masks of certain modalities. It could come with any kind of modality. My goal was to transform the ROI masks following the same transformation (co-registration, atlas spacing, and correction) of the corresponding modalities. With this, we can maintain the ROI's physical location for further analysis in BraTS atlas. I could modify some classes in your current software to achieve the goals. It was really useful! I sincerely appreciate it.

[min9kwak]

Dear @neuronflow,
I have created a derivative work based on your project. In compliance with the AGPL-3.0 license, I have made the source code of this work publicly available, and you can find it at the following link.

I appreciate your excellent project and hope that my work will be useful. If there are any issues or concerns, please let me know.

Thank you.

[neuronflow]

Sweet, I see why this is useful. The current version of our preprocessing package saves all the transformation matrixes to enable such processing downstream.

I wonder what is the most elegant way to achieve what you want. I believe introducing a second class that is linked to a Modality might be a good idea?

Would you be interested in working on a PR to integrate this feature into our package?

[min9kwak]

Thanks for your suggestion. I could PR for the integration of the current implementation. Since the current version is under a separated directory, I should further modify for the code consistency.

[MarcelRosier]

The package now supports this kind of feature via the Transform class.
A possible workflow could look like this:

from brainles_preprocessing.preprocessor import AtlasCentricPreprocessor
from brainles_preprocessing.transform import Transform

# execute regular preprocessing pipeline and add parameter to save transformations
# ...
preprocessor = AtlasCentricPreprocessor(...)
preprocessor.run(
    ...,
    save_dir_transformations="path/to/transformations"
)


# use transforms for extra files like binary masks
transfom = Transform(
    transformations_dir="path/to/transformations",
)
transformed_segmentation = transfom.apply(
    target_modality_img="target.nii.gz", # path to the transformed modality image (result from the preprocessor)
    target_modality_name="t1c", # name of the modality that the mask belongs to
    moving_image="mask.nii.gz", # input mask
    output_img_path="transformed_mask.nii.gz", # transformed mask
    log_file_path="transform.log",
)

What do you think @min9kwak?
Feel free to re-open the issue if this is not a viable fix in you opinion.