Output file generated by transform_images is corrupted.

[grst]

• wsireg version: 0.3.8
• Python version: 3.10.12
• Operating System: Scientific Linux el7

Description

I am trying to register these two images, following the tutorial.

When saving the aligned images with transform_images(), the generated output ome.tiff file is corrupted. Opening with napari fails, when I open it in Zeiss ZEN it looks like this:

As a side, the function docstring suggest that images can be saved to zarr by specifying file_writer="zarr", but this doesn't work and the source code suggests this is not implemented.

What I Did

The input images are available here: input_images.zip

# initialize registration graph
reg_graph = WsiReg2D("my_reg_project", wsireg_dir)
reg_graph.add_modality(
    "img0",
    wsireg_dir / "./img0.tif",
    preprocessing={
        "image_type": "BF",
        "as_uint8": True,
        "invert_intensity": True,
    },
)

reg_graph.add_modality(
    "img1",
    wsireg_dir / "./img1.tif",
    preprocessing={
        "image_type": "BF",
        "as_uint8": True,
        "invert_intensity": True,
    },
)

reg_graph.add_reg_path(
    "img0",
    "img1",
    thru_modality=None,
    reg_params=[
        "rigid",
        "affine",
        # "nl",
    ],
)

reg_graph.register_images()
reg_graph.save_transformations()
reg_graph.transform_images(file_writer="ome.tiff")

[NHPatterson]

Hi Gregor,

I think your registration worked fine, I was able to do it on my end. There are just issues with the various software you have used to view the images.

Opening with napari fails

This is because the standard wsireg output is a pyramid OME-TIFF (even for small images) which the tiff readers of napari likely don't understand.

Zeiss ZEN fails

Likewise, Zeiss ZEN doesn't understand this format.

The images open fine in QuPath and if you read them as numpy arrays, you can open them in napari. Below extends your current script and shows how to open the images in napari. There are some additional things you can do if you are working with WSIs to open image pyramids in napari.

from wsireg import WsiReg2D
import napari
from tifffile import imread

img0_fp = "/Users/nhp/Downloads/input_images/img0.tif"
img1_fp = "/Users/nhp/Downloads/input_images/img1.tif"
out_dir = "/Users/nhp/Downloads/input_images"

# initialize registration graph
reg_graph = WsiReg2D("my_reg_project", out_dir)
reg_graph.add_modality(
    "img0",
    img0_fp,
    preprocessing={
        "image_type": "BF",
        "as_uint8": True,
        "invert_intensity": True,
    },
)

reg_graph.add_modality(
    "img1",
    img1_fp,
    preprocessing={
        "image_type": "BF",
        "as_uint8": True,
        "invert_intensity": True,
    },
)

reg_graph.add_reg_path(
    "img0",
    "img1",
    thru_modality=None,
    reg_params=[
        "rigid",
        "affine",
        # "nl",
    ],
)

reg_graph.register_images()
reg_graph.save_transformations()
out_images = reg_graph.transform_images(file_writer="ome.tiff")

# read in transformed image and original target image
img0_tformed = imread(out_images[0])
img1 = imread(img1_fp)

viewer = napari.Viewer()
viewer.add_image(img0_tformed, rgb=True)
viewer.add_image(img1, rgb=True)

https://github.com/NHPatterson/wsireg/assets/17855764/b3c61644-f29b-40cd-9c43-3def32291dc7

All the best, Heath

[grst]

Hi Heath,

thank you so much for your quick response! This works for me :)

Cheers, Gregor