issue with deep_flash utility

[aa2782]

I was trying to implement the 'deep_flash' utility from the ANTspynet to segment the Medial Temporal Lobe(MTL) from my T1 MRI images. I ran it using both raw and skull-stripped images but the issue remains. For some reason, I am not getting the segmented images correctly. After running the code, I got a dictionary with a bunch of probability_images and one segmented_image, which all contain zeros. I am not sure what steps I could perform to get my segmented regions correctly.
If there is a particular way to obtain the segmented images using the results that I got, please let me know.

[ntustison]

What happens when you run this example?

[aa2782]

This is how i ran the code. I had a raw ANTs image. After running, I got this result:

I got this dictionary with a lot of key/value pairs similar to above image. Then, I wanted to see if any of the values have non-zero voxels in them, which would mean that the algorith did the segmentation. I used nibabel to do the voxel count like this and keep getting 0 like this:

In the antspynet documentation, it says after running the code i'll get a list consisting of the segmentation image and probability images for each label and foreground.

I was expecting I'd be able to plot the segmentation image for each subregion right away using ants.plot but when I do, its just a black figure.

I'm pretty new to antspynet, so if there is anything i need to install or if there is a special way to view the segmented images using the results i got, please let me know. Also, I ran this both on Google Colab and local jupyter notebook in linux.

[ntustison]

I don't see where you ran the code exactly as it is shown in the example. Did you actually run the example?

[aa2782]

Would it be easier for you to analyze if I sent you my code and the raw ants image?

[aa2782]

To answer your question, yes i ran the code in google colab. Infact I just did again and the result is still the same.

[aa2782]

I just realized where the problem was. I had reoriented my ANTs file when I loaded it. When I removed the orientation parameter, the deep_flash function worked just fine. Thank you so much for your time!!

[ntustison]

Great. But given that it was an orientation issue and that I see you changed the default do_preprocessing = True to do_preprocessing = False in your run above, my guess is that you never ran the example like I had requested. In the future, my suggestions would be to:

1) Run the examples exactly as written. In this case, it's a simple cut-and-paste. 2) In general, always do an initial run using the default parameters. In this case, I strongly recommend against using do_preprocessing=False.