Medical imaging implementation

[Samy-mri]

Hello everyone,

I've recently started using Anomalib to detect morphological abnormalities in T1-weighted MRI images, in patients with movement disorders.

To implement this I added a few lines of code to handle NIFTI reading (src/data/utils/image.py).
and to handle gray-scale images I am using 3 neighbouring slices to accomodate for the pretrained (RGB) models, which suited my application. Another trick would be to unsqueeze in the channel dimension.

I also added some code to handle non-int8 data (/src/post_processing/post_process.py and visualizer.py).

I am wondering if there is interest to implement this properly, which would also mean accomodating for single-channel uint16 data.

Happy to share what I have already.
Best,
Samy

[yarinbar]

Do you have any update or tips? I am trying to use it too on medical data :)