Add Medical Image Denoising Demo Using Anisotropic Diffusion

[m-wisell]

Summary
I'd like to add a medical image processing demo implementing anisotropic diffusion (Perona-Malik model) for edge-preserving denoising of mammograms and other medical images. I believe this would be the first biomedical imaging example in the demo collection. This would not only demonstrate Dolfinx's applicability to biomedical research, but also show implementation of nonlinear, spatially-varying diffusion coefficients.

Proposed Contribution

1. Demo Program

• implements Perona-Malik diffusion
• uses publicly available medical datasets, such as CBIS-DDSM
• before/after visualization

2. Documentation

• PDE formulation and edge-preserving properties
• image-to-mesh conversion workflow
• parameter selection guidelines

3. Testing will include unit tests for diffusion operator and edge preservation

Questions
Before starting, I'd appreciate any guidance or preferences regarding:

1. Should this go in python/demo/ or a separate python/demo/medical_imaging/ directory?
2. For dependencies, is PIL/scikit-image acceptable for image I/O?
3. Should I include some of the same images from a dataset (CBIS-DDSM) or just the script for the demo?

[chrisrichardson]

Hi @m-wisell - this sounds great. To answer some of your questions: I would think that a sufficently complex demo, as you are suggesting, could contain multiple files, and therefore need its own directory.
Adding extra dependencies is not a huge problem, but best avoided if possible, as it makes the CI testing more complex.
Finally, we wouldn't add large datasets to the main repository, maybe just a reference that can be downloaded at runtime.

It would be worth discussing in more detail on the fenics slack - e.g. in the #development channel.