Surface measurements are very tricky

[tischi]

see here: https://github.com/tischi/spindle3d/issues/30 we should add this to the training material somehow. cc @manerotoni

[manerotoni]

Hello @tischi, looking now at the object measurements and I realize that surface measurements is not so straight forward in python. I think you have to compute for every object the isosurfaces meshes and then compute the area using: https://scikit-image.org/docs/stable/api/skimage.measure.html#skimage.measure.marching_cubes

and

https://scikit-image.org/docs/stable/api/skimage.measure.html#skimage.measure.mesh_surface_area

Not 100% sure and I want to try.

It would be nice to include a module on surface measurements with all caveats. What do you think?

If yes I create an issue with an extra module development part.

[tischi]

Cool that you are exploring this!

Whether adding an extra module or adding this as another activity of the shape measurements is a tough choice. I feel if you want to introduce the concept of meshes it should probably go into a separate module....

[manerotoni]

I think it should be a separate module. Is the computation of surface in morpholibJ also relying on meshes?

[tischi]

I don't know. @dlegland do you know ☝️ ?

[dlegland]

Hi,

the surface area measurement in MorphoLibJ relies on counting local 2-by-2-by-2 configurations of voxels, and associating a weight to each of them. Weights are computed according to a discrete version of Crofton formula.

There are many methods for computing surface area... using meshes is not always the best! My advices:

• if your input are grayscale or intensity images, and that you can extract an isosurface (using for example MarchingCube algorithm), then you can think on computing surface are of the mesh
• if your input is binary (or label) image, do not use marching cubes!! As the faces have only a limited number of orientation, the resulting surface area is largely overestimated. This has strong implication when computing shape factors based on surface area. Instead, use local configuration based method. Typical references are Lindblad, Mullikin, or Ohser (book on microstrutures).

I have a paper draft about this with the references and some accuracy evaluations, I will try to put in on bioarxiv soon (hope in the beginning of this summer).

David