Projections module: add biological relevance

[tischi]

@manerotoni @tibuch @grinic

Do you have some suggestions for how to improve the biology of the projections module?

Like:

1. When would you use max projection and when sum projection? Would we have example data for this (not for tomorrow, but for the next course)?
2. Why and when using projections at all?

Just some written text here would be already nice for tomorrow such that Arif can tell the participants.

[manerotoni]

Hi @maulakhan, Projection helps in dimension reduction as it is intrinsically hard to visualize a 3D image. It is not limited to XYZ, you can also for example perform a projection along T of a XYT image and so get some information of the movement in XY.

Projections in general is one way to render a multidimensional image on paper. Each projection type will emphasize different features of our data set. It is up to the scientist to decide what is best for their data set (typical answer: What should I use? It depends!)

Maximal projection In Light microscopy what is bright is relevant. Maximal projection shows in 2D image all the relevant information. Max projection is done very often when showing an image in a publication or presentation.

Min projection In Electron microscopy what is relevant is dark. So a min projection can help showing the 3D data (not 100% sure if people really do this)

Sum projection For instance to show the total amount of signal in a 3D object. Caveat you also add up background if you did not subtract before.

Other type of projections I do not use so much.

[tibuch]

~Sometimes a single layer of cells is not exactly flat and you would acquire a small stack to make sure that you cover it. Then you can use a max projection to get a nice 2D image.~ Edit: @grinic's wording is much nicer! :arrow_double_down:

Slightly more involved would be to compute the variance image for each slice and then take the pixels from the original stack where the variance is maximized over the whole stack. I think this is also known as best focus projection.

[grinic]

A more technical example is a cell mono layer that is slightly tilted relative to the focal plane. Then you would acquire a Z stack and make a max projection to recover all information in the same 2D image.

[grinic]

~Sometimes a single layer of cells is not exactly flat and you would acquire a small stack to make sure that you cover it. Then you can use a max projection to get a nice 2D image.~ Edit: @grinic's wording is much nicer! ⏬

I realize just now my comment was redundant, sorry

[tischi]

Thank you all for the nice suggestions!

I updated the preamble of activity 1 to try including some of these notions:

### Explore max and sum projections
- Open [xyz_16bit__spots.tif](https://github.com/NEUBIAS/training-resources/raw/master/image_data/xyz_16bit__spots.tif)
- This 3-D image contains sparse dot-like fluorescence signals
- Appreciate that viewing data in 3-D is intrinsically challenging
- Understand that due to the fact that the spots are brighter than the background a maximum projection along the z-axis can be useful to obtain a quick overview of all spots in the image
- Discuss under which circumstances the number of dots in the projections equals the total number of spots
- Also perform a sum projection and discuss why this is may be better than a maximum projection in terms of quantifying the total signal contained in the spots
- Understand that the data type of a sum projection typically needs to be adapted
- Project along the x and y axis
- Understand the projections along the x or y axis typically yield an anisotropic images (due to an initial xy/z anisotropy)
- Resample or rescale the x and y projections for correct appearance in physical space

I am thinking now that maybe in general we should try to put more bio-physical information like this into all of our activity preambles.

@maulakhan if you agree with them it would be nice if you could discuss these points with the participants tomorrow.

[tibuch]

Another example would be background subtraction in movies. Taking an average projection over time reveals the static background in a movie. This average projection can then be subtracted from all movie frames and only the moving objects remain. This works very well if the illumination is constant over all movie frames. Which can be tricky for live microscopy.

[manerotoni]

In the line of @tibuch, I think correction of illumination (transmission, wide field) can make use of mean projections. Acquire many transmission images, compute mean, divide images by "background".

One suggestion would could be to improve the presentation slides, with more examples. Now it is very technical, we should add example applications of what you can do with it.

[tischi]

Now it is very technical, we should add example applications of what you can do with it.

You mean adding more acitivities?

[tischi]

• [ ] we should add some biologically interesting examples to the Figure