Infected cell detection

[tischi]

Since the cell segmentation works very well, I think we could try make use of all the pixels outside the cells in order to determine an automated intensity threshold for whether a cell is infected or not.

For example something like:
some_quantile_inside_cell > bg_median + N * bg_mad

Each well could have its own computation of the bg values but maybe we could even use the same across the whole plate.

[imagirom]

For comparison, I am currently using:
some_top-k_inside_cell > bg_mean + fixed_offset

What are N and bg_mad in your equation?

[tischi]

1. bg_mad = https://en.wikipedia.org/wiki/Median_absolute_deviation
   (robust version of sdev)
2. N is a parameter, i.e. how many mad (or sdev) do we want to be away from the background.

So essentially your fixed offset would be replaced by some automated measure of the noise level in the background in the respective plate. Thus it would an adaptive threshold.