2. Spot prediction

This is the step performed before detecting the centers of the spots and the is a semantic segmentation masks of the areas were SpotMAX will later look for spots.

This means that if an area in the image is marked as background in this step, spots will not be searched and detected there.

Segmentation of the spots image is performed in two steps:

  1. Image pre-processing

  2. Semantic segmentation

2.1. Image pre-processing

SpotMAX provides two filters for image pre-processing of the spots signal: blurring (gaussian filter) and a spot detection filter (i.e., Difference of Gaussians).

The gaussian filter can be applied with a different sigma for each dimension, which is very useful when working with anisotropic 3D z-stack data.

The spot detection filter is a Difference of Gaussians filter whose parameters are automatically calculated from the expected spot size. See the description of the parameter Sharpen spots signal prior detection for more details.

2.2. Semantic segmentation

The pre-processed image is used as input to the segmentation model of your choice. SpotMAX will segment the reference channel by using automatic thresholding, SpotMAX AI models, Spotiflow, or any of the models available on the BioImage.IO Model Zoo.

For more details about the available methods for automatic thresholding see this guide Thresholding (scikit-image).

One crucial aspect of SpotMAX is that you can apply the segmentation model on each input segmented object (e.g., the single cells, a.k.a. “Local”) or on all the objects in the image (a.k.a. “Aggregated”). See the parameter Aggregate cells prior analysis to know how to toggle these two modes.

We recommend testing with both modes, but as a rule of thumb if all the reference channel structures are present in all the cells but you have a large intensity variation between objects then using the “Local” mode could be beneficial.

On the other hand, if some of the objects are completely devoid of any reference channel structure aggregating the objects might be the only option.

Note

If you are working with the S. cerevisiase model organism, most of the times small buds do not have any structure. However, the “Local” will still work if you annotate mother-bud relationship using our other software Cell-ACDC. This is because SpotMAX will consider the mother-object as a single object while the bud is still attached to the mother (i.e., before division is annotated). Make sure that you provide the annotations to SpotMAX with the parameter Table with lineage info end name.