Has anyone used Mesmer in DeepCell for simple grayscale segmentation of cultured cell?

I was looking for some help from someone familiar with Mesmer (Van Valen group).

We have some fluorescent movies of cultures cells stimulated to transiently increase cytoplasmic calcium (calcium reporter lights up). Each movie is around 121 or so frames, only gray scale and there are a few dozen cells/frame. There is relatively little cell movement during the course of the movie (CHO cells are well-attached during the 3-30 min movies). I would like to segment the cells based on the fluorescence images (all cells have some internal cytosolic fluorescence). As I said, a rather simple problem compared to many. I have been trying to use Cell Profiler, but it is difficult to get most of the cells identified for the mask (sample image attached). I could do a lot of hand adjustments, but then I saw David Van Valen give a wonderful presentation of his AI segmentation program MESMER and wanted to try to use it. He has set up a detailed web site, but I thought it might be a bit faster for a novice such as myself to get some direct help from someone who has experience using MESMER and would be willing to give me a hand (tell me what exactly to do using the pre-trained models in the cloud). The analyses we want to perform are straightforward (change of cell fluorescence intensity (calcium level) with time. We just need to generate the segmentation (masks for each cell). The data are from a delta vision microscope in “.dv” files. I was hoping there was a very simple way for me to try to use MESMER for a quick initial try to analyze our data, just in case it really isn’t the way to go for us. I have attached a couple of framesS16-chol_FirstFrame.tiff (8.0 MB) to give you a sense of the data.


Addendum: can MESMER be used if there is no nuclear (DAPI) staining, just monochrome (calcium reporter in the cytosol)?

Never Tried Mesmer, but the results in CellPose don’t seem too bad.

1 Like

Thanks for the suggestion. Turns out MESMER is NOT appropriate for these data without nuclear staining. Thanks for the suggestion.