robustness of automated cell labelling about suite2p HOT 3 CLOSED

Hi Marius,
I've just started using Suite2p and I'm having some issues with Cell Detection.
I'm processing a FOV acquired at 30hz, obj 16x, digital zoom 2, 512x512.
Please find attached a picture showing the z-stack of my field and next to it the GUI. As you can see, the algorithm finds very few cells (which may actually not even be neurons) plus a several other small clusters of pixels. This is very far from the result I should obtained. Would you be able to explain why am I having this issue? Is there anything I can do to improve it? As things are now, I'm not even able to select the real ROIs on the right side of the GUI and move them to the left side.
Some info:
in make_db I set db(i).diameter = 12; %this may need to be smaller? it is in pixels, right?
as suggested in the readme, I set ops0.nSVDforROI to anumber close to the expected number of cells in my FOV. In this case 50. However, this doesn't seem to make cell detection easier.
Is there anything else I can do?

many thanks,
mariangela

from suite2p.

Hi Mariangela, I don't think this has anything to do with the previous issue in the thread.

In any case, it looks like something went horribly wrong. Perhaps the registration failed, or perhaps the dimensions of the tiffs have not been detected correctly. Are there extra channels in your tiff? Single plane? Try to plot ops.DS (that shows the offsets computed by the registration). These offsets should be relatively smooth in time. If they are all over the place, something is wrong. Also look at the target image for registration (ops.mimg). That should look like a crisp mean image.

Try also to set ops0.nSVDforROI at a larger value, maybe around 300-500. You won't really see benefits below that value, and I'm not sure if a small number confuses it.

from suite2p.

Hi Marius,
many thanks fro your reply. I think the issue came up because I used a tiff file that had been previously processed in imageJ. The original file contained acquisitions from 2 channels and two different planes. I then cut this to obtain a final file with only green channel and one plane. This is the one I used in suite2p.

from suite2p.