Question

[Joseph Ronald Rubin]

Hello,

I intend to use rapidSTORM to obtain localizations of particles seen through TIRF and use those to calculate residence times of particles binding to the surface.  My images have an uneven background.  Due to the uneven background, the fixed global threshold does not work.  I attempted to use the local relative threshold but the tracker will only track a very small percentage of the particles.  I tried adjusting the spot search eagerness and it still does not track most of the particles.  I also adjusted the camera response to photon and the signal to noise ratio.  Do you have any advice on what other specific settings I should change to track more particles accurately?

Thank you so much,

Joseph Rubin

[Sven Proppert]

Hi Joseph,

I'm out of the game for quite a while now but I remember that we had a similar problem when imaging tissue. Steve suggested that we (and now you) could try the following:

Set "User level" (top menu) to "Expert"

Set "Spot finding method" to "Smooth by difference of averages"

Then two new input fields for "Foreground-" and "Background mask width" will appear. Play around with the values.

Unfortunately, I can't explain anymore what these values actually do. @Steve: Do you remember and would perhaps like to jump in?

I only remember, that we we got some reasonable results with values that seemed not to make too much sense. I think it was Foreground: 10 and Background: 12.

But as these settings behaved a little unexpectedly, it would not surprise me if totally different values gave better results with your data.

An additional thing you could try is to increase the "Spot search eagerness" (we took 15). But be aware that this significantly slows down the evaluation process. So you should have an up to date multicore machine at hand. By that I don't mean an Intel i5 1,6 Ghz dual core but also does not need to be a 10-20k$ workstation. By the way: GPU's will not help you in computing. rapidSTORM relies on CPU (and fast SSDs) as it was unfortunately written before the rise of GPUs :-/

I will try to reach out to my former collegues and try to get some info whether or not my memories are correct. If I don't post an erratum within the next weeks you can assume, there weren't any flaws. (or contact me to make sure, I didn't forget it...)

Keep us posted if you can make any progress.

Cheers,

Sven

[Steve Wolter]

Thanks for the great answer, Sven! You made my day, this kind of help is what I hoped for with the list :).

The problem is probably that rapidSTORM won't select the right positions to try fitting. By default, it applies a low-pass filter and then looks for the brightest points in the image. With strongly uneven background, it will try to fit the bright background first before it finds any spots. Sven suggested the right strategy - the difference of averages usually gives better results because it looks for the largest difference between the results of a weak averaging mask (should be around the spot size, so I'd try with 5) and a large averaging mask (should be comfortably larger than spots, Sven's 12 should be a good value).

The parameters you changed all influence the fitting algorithm and its results, that's why they don't make much of a difference when spot search is failing.

--
You received this message because you are subscribed to the Google Groups "rapidstorm-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rapidstorm-disc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rapidstorm-discuss/21690e46-8c26-4739-980d-49bfb5664bd1%40googlegroups.com.

[Sven Proppert]

Thank you, Steve! :-) *blushing*

And thank you for the explanation. Indeed, the sorted list of possible localisations was the link I was missing.

I called my former colleague Martin Pauli yesterday and in fact, Steve was right all the time and I remembered incorrectly. Martin said, they are using the standard settings for "difference of average" (5 and 15) and are getting reasonable results. Please don't ask, why I was quite sure some weird stuff happened and 10 and 12 were best. Maybe that was a freak measurement. Anyway: Go with the standard settings (5 and 15) and tweak them a little (only) if necesary.

A remark to the Spot search eagerness: Steve pointed out the mechanisms of this value in another thread. Please be aware that higher values can lead to more false positives. On the other hand, at least in my understanding, if the data is noisy with uneven background, a low value should lead to false negatives, as "the "spot search eagerness" is the amount of consecutive bad spots that the algorithm will look at". Correct me if I'm wrong, Steve.

My idea is, that if you can't sufficiantly get rid of the cell background, there will be some residual bad candidates (coming from bckground and not real spots) pretty much at the top of the sorted list of candidates. So with a low value, you have the risk to skip all the good ones - presumably - in the middle of the list (false negatives). On the other hand, if the value is too high the fitter might even take some of the rubbish at the bottom (false positives).

So either way, I think the "spot search eagernes" is a risky thing. I suppose the best way to find a good value is to artificially create ("simulate") some data and assess false positives, - negatives, precision, recall and the like. I think, I would go with a mockup measurement (the same kind of cell, but no particle marker in it) and than artificially add some spots (at least a few hundred) with a reasonable distribution (I would take random), size(, shape) and brightness (as in your real measurements).

Well, another question just arises: Why do you have background anyway, if you are in TIRF illumination. The penetration depth should be at most ~300 nm. Is there so much (bio-) fluorescence going on in your cells? Or is it more like a HILO illumination? How do you set your microscope to TIRF? And what kind of microscope is it? Self-assembled (you or somebody else) or commercial? Please don't get me wrong. I just want to make sure that your setup is not corrupted. It would not have been the first time to fight data algorithmically because something with the aquisition went wrong. Happened to me once or twice and it took very long to find the bugs ;-)

To unsubscribe from this group and stop receiving emails from it, send an email to rapidstorm-discuss+unsub...@googlegroups.com.