Question about rapidSTORM

[Steve Wolter]

Hi Judit,

welcome! I'm answering the question you asked in your group join request:

> Hello all! I'm using rapidSTORM as to find locations and their PSF and then run an in-house made software for SPT. In the input options, we're using fixed global threshold and calculating the threshold as the mean of the main grey value of a ROI over time plus two times the mean of the SD in each ROI over time. We also select smooth by average as spot finding method with a mask width=5. Although the background values we obtain make sense we're still finding many spots that are not real.In fact, we produced some simulated images and as soon as we add noise we have to introduce a factor of 5 to the threshold value to find the correct localisations. The question is: are we understanding correctly the threshold meaning in rapidSTORM? Furthermore, could the finding method and mask width have something to do with this? Thank you very much, Judit Sastre

The threshold in rapidSTORM is measured as the amplitude of the fitted Gaussian function, so the volume of the fitted PSF without the fitted background. The value of 2 SDs of the background sounds a bit low, though: As few as 2-3 outlying pixels can cause a high amplitude in fitting and surpass a threshold of 2 SDs easily. My diploma thesis had some data on this on pg. 69.

Overall, three factors will influence false positive spots:

1. before fitting, the smoothing and spot finding methods' goal is to pre-filter spot locations and output a sorted list of candidates, which hopefully has the true spots very close to the front. If this fails, later stages will have to trade recall against precision.

2. after fitting, the fit judger will decide based on the fit result whether a spot is a localization or not. This is where the threshold factors in.

2. the "spot search eagerness" is the amount of consecutive bad spots that the algorithm will look at. A high value will cause more false positive localizations because more spots are looked at.

Is this helping? Usually, a higher threshold was a good solution for the problem. I found a histogram of localization intensities to be quite useful for choosing a threshold because you usually see two clear regimes of good and bad localizations.

Best, Steve