Fig. 2
Benchmarking DHPSFU performance against established algorithms. (A) The Jaccard index (the number of true localisations divided by the total number of true plus false localisations detected) for each method. (B) The dependence of each method’s sensitivity on the z-coordinate of the localisation. The two images illustrate the near-horizontal orientation of the DH-PSF close to the extremes of the z-range in the ‘Microscope 2’ dataset. The dashed lines indicate the z-range to which the plots in C are restricted. (C) The performance of the algorithms in the “optimal” z-range. Left, a plot of sensitivity (the number of true localisations detected divided by the total number of true localisations) against precision (the number of detected true localisations divided by the total number of detected localisations) for the three processing methods. Right, the Jaccard index for each method. In A and C, the lighter points and bars represent the values taken from the Sage et al. dataset20, where the Sage et al. datasets were analysed by the method developers themselves. Bottom left, description and sample PSF close to z = 0 for the three microscopes used in this study. (D) The localisation accuracy in XY (left) and Z (right) for each microscope and algorithm. (E) Close-up views of regions from the Sage et al. datasets showing: (1) the splitting of a single localisations into two, merging of two localisations into one, or missed localisations by SMAP (blue arrows); and (2) the localisations missed by EasyDHPSF (orange arrows). Scale bars, 1 μm.
