Fig. 2: HAWKMAN analysis of simulated microtubule data from the Localisation Microscopy Challenge.

Comparison of SE and ME fitting of high emitter density data and equivalent low density data. A section of the reconstructions is shown in (a–c) for HD:SE fitting, HD:ME fitting and LD:SE fitting, respectively, which contain artificial sharpening for both high density methods. The results of HAWKMAN analysis are shown in (d–l) for each of the three test reconstructions with HAWK used as the reference. The sharpening maps (d–f) and structure maps (g–i) are shown at a length scale of 20 and 80 nm, respectively. In both the SE and ME cases the magenta areas (test image only) in the sharpening maps (d, e) and the structure maps (g, h) indicate substantial sharpening along with missing structure (cyan), which is more severe (as expected) in the SE case. Both these effects are absent for the low density reconstruction (f, i). These results are reflected in the confidence maps (j–l, 30 nm scale). These show substantially reduced confidence (strong artefacts, highlighted in red) for both high density methods (j, k), whereas for the low density data the confidence (l) is high everywhere. The differences in accuracy of the three reconstructions is further highlighted by the corresponding artefact scale maps (m–o). These show the maximum scale of artefacts is larger for the SE reconstruction than the ME and that the LD is virtually artefact free. The colour scale ranges from length scales 1 reconstruction pixel (10 nm red) to 27 (270 nm purple). Scale bar 500 nm.