Figure 4. Artefact identification using random forest classifier.

Re-classification of the data allows artefacts to be identified and removed. A fixed cell sample, with microtubules labelled with Alexa 647, was used to acquire a localization microscopy data set. (a) All localized positions found by the ThunderSTORM, (b) only those classified as accurate and (c) those classified as inaccurate. Although the structure indicated by an arrow in a appears to consist of three strands, the middle one appears only in the image made from misfits due to overlapping fluorophores PSFs, revealing it to be an artefact. The image reconstructed from accurate fits; (b) only two strands. (d) A false colour image of the same data, with insets (i) and (ii) shown in more detail. Scale bar, 1.5 μm.