Boundary artifacts and sensitivity to small boundary perturbations. (A–C) The same cell image with the boundary (cyan) identified at three different threshold values. Bars, 5 µm. (A) Overthresholded case, in which thin filopodia (white arrows) are not properly resolved. Red and yellow lines in A–C indicate central lines (skeletons) of protrusions that are longer and shorter than the critical length L_cr, respectively. (B) Optimally thresholded case, in which all boundary features are properly captured. (C) Underthresholded case, in which filopodia start to merge (white arrows) and a small region of background noise (white rectangle) becomes a part of the misrepresented cell boundary. In all cases, FiloTrack identifies and tracks filopodia for a given boundary representation, but improperly captured cell features may reduce the accuracy of the final statistics. MovThresh is designed to find optimal thresholds and minimize possible boundary artifacts. (D) Two overlaid MATs (black and white) show small perturbation in cell boundaries can cause significant differences in the tree graph. Random noise (<1-pixel length) creates significant differences in the location of the graph edges and their number so that direct comparison of two MATs is problematic. (E) Rather than comparing MATs directly, we use extracted boundary profiles. This illustrates 100 overlaid boundaries that were perturbed within 1-pixel size. Red dots show protrusion tips on each boundary as determined by the corresponding boundary profiles (H). Boundary profiles are similar despite differences in tree graphs. (F and G) Expanded views of two regions highlighted in E. The spread of the detected protrusion tips (red) is on the same scale as the boundary perturbations, showing that small perturbations do not drastically change protrusion identification when using the boundary profiles. (H) 100 overlaid boundary profiles of perturbed cell outlines. Red dots show local maxima of the profiles. (I and J) Expanded views of two regions highlighted in H and corresponding to the protrusions shown in F and G. Hypersensitivity of the MAT to the boundary perturbation does not translate into hypersensitivity of protrusion tip detection using the boundary profiles, even though the latter are created using the MAT.