Computational methods for identifying cells with a stellate morphology: adaptive corner detection and a Fourier analysis. (A and B) Selected images of S2 cells displaying normal and stellate (Wave/Scar depleted) phenotypes. Bars, 12 µm. (C and D) An adaptive corner detection algorithm (He, X.C., and N.H.C. Yung. 2004. Proceedings of The Pattern Recognition 17th International Conference) applied to the same images as in A and B, where cell area was identified by thresholding. Detected corners are depicted by red boxes. (E) Quantification of the number of corners detected on the specific cells in C and D (blue bars) and means for Wave/Scar-depleted stellate cells and wild-type cells (red bars, n = 144 Wave/Scar depleted; n = 150 wild type). (F and G) A new method for analyzing the topology of the cell perimeter using power spectral analysis. The technique first traces the cell perimeter to create a matrix of all perimeter pixels and measures the distance from the cell centroid to each of these pixels. The insets show the distance from centroid values for the 140 perimeter pixels between the lines labeled D1 and D140. The Wave/Scar cell has been resized (not depicted) so that the perimeters of the two cells are both equal to ∼500. These values are Fourier transformed, and the complex modulus was calculated to yield the power spectrum for the perimeter of the cell. This power spectrum can then be binned into discrete frequency ranges and used to compare the membrane topology of cells by analyzing which frequencies are enriched in each phenotype. (H) Quantification of the integrated value of the power spectrum from frequencies 12–16 (Fig. S1) on the specific cells in F and G (blue bars) and means per plate well for Wave/Scar-depleted stellate cells and wild-type cells (red bars). Error bars indicate mean ± SD.