Fig. 6

Results for real images. Examples of hiPSC-CMs from validation set 1 (a) and validation set 2 (b), imaged using fluorescence microscopy, are shown. An example of hiPSC-PSNs, from validation set 3, imaged using phase contrast imaging is shown in (c). The distributions of absolute differences in degrees between mean orientations estimated by CytoSpectre and by each human expert are plotted in (d), (e) and (f) for images of validation set 1 (N = 11), validation set 2 (N = 10) and validation set 3 (N = 10), respectively. If ¼ or more of the human experts were unable to specify the mean orientation for an image, the image was excluded from this analysis. The red line is the median, the edges of the blue box are the 25th and 75th percentiles and the whiskers extend to 1.5 times the interquartile range (corresponding to approximately 99.3 % coverage for normally distributed data). Values beyond these limits are considered outliers and are plotted as red crosses. Pearson’s linear correlation coefficients between anisotropy rankings estimated by human experts and circular variance values estimated by CytoSpectre (CS) for each human-human or CS-human pair are shown as correlation matrices in (g), (h) and (i) for images of validation set 1 (N = 15), validation set 2 (N = 15) and validation set 3 (N = 15), respectively