Figure 2

Conversion of linear to categorical scoring. The intensity histogram (panels inside A and B) obtained from a whole slide image was modeled using the Gaussian mixture. The Expectation-Maximization (EM) algorithm was applied to model three Gaussian components with the means and standard deviations for strong, moderate and weak intensity ranges. The white pixel intensity is plotted on the X axis with high values corresponding to less coloration. The probability density (Y axis) represents the frequency of pixels. A background cutoff was chosen arbitrarily at 230 white pixel intensity in DAB_HER2 and DAB_ER images. Pixel intensities indicated by thresholds t₁ and t₂ separate 3+ from 2+ or 2+ from 1+/0 respectively.