Mean depth-width ratio of extrema as textural feature for automated cell proliferation analysis

As a step towards automation of Mitotic Index estimation for cell proliferation studies, we introduce in this work a roughness feature of surface-intensity images: the mean depth-width ratio of extreme, (MDWRE). This feature allowed identification of variable-shaped metaphases and interphase nuclei in the presence of many artifacts (one metaphase per hundreds of nuclei and thousands of artifacts). The texture of the cytological objects (seen as rough surfaces) was quantified by scanning in one dimension the lines contained in a closed contour. MDWRE resulted suitable for image magnifications as low as (x10), making possible a faster scanning of the slides. The use of this feature gave + 14%, +65%, +133% and +133% better performance figures than classical textural features derived from cooccurrence matrices such as Contrast, Energy, Entropy and Angular 2nd Moment respectively, and +51% better than the Relative Extrema Density (RED). The MDWRE per object and the shape of the histogram of the depth-width ratio of gray-level roughs, have shown to be very useful as textural features for the classification of metaphase images.