The ScanScope CS microscope slide scanner (Leica Biosystems) was used to digitize whole slide images of histological sections at 20 × (0.5 µm/pixel) and record them as tiled tiff images.
For each image, regions of interest (ROIs) were drawn using the ImageScope software (Leica Biosystems) in order to select only tumor tissues and remove the artifacts. The images were processed as reported in the previous study (25 (link)). Briefly, squares of 2000 pixels size corresponding to 1 mm2 area were used in this study. The squares were generated to fit the area of the ROI. A ratio between the stained area (brown color) and the surface of tissue was computed and assigned to each square based on their coordinates. Local ratio computed for each square was ranked according to the following ten intervals: level 0 (0–10%), level 1 (>10–20%), level 2 (>20–30%), level 3 (>30–40%), level 4 (>40–50%), level 5 (>50–60%), level 6 (>60–70%), level 7 (>70–80%), level 8 (>80–90%), and level 9 (>90–100%). The ranks then formed the basis for the co-occurrence matrix used to compute Haralick texture parameters. The classical Haralick parameters (26 (link)) were computed from the normalized co-occurrence matrix: contrast, homogeneity, dissimilarity, entropy, energy, and correlation. The descriptors of the distribution shape were also computed: skewness and kurtosis.
For each image, regions of interest (ROIs) were drawn using the ImageScope software (Leica Biosystems) in order to select only tumor tissues and remove the artifacts. The images were processed as reported in the previous study (25 (link)). Briefly, squares of 2000 pixels size corresponding to 1 mm2 area were used in this study. The squares were generated to fit the area of the ROI. A ratio between the stained area (brown color) and the surface of tissue was computed and assigned to each square based on their coordinates. Local ratio computed for each square was ranked according to the following ten intervals: level 0 (0–10%), level 1 (>10–20%), level 2 (>20–30%), level 3 (>30–40%), level 4 (>40–50%), level 5 (>50–60%), level 6 (>60–70%), level 7 (>70–80%), level 8 (>80–90%), and level 9 (>90–100%). The ranks then formed the basis for the co-occurrence matrix used to compute Haralick texture parameters. The classical Haralick parameters (26 (link)) were computed from the normalized co-occurrence matrix: contrast, homogeneity, dissimilarity, entropy, energy, and correlation. The descriptors of the distribution shape were also computed: skewness and kurtosis.