Correlating Changes in the Epithelial Gland Tissue With Advancing Colorectal Cancer Histologic Grade, Using IHC Stained for AIB1 Expression Biopsy Material

Abstract

A systematic statistical analysis of AIB1-stained biopsy material showed that high-grade CRC lesions contain higher intensity levels, appear coarser, are more homogeneous with smooth variation across the image, have lower contrast that is slowly varying across the image, have lower AIB1 staining, and have lower edges. A combination of textural and color attributes, evaluating image gray-tone distribution, textural roughness, inhomogeneity, AIB1 staining, and image coarseness should be considered in evaluating AIB1-stained CRC lesions.

Employing machine learning and microscopy images of AIB1‐stained biopsy material to assess the 5‐year survival of patients with colorectal cancer

“…We generated a total of 69 features, 53 features quantified textural properties that were computed from the grayscale version of the image ROI and 16 features evaluated color properties from the original RGB image. All 69 features have been explicitly defined in a previous study by our group that concerns epithelial gland changes with CRC histologic grade (Theodosi et al, 2018). Regarding textural features (S. Theodoridis & Koutroumbas, 2003), 4 features were first‐order statistics features computed from the image‐ROI's grayscale histogram, 14 features were computed from the image‐ROI's co‐occurrence matrices, 5 features were calculated from the image‐ROI's run‐length matrices, 24 features were quantified from the two‐dimensional, discrete wavelet transform using Daubechies wavelets (S. Theodoridis & Koutroumbas, 2003), and 6 features were calculated in accordance with the Tamura method (Tamura, Mori, & Yamawaki, 1978).…”

“…Regarding textural features (S. Theodoridis & Koutroumbas, 2003), 4 features were first‐order statistics features computed from the image‐ROI's grayscale histogram, 14 features were computed from the image‐ROI's co‐occurrence matrices, 5 features were calculated from the image‐ROI's run‐length matrices, 24 features were quantified from the two‐dimensional, discrete wavelet transform using Daubechies wavelets (S. Theodoridis & Koutroumbas, 2003), and 6 features were calculated in accordance with the Tamura method (Tamura, Mori, & Yamawaki, 1978). Regarding color features, we calculated 16 features from the L*a*b* version of the RGB color image (Theodosi et al, 2018).…”

“…However, the number of features in this scenario was very large (3 × 69 = 207). This fact renders optimal SML‐system design nearly impossible if one considers that we have to choose a maximum of nine‐feature combination from within a total of 207 features for avoiding over‐fitting (Theodosi et al, 2018). Consequently, we applied feature reduction by first ranking our data employing a criterion (S. Theodoridis & Koutroumbas, 2003) formed by each feature's between‐class discrimination, using the Wilcoxon statistical test, and the feature's correlation with the rest of the features.…”