Cellular Architecture on Whole Slide Images Allows the Prediction of Survival in Lung Adenocarcinoma

Pathomic Features Reveal Immune and Molecular Evolution From Lung Preneoplasia to Invasive Adenocarcinoma

“…Besides, we embedded each ROI into smaller maps to summarize the cellular composition and architecture of each cell type. 41 , 42 Here, we took LYMs to illustrate the generation process of embedded maps. We glided through the ROI from the image top-left to bottom-right with a window size of 50 × 50 pixels and a stride of 50 pixels.…”

“…For each embedded map, we first calculated its gray-level co-occurrence matrix (GLCM). The contrast and energy of the GLCM matrix are then computed using the default scikit-image 42 parameters to serve as a compact representation of the embedded maps. Supplementary Table S3 summarizes the 9 pathomic features extracted and used in our analysis.…”

Automated Cellular-Level Dual Global Fusion of Whole-Slide Imaging for Lung Adenocarcinoma Prognosis

“…Since we are more focused on the global architecture and distribution of different cell populations than on the contours and textures of individual nuclei, our approach is to compress the local nuclear information to embed cell types in the WSI. First, we encode the abundances of the three cell types (neoplastic, inflammatory, and miscellaneous) for each 64 × 64 pixel window in three separate channels following Chen et al [7]. Then, an embedded map of different cellular types (Figure 2F) with a 64-fold dimensional reduction from the raw WSI is obtained.…”

“…Pathological assessments have been the gold standard for cancer diagnosis, and their role also extends to prognostication [6,7]. With the application of the whole-slide imaging scanner, digital pathologic images from tissue glass slides have markedly aided the development of computational pathology, including the prediction of the survival risk of patients [8].…”

“…In parallel, Alsubaie et al [23] characterized the heterogeneity of LUAD by using morphological features of segmented tumor cell nuclei, then generated a heat map of the entire WSI and used Cox proportional risk regression models to find the most important discriminatory features to implement prognostic analyses. Another study by Chen et al [7] was based on calculating partial eigenvalues for the cell nuclei to represent certain textural features, and ultimately selecting a subset of features for prognostic analysis. These models have some important limitations.…”