Fig. 2.

Overall framework for extracting features from the collagen and tumor-infiltrating lymphocyte (TIL) components within the tumor microenvironment (TME) of H&E images. a H&E images from both pre- and post-NACT stages of the same patient were acquired. The pre-NACT H&E image was derived from a biopsy, whereas the post-NACT H&E image originated from a surgical specimen. b Tiles of size 3000x3000-pixel were extracted from each pre- and post-NACT H&E image. These tiles underwent preprocessing steps for feature extraction, including epithelium/stroma segmentation and nuclei segmentation. c The first set of features focused on quantitative aspects of Collagen Fiber Orientation Disorder (CFOD) in stromal regions. Collagen fiber orientations were determined using a derivative-of-Gaussian model. An orientation co-occurrence matrix was generated, where brighter on-diagonal cells indicated higher co-occurrence of collagen fibers with similar orientations. A feature quantifying the disorder of collagen fiber orientations was computed from this matrix. Warmer colors in the feature map represented greater disorder in collagen fiber orientations. d Another set of features characterized TIL architecture quantitatively in epithelial and stromal regions. HGSOC, high grade serous ovarian carcinoma; TIL, tumor-infiltrating lymphocytes; CFOD, collagen fiber orientation disorder.