Cells are Actors: Social Network Analysis with Classical ML for SOTA Histology Image Classification

Digitization of histology images and the advent of new computational methods,\nlike deep learning, have helped the automatic grading of colorectal\nadenocarcinoma cancer (CRA). Present automated CRA grading methods, however,\nusually use tiny image patches and thus fail to integrate the entire tissue\nmicro-architecture for grading purposes. To tackle these challenges, we propose\nto use a statistical network analysis method to describe the complex structure\nof the tissue micro-environment by modelling nuclei and their connections as a\nnetwork. We show that by analyzing only the interactions between the cells in a\nnetwork, we can extract highly discriminative statistical features for CRA\ngrading. Unlike other deep learning or convolutional graph-based approaches,\nour method is highly scalable (can be used for cell networks consist of\nmillions of nodes), completely explainable, and computationally inexpensive. We\ncreate cell networks on a broad CRC histology image dataset, experiment with\nour method, and report state-of-the-art performance for the prediction of\nthree-class CRA grading.\n