Tissue Region Growing for Hispathology Image Segmentation

Abstract

The accurate identification of the tumour tissue border is of crucial importance for histopathology image analysis. However, due to the high morphology variance in histology images, especially in border regions where cancer tissue interfere into the normal region, it is challenging even for the pathologists to define the border, not to say for the machine. In this paper, we present an innovative framework to semantically segment the tumour border area in colorectal liver metastasis (CRLM) on pixel level by integrating the features from deep convolutional networks with spatial and statistical information of the cells. With annotations from the pathologists, a two-level deep neural network including a cell-level model and a tissue-level model, is trained to classify patches from the whole slide scan image. Based on the prediction of trained models, a growing-style algorithm is proposed to finalize the segmentation by leveraging the statistical and spatial properties of the cells. Evaluated against the ground truth created by the experts, the framework demonstrates a significant improvement over a conventional deep network model on the cell-level model or the tissue model alone.