Context-Aware CNN

Table of Contents
Introduction
Demo
Citation
Dataset
Code
Method
Results

Context-Aware Convolutional Neural Network for Grading of Colorectal Cancer Histology Images

View the Publication and the Supplementary Document.

Introduction

Digital histology images are amenable to the application of convolutional neural network (CNN) for analysis due to the sheer size of pixel data present in them. CNNs are generally used for representation learning from small image patches (e.g. 224x224) extracted from digital histology images due to computational and memory constraints.

We propose a novel framework to incorporate larger context by context-aware neural networks based on images with a dimension of 1,792x1,792 pixels for classification and segmentation tasks. The proposed framework first encodes the local representation of a histology image into high dimensional features then aggregates the features by considering their spatial organization to make a final prediction.

Demo

A colab based demo on a set of images is available here

Citation

The journal paper on this work has been publised in IEEE Transactions on Medical Imaging. If you want to refer this work in your research, please cite our journal paper on this work:
@article{Context-aware convolutional neural network for grading of colorectal cancer histology images},
  author={Shaban, Muhammad and Awan, Ruqayya and Fraz, Muhammad Moazam and Azam, Ayesha and Tsang, Yee-Wah and Snead, David and Rajpoot, Nasir M},
  journal={IEEE Transactions on Medical Imaging},
  year={2020}
  publisher={IEEE}
}

Dataset

The CRC and Exteded CRC datasets are publicly available on TIA-Lab dataset page

Code

The code will be released by Mid of July 2020.

Method

Flow diagram of the proposed context-aware framework for CRC grading. The top row shows the local representation learning. The bottom row illustrates the network architecture for representation aggregation learning which consists of multiple context blocks and other standard layers. Dashed lines represent the blocks of a specific network design whereas solid lines represent the common blocks.

Results

Visual results of CRC grading are shown for patch classifier, existing context, and the proposed method on an image of size 1,792x1,792. The stride size for context networks is equal to the size of patch (224x224) used for patch classifier. Green, blue and red colors of overlaid rectangular boxes show the normal, low and high-grade predictions respectively, whereas empty box areas represent non-glandular/background regions.