Classification of Colorectal Cancer Histology Images Using Image Reconstruction and Modified DenseNet

Colorectal cancer is one major cause of cancer-related death around the globe. Recent breakthroughs in deep learning have paved the way to apply it for the automation of histopathology images as a tool for computer-aided diagnosis of medical imaging. Here we have presented a novel state of the art classification model for classifying the colorectal histopathology images into 9 classes. All the traditional approaches like texture-based classification, transfer learning etc. already has been used to achieve a state-of-the-art result, but these have some limitations. Rather than using conventional mechanisms, we have proposed a methodology that can interpret the histopathology images in a more generalized way without image preprocessing and augmentation mechanisms. A combination of two deep learning architectures i.e., an encoder unit of autoencoder module and a modified DenseNet121 architecture are used for this purpose. An accuracy of 97.2% on Zenodo 100k colorectal histopathology dataset has been reported. The presented result is better than most of the contemporary works in this domain. We have also evaluated the effectiveness of the current approach for the low-resolution histopathological images and achieved good recognition accuracy.