TAG: Nucleus Detection in Colorectal Adenocarcinomas Histology Images using Local Texture, Appearance, and Gradient Features

This work considers the problem of detecting nuclei in H&E stained images. The high intra-class and inter-image variability of nuclei of the images itself calls for a more robust method that is able to handle this high variability. Recently, deep learning has become the most popular method in computer vision with many systems achieving state-of-the-art results for a wide array tasks. However, as a trade-off, deep learning methods require tons of data and computational resources that some may not posses. Thus, this work proposes a traditional computer vision pipeline along with TAG, a local feature-based approach that combines texture, appearance, and gradient features which is used in the task of pixel-per-pixel prediction in the virtue of semantic segmentation. Based from the results, the method was shown to be effective despite the challenges of high variability in the dataset.