Nuclei Classification in Histopathological Images via Local Phase Quantization and Convolutional Neural Network

Classification of nuclei patterns in histopathological images has a significant interest in a wide range of biomedical research. In recent years, deep learning approaches have solved different issues and shown promising results in histopathological images. These developments have assisted researchers to distinguish different types of pathological nuclei samples without high labeling cost. In the normal routine, examination of histopathological slides under a microscope might be sensitive, slow and inaccurate due to subjective evaluations. In this study, we aim two different types of machine learning approaches for automatic nuclei classification in histopathological images on University of Warwick's colon data set with four nuclei types: Epithelial, inflammatory, fibroblast and other. The first approach was based on a convolutional neural network-based feature extraction with support vector machine-based classification. In the second approach, local phase quantization (LPQ) is applied along with support vector machine-based classification with data set augmentation. The obtained results achieved high accuracy rates on the classification of nuclei in histopathological images.