Classification of Cell Types on Histopathological Images Using Local Binary Patterns

Abstract

In recent years, the use of computer aided diagnostic (CAD) systems has been increasing with a high acceleration in the field of digital pathology. Application and study areas are expanding over time include the detection, classification and segmentation of nuclei. In this study, various traditional machine learning methods (k-closest neighborhood, random forests and support vector machines) and deep learning (convolutional neural network) were used comparatively on CRC colorectal adenocarcinomas dataset. Since conventional machine learning algorithms do not receive a two-dimensional input such as convolutional neural network, local binary images are utilized. As a result, when the feature extraction for machine learning algorithms is performed, KNN and RF algorithms provide very successful results, whereas CNN algorithm gave better results without making any feature extraction.