Transfer Learning Methods for Using Textural Features in Histopathological Image Classification

Abstract

As the technological advances in computer hardware and machine learning have increased significantly, deep learning models have also been used in many different areas. Examples of these areas are image recognition, face detection, natural language processing, toxicology, suggestion systems, anomaly detection and disease diagnosis in the health sector. This study focuses on studies on disease prediction and diagnosis through histopathological images. The main purpose of the study is to apply deep learning models that can classify cancerous tissues with high accuracy. Besides that, implementation of deep models are done with a low computational cost so that models can be trained in a fast manner. Within the scope of this subject, the convolutional neural network models, which are very popular in image classification in the deep learning world, have been realized by applying transfer learning technique. In addition to these models, a deep learning model called CAT-Net is used to compare and evaluate the success of the transfer learning method. The results of the study are compared with overall accuracy, precision, recall, and F1 score metrics for each model.