Analysis of colon cancer disease images using deep learning approach

Colon cancer ranks third in deaths due to cancer worldwide. Early diagnosis of colon cancer is of great importance in effective controlling of the disease. In recent years, significant advances on Deep Learning (DL) technology have been carried out, and this technology might be frequently used in the fields of detection, diagnosis, classification, segmentation and prediction on types of cancer. In this study, histopathological colon tissue image samples were classified as cancerous or normal by using Convolutional Neural Networks (CNN), which is a neural network frequently preferred in the field of image processing among DL approaches. The data sets used in this study were obtained from the open sources and were randomly divided in order to train and test in DL models. The prepared image data were classified as cancerous or normal using current DL algorithms such as AlexNet, GoogLeNet, ResNet50, ResNet101 and Xception. The parameters of accuracy, sensitivity, specificity and F1 score were used in measuring of the performance of DL algorithms. As a result, accuracy rates of AlexNet, GoogLeNet, ResNet50, ResNet101, and Xception were 99.73%, 99.9%, 99.93%, 99.87% and 99.93%, respectively. These results indicate that the models used in this work may be useful in colon cancer diagnosis and classification.