Detection of Lung and Colon Cancer from Histopathological Images: Using Convolutional Networks and Transfer Learning

Abstract

Analyzing histopathology images to detect the presence of cancer cells is a very important task during cancer treatment. This task has traditionally been largely done by manual methods. Therefore, the results of these analyzes are highly dependent on the pathologist's skills and professional experience, wasting time and manpower. Automating this task using deep learning techniques will speed up the early detection of cancer cells. Interestingly, these techniques have led to impressive advances in image processing in various fields, including the medical field. In this paper, we first attempt to highlight the importance of using deep learning techniques to classify histopathological images, and have cited studies using LC25000 datasets to accomplish this task. We then compared twelve models based on pretrained VGG-16, ResNet, DenseNet and NasNet models. The overall accuracy in this study ranged from 95.99% to 99.98%, reaching 100% for some categories. The purpose of this article is to compare pretrained models, examine the impact of the number of layers on the performance of built models, and highlight the importance of using transfer learning techniques.