Enhancing Histopathological Colorectal Cancer Image Classification by using Convolutional Neural Network

Colorectal cancer (CRC) also known as bowl cancer is one of the leading death causes worldwide. Early diagnosis has become vital for a successful treatment. Now days with the new advancements in Convolutional Neural networks (CNNs) its possible to classify different images of CRC into different classes. Today It is crucial for physician to take advantage of the new advancements in deep learning, since classification methods are becoming more and more accurate and efficient. In this study, we introduce a method to improve the classification accuracy from previous studies that used the National Center for Tumor diseases (NCT) data sets. We adapt the ResNet-50 model in our experiment to classify the CRC histopathological images. Furthermore, we utilize transfer learning and fine-tunning techniques to improve the accuracy. Our Experiment results show that ResNet_50 network is the best CNN architecture so far for classifying CRC histopathological images on the NCT Biobank open source dataset. In addition to that using transfer learning allow us to obtain 97.7% accuracy on the validation dataset, which is better than all previous results we found in literature.