Deep learning-based colon cancer diagnosis

Colon cancer is a type of cancer that occurs in colon. In 2020, nearly 2 million cases of colon cancer were diagnosed worldwide. In Turkey, the incidence of colon cancer is fourth in men and second in women. In the last decade, significant advances have been achieved in the field of deep learning. Human image classification performance has been surpassed via deep learning methods. The image classification performances of convolutional neural network (CNN)-based models have been surpassed by the vision transformer (ViT) model proposed in 2021. Numerous deep learning-based methods have been developed to date for automatic classification, analysis and interpretation of medical images. The successful use of deep learning in medical imaging methods has brought deep learning research in colon cancer to the agenda. In this study, the automatic colon diagnosis performance of ViT and CNN-based transfer learning models were investigated. The effects of different patch sizes on the performance of ViT models have been demonstrated. During the model training and testing, the Kather-5k dataset containing 5000 histopathology images was used. As a result of the performance comparison between ViT and CNN-based transfer learning methods, the highest accuracy was obtained as 94,93% by the EfficientNetB4 transfer learning model. Using the CNN-based transfer learning method, the workload of pathologist doctors may be reduced and the performance of colon cancer diagnosis may be improved.