抄録
Currently, there is a problem with the lack of doctors who diagnose colorectal cancer. Especially in Japan, colorectal cancer patients are increasing with the aging society. Therefore, the construction of automatic diagnosis support system that substitutes double and triple check in colorectal cancer cytology will be indispensable. In this study, we constructed a system to automatically identify colorectal cancer, adenoma and normal cells by changing the layer structure based on Convolutional Neural Network and Fully Convolutional Network. As a result, the maximum value of classification accuracy is 95.22% by VGG16 based Convolutional Neural Network and 96.52% for U-Net based Fully Convolutional Network. The result shows that this system opens up the possibility of the system which substitutes double check and triple check of colorectal cancer cytology.
