Colorectal polyp segmentation using a fully convolutional neural network

Abstract

Colorectal cancer is the third common cancer in the United States and most colorectal cancer is associated with colorectal polyps. In hospital, colonoscopy is a common way to detect colorectal polyps. Colorectal polyps segmentation plays an important role in the diagnosis and prevention of digestive system related diseases. Therefore, there is a pressure-need for polyp segmentation computer-aided system to help doctors in diagnosis. In this paper, we propose a new, end-to-end fully convolutional neural network structure for segmenting colorectal polyps. This method can directly output a prediction map of the same size as the original image of the input network. We use the CVC-ClinicDB database to evaluate our method. Proposed method achieves accuracy values of 96.98%, F1score values of 83.01%, sensitivity values of 77.32% and specificity values of 99.05%.