Improving the Polyp Image Segmentation Based on Parallel Reverse Attention Network

Polyp image segmentation helps diagnose and treat tumors in a patient’s body. The big challenge of it is the similarity in color, texture and intensity compared to neighboring points. This paper method proposed to segment polyp images by improving Parallel Reverse Attention Network methods. The proposed method includes three stages as: (i) partial decoder to identify the global map by Res2Net; (ii) downsampling for reverse attention from the global map; (iii) upsampling with adding ground-truth for sigmoid function for polyp segmentation. The experiments are performed on public datasets such as Kvasir-SEG and EndoTech with IoU and Dice Coefficient evaluation metric. The results of the proposed method better the other methods in these datasets.