A Novel Neural Network Based on Transformer for Polyp Image Segmentation

The segmentation of medical images is crucial in clinical medicine for diagnosis and treatment. Precise segmentation of polyp images is of significant importance in colonoscopy to decrease the missed detection rate and lower the risk of polyps progressing into colon cancer. Although Convolutional Neural Networks have provided an automated solution for segmentation and feature extraction, the disadvantages of this approach are lengthy training times and loss of local details. Conversely, Transformer excel in capturing global features and parallel processing, but require some artifacts. This paper introduces a novel image processing architecture that combines CNN and Transformers in a unique manner. The proposed model is evaluated on five polyp segmentation datasets, and the experimental results demonstrate that it achieves faster reasoning speed and higher accuracy compared to conventional CNN.