Intestinal polyp segmentation based on U-net model

About 90% of colorectal cancer is caused by the malignant transformation of adenomatous polyps. Early resection can increase the five-year survival rate to over 90%. In view of the high missed detection rate (25%) of traditional colonoscopy and the difficulty in segmenting tiny polyps (Dice coefficient less than 65%), this study introduced multi-scale feature fusion and attention mechanism based on U-Net architecture to achieve automatic segmentation of polyps. The Dice coefficient reached 77.016%, and the measurement accuracy met clinical needs, providing effective support for auxiliary diagnosis.