A novel polyp segmentation approach using U-net with saliency-like feature fusion

Abstract

Polyps that pose a serious threat to human life, some of which are not detectable during a colonoscopy. This situation not only reduces the quality of life but also significantly raises healthcare costs. Polyp detection early reduces the risk of colorectal cancer and increases the likelihood of successful treatment. However, the complex environment of colonoscopic polyps, as well as various disturbing factors, make this process extremely difficult. Although advancements in computer-aided diagnosis systems have increased success in polyp detection tasks, there are still critical gaps today. In this study, a U-Net supported visual saliency approach is proposed to successfully determine polyp boundaries. Color space information, gradient information, and saliency are used as U-net input for this purpose. When compared to the original polyp images, the proposed end-to-end segmentation method has very high segmentation performance. Furthermore, its dice performance outperforms current state-of-the-art methods in the literature. © 2023 Elsevier Inc. All rights reserved.