Colorectal polyp region extraction using saliency detection network with neutrosophic enhancement

• Published in: Computers in biology and medicine 2022-08, Vol.147, p.105760-105760, Article 105760
• Main Authors: Hu, Keli, Zhao, Liping, Feng, Sheng, Zhang, Shengdong, Zhou, Qianwei, Gao, Xiaozhi, Guo, Yanhui
• Format: Article
• Language: English
• Subjects: Automation; Cancer; Colon; Colonoscopy; Colorectal carcinoma; Colorectal polyp; Decision making; Deep learning; Feature extraction; Image enhancement; Image segmentation; Medical prognosis; Polyp recognition; Polyp segmentation; Polyps; Salience; Saliency detection; Short connection; Specular reflection
• ISSN: 0010-4825; 1879-0534
• DOI: 10.1016/j.compbiomed.2022.105760

Colorectal polyp recognition is crucial for early colorectal cancer detection and treatment. Colonoscopy is always employed for colorectal polyp scanning. However, one out of four polyps may be ignored, due to the similarity of polyp and normal tissue. In this paper, we present a novel method called NeutSS-PLP for polyp region extraction in colonoscopy images using a short connected saliency detection network with neutrosophic enhancement. We first utilize the neutrosophic theory to enhance the quality of specular reflections detection in the colonoscopy images. We develop the local and global threshold criteria in the single-valued neutrosophic set (SVNS) domain and define the corresponding T (Truth), I (Indeterminacy), and F (Falsity) functions for each criterion. The well-built neutrosophic images are processed and employed for specular reflection detection and suppressing. Next, we introduce two-level short connections into the saliency detection network, aiming to take advantage of the multi-level and multi-scale features extracted from different stages of the network. Experimental results conducted on two public colorectal polyp datasets achieve 0.877 and 0.9135 mIoU for polyp extraction respectively, and our method performs better compared with several state-of-the-art saliency networks and semantic segmentation networks, which demonstrate the effectiveness of applying the saliency detection mechanism for colorectal polyp region extraction. •We present a polyp region extraction strategy using the scheme of saliency detection and the theory of neutrosophic.•We transfer the specular detection problem to the SVNS-based multi-criteria decision-making problem.•We apply the CNN-based saliency detection network to deal with the colorectal polyp region extraction problem.•We propose a saliency detection-based polyp extraction network with two-level short connections.