Segmentation of polyps based on pyramid vision transformers and residual block for real-time endoscopy imaging.

Autor: Nachmani R; Department of Electrical and Electronics Engineering, Ariel University, Ariel 407000, Israel., Nidal I; Department of Surgery, Hasharon Hospital, Rabin Medical Center, affiliated with Tel Aviv, University School of Medicine, Petah Tikva, Israel., Robinson D; Department of Orthopedics, Hasharon Hospital, Rabin Medical Center, affiliated with Tel Aviv, University School of Medicine, Petah Tikva, Israel., Yassin M; Department of Orthopedics, Hasharon Hospital, Rabin Medical Center, affiliated with Tel Aviv, University School of Medicine, Petah Tikva, Israel., Abookasis D; Department of Electrical and Electronics Engineering, Ariel University, Ariel 407000, Israel.; Ariel Photonics Center, Ariel University, Ariel 407000, Israel.
Jazyk: angličtina
Zdroj: Journal of pathology informatics [J Pathol Inform] 2023 Jan 26; Vol. 14, pp. 100197. Date of Electronic Publication: 2023 Jan 26 (Print Publication: 2023).
DOI: 10.1016/j.jpi.2023.100197
Abstrakt: Polyp segmentation is an important task in early identification of colon polyps for prevention of colorectal cancer. Numerous methods of machine learning have been utilized in an attempt to solve this task with varying levels of success. A successful polyp segmentation method which is both accurate and fast could make a huge impact on colonoscopy exams, aiding in real-time detection, as well as enabling faster and cheaper offline analysis. Thus, recent studies have worked to produce networks that are more accurate and faster than the previous generation of networks (e.g., NanoNet). Here, we propose ResPVT architecture for polyp segmentation. This platform uses transformers as a backbone and far surpasses all previous networks not only in accuracy but also with a much higher frame rate which may drastically reduce costs in both real time and offline analysis and enable the widespread application of this technology.
Competing Interests: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(© 2023 The Authors.)
Databáze: MEDLINE
Externí odkaz: