Polyp segmentation in medical imaging: challenges, approaches and future directions

Abstract Colorectal cancer has been considered as the third most dangerous disease among the most common cancer types. The early diagnosis of the polyps weakens the spread of colorectal cancer and is significant for more productive treatment. The segmentation of polyps from the colonoscopy images is very critical and significant to identify colorectal cancer. In this comprehensive study, we meticulously scrutinize research papers focused on the automated segmentation of polyps in clinical settings using colonoscopy images proposed in the past five years. Our analysis delves into various dimensions, including input data (datasets and preprocessing methods), model design (encompassing CNNs, transformers, and hybrid approaches), loss functions, and evaluation metrics. By adopting a systematic perspective, we examine how different methodological choices have shaped current trends and identify critical limitations that need to be addressed. To facilitate meaningful comparisons, we provide a detailed summary table of all examined works. Moreover, we offer in-depth future recommendations for polyp segmentation based on the insights gained from this survey study. We believe that our study will serve as a great resource for future researchers in the subject of polyp segmentation offering vital support in the development of novel methodologies.