AGNet: An Alternate Guidance Network for Colorectal Polyp Segmentation and Detection

Colorectal cancer remains a significant health threat, but early detection through colonoscopy can lead to effective treatment. Precise polyp segmentation in colonoscopy images is crucial but difficult due to the variation in polyp characteristics, mucosal boundaries, imaging constraints, and computational complexity. This paper presents a novel and efficient Alternative Guidance Boundary-Aware Polyp Segmentation Network, which integrates Feature Selective Models, Multi-Scale Convolutions, and Alternate Guidance modules. The Feature Selective Model effectively selects highly discriminative features while reducing background noise in each side output. Comprehensive experimental results show that the proposed method surpasses six state-of-the-art techniques in polyp segmentation across five public datasets, evaluated using four widely recognized metrics. In particular, on the CVC-300 dataset, it exhibited the strongest performance, with an F-measure (F<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">β</inf>) of 0.98, S-measure (S<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">α</inf>) of 0.994, E-measure (Eδ) of 0.994 and Mean Absolute Error(M) of 0.002. These results underscore the effectiveness and robustness of the proposed approach in improving polyp detection, and its potential to contribute significantly to early colorectal cancer diagnosis and prevention.