Improved Detail Enhancement Based on Mamba and Edge-Guided Polyp Segmentation Model

In early colorectal cancer screening, colonoscopic images are generally used for automated polyp segmentation and detection in order to improve diagnostic accuracy and efficiency. Automated segmentation of polyps remains challenging due to the generally complex environment inside the human gut and the quality limitations of the image output. To address this problem, a polyp segmentation model (DGAM-Net) is proposed based on the improved detail enhancement as well as edge guidance of Mamba. The model is improved in terms of both detail enhancement as well as edge segmentation effects. The extraction of global feature capability is enhanced in the encoder stage by introducing a state-space model; detail-enhanced convolution (DEConv) is introduced in the decoder to compensate for the lack of detailed features in the encoder; and an edge-guided attention (EGA) mechanism is introduced at the jump junctions, which enhances the ability to capture contextual information while helping the network to focus on the important regions. The proposed method is tested on the Kvasir and CVC-ClinicDB public polyp segmentation datasets. In these 2 authoritative datasets there is this certain improvement over the original model in its 6 evaluation metrics. It shows that the proposed algorithm has some effect on improving the original model with higher segmentation accuracy.