CrossFormer: Multi‐scale cross‐attention for polyp segmentation

Colonoscopy is a common method for the early detection of colorectal cancer (CRC). The segmentation colonoscopy imagery valuable examining lesion. However, as colonic polyps have various sizes and shapes, their morphological characteristics are similar to those mucosa, it difficult segment them accurately. To address this, novel neural network architecture called CrossFormer proposed. combines cross-attention multi-scale methods, which can achieve high-precision automatic polyps. A module proposed enhance ability extract context information learn different features. In addition, channel enhancement used focus on useful information. model trained tested Kvasir CVC-ClinicDB datasets. Experimental results show that outperforms most existing methods.