Fuzzy Energy Competition Active Contour Network for Polyp Segmentation

Accurate polyp segmentation in colonoscopy images is important for early detection of colorectal cancer, but speckle noise, variable lesion size, low contrast, and illumination changes can obscure boundaries. Learning-based models provide strong semantic representations. However, their masks may remain spatially smooth or uncertain at weak boundaries, whereas conventional active-contour models lack task-specific deep features and often require iterative optimization. We propose the Fuzzy Energy Competition Active Contour Network (FECAC-Net), an end-to-end framework that combines a Progressive Multi-Scale Attention (PMSA) network with a Fuzzy Energy Competition Curve Evolution (FE2CE) network. PMSA uses a four-stage PVTv2 encoder and a lighter three-stage decoder with dense cross-scale connections to localize polyps and preserve fine details. FE2CE converts the PMSA mask into a pseudo level-set function and applies a strictly convex fuzzy-energy difference rule to refine membership values. On Kvasir-SEG, CVC-ClinicDB, CVC-300, CVC-ColonDB, and ETIS, FECAC-Net obtains mean dice scores of 0.920, 0.941, 0.893, 0.801, and 0.773, respectively. Its rapid convergence underscores its potential for clinical deployment in polyp segmentation tasks.