Adaptive Collaborative Learning Polyp Segmentation Framework Based on Contour-Driven Parallel Attention Network

Automatic polyp segmentation plays a crucial role in the early diagnosis and treatment of colorectal cancer (CRC). However, this technology often faces challenges such as complex polyp morphology, unclear boundaries, and limited generalization capability of the algorithms. In response, a novel end-to-end polyp segmentation framework incorporating a parallel attention mechanism was proposed. The backbone collaborates with an adaptive active contour model that dynamically updates model evolution parameters during training to achieve optimal performance. This strategy enables mutual reinforcement and supervision between models. Additionally, the fractal geometric properties of the object were accounted during training, which reduces image noise while enhancing the model's sensitivity to polyp morphology. Our algorithm outperforms current state-of-the-art methods on multiple authoritative datasets, demonstrating significant advancements in both accuracy and generalization, particularly in handling polyps with complex boundary information.