TransNet: Advancing Colonoscopy Polyp Segmentation Through Transformer Integration

Colorectal cancer remains a global health concern, necessitating innovative solutions for early detection. In response, we present a novel polyp segmentation model for colonoscopy images, uniting the Vision transformer and Swin transformer architectures. Our model demonstrates significant segmentation accuracy improvements, validated through extensive evaluation on prominent public datasets, consistently outperforming existing models. A key factor in our model’s success is the integration of a Convolutional Neural Network (CNN)-based decoder network, enhanced with a Convolutional Block Attention Module (CBAM) for efficient feature map upsampling. This integration marks a pioneering advancement in polyp segmentation techniques. The superior performance of our model, validated by high metric values, positions it as a robust tool for practical applications in colorectal cancer diagnosis.