RESPA-Net: An Efficient Model for Polyp Image Segmentation Based on Multi Residual and Attention Mechanism

Colorectal cancer is a type of cancer that attacks the colon and rectum. This disease is characterized by the appearance of polyps that can develop into cancer over time. Early detection of polyps is crucial to prevent the possibility of developing into cancer, but manual colonoscopy involving experts is prone to human error. Therefore, a deep learningbased solution is needed to improve detection accuracy. However, the previous method has many parameters, increasing the computational load. This study presents RESPA-Net, a Half-UNet-based architecture enhanced with Multi-Residual Block, Squeeze-and-Excitation (SE) Block, and Pixel Attention to improve segmentation accuracy while ensuring computational efficiency. This model was trained and evaluated using CVCClinicDB, producing a dice score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 4. 3 5 \%}$</tex> and an IoU of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 9. 6 3 \%}$</tex>, outperforming previous methods such as U-Net, UNet++, Half-UNet, ColonSegNet, and DoubleUNet. In addition, RESPA-Net has only 0.44 million parameters, making it an efficient and accurate solution for polyp segmentation.