Enhancing Polyp Segmentation Efficiency Using Pixel Channel Attention HalfU-Net

Polyps, though often benign, can lead to colorectal cancer, which accounts for 9.4% of cancer deaths worldwide. Early detection through colonoscopy is vital, but current methods are limited in accuracy and time efficiency. Deep Learning (DL) has demonstrated superior performance in medical image segmentation, with U-Net being a widely adopted architecture. However, U-Net++ and similar architectures involve significant computational complexity. To address this, Half-UNet was proposed, offering a significant reduction in model parameters without sacrificing performance. Building on these advancements, this study introduces the Pixel Channel Attention Half-UNet (PCA HalfU-Net), which integrates parallel attention mechanisms to improve segmentation accuracy while reducing computational demands, aiming for more effective and efficient polyp detection. Our model outperforms U-Net, improving DSC and IoU metrics by 94.6% and 90.1%, respectively. In addition, the number of parameters in the proposed approach is significantly lower than that of previous models, amounting to only 0.266 million. This indicates that PCA HalfU-Net offers an optimal balance between computational complexity and segmentation accuracy. This compact architecture enhances its suitability for deployment in environments with limited computational resources without compromising the overall effectiveness of the polyp segmentation task.