<p>Segmentation of the polyps in colonoscopy images is vital in the early diagnosis and prevention of colorectal cancer. Despite the advances made by current deep learning models, the accuracy and efficiency of the segmentation can still be improved. This paper presents an attention-enhanced network with attention gates and Squeeze-and-Excitation (SE) blocks for polyp segmentation. In the proposed model, we adopted the VGG16 architecture (without FC layers) as our encoder, training all weights from scratch. The decoder part also has the U-Net structure, and uses skip connections to maintain high-resolution details. Attention gates are incorporated at each level of the decoder to learn and filter out irrelevant features. Furthermore, SE blocks are used to re-normalize channel-wise feature responses and enhance the useful features while suppressing the less useful ones. The performance of the proposed model has been assessed on the Kvasir-SEG data set, which is widely used for polyp segmentation. The proposed approach achieved a Dice score of 0.8549, demonstrating superior segmentation performance compared against state-of-the-art methods. Notably, the proposed model surpassed the baseline architecture in the DSC with an improvement of 17.92%. The proposed model’s balance of high accuracy and real-time performance (&#xa0;82ms/image) shows clinical deployment potential.</p>