Deep learning has emerged as a transformative approach in the realm of polyp segmentation within colonoscopy images. In the pursuit of advancing this segmentation approach, this study introduces a novel deep learning architecture that synergistically combines the concept of EfficientNet-B7 and U-Net and propose a new architecture named EU-Net. The EfficientNet-B7, recognized for its scalable and powerful feature extraction capabilities, serves as the encoder of EU-Net. In contrast, the U-Net, renowned for its capability in biomedical image segmentation, functions as the decoder. The prime contribution of the proposed model lies in the analysis of different attention mechanisms before the concatenation of skip connection and the upsampling feature. This attention mechanism enhances the model’s focus, ensuring it retains critical regions within images, thereby boosting the accuracy of segmentation tasks. Furthermore, the EU-Net model is enriched with dilated convolutions, enabling it to capture multi-scale context without the loss of resolution, a paramount feature for detecting polyps of varying sizes. The proposed model was rigorously evaluated using the CVC-ClinicDB, a benchmark database for colonoscopy image analysis. The proposed model achieves an IoU of 0.8863. Preliminary results showcase a substantial improvement in segmentation performance, highlighting the potential of the proposed Eu-Net in advancing colorectal cancer prevention efforts.