Deep learning architectures based on convolutional neural networks have achieved remarkable success in the domain of medical image segmentation and have gained widespread application in practical scenarios. However, the escalating complexity of medical images presents novel challenges to current methodologies. In this study, we propose an innovative architecture named the Attention-based Multi-Scale Feature Conservation Network (AMFCNet), explicitly tailored for medical image segmentation. Leveraging the characteristics of attention mechanisms, AMFCNet incorporates Efficient Channel Attention (ECA), utilizing channel attention mechanisms to concentrate the model on the crucial feature channels for segmentation tasks. We integrate the Receptive Field Block (RFB) module to enhance the network’s receptive field, allowing the model to adapt more effectively to features at different scales and levels. Additionally, we devise the Primary Feature Conservation (PFC) module, dedicated to capturing the primary features of images to attain more precise segmentation results. Comprehensive experimental results affirm that AMFCNet surpasses state-of-the-art methods in medical image segmentation. The proposed model underwent evaluation on the ISIC2018, BUSI, CVC-ColonDB, and CVC-300 datasets, achieving Dice similarity coefficients of 91.19%, 81.72%, 91.12%, and 90.30%, respectively.