Abstract: The automatic segmentation technology of medical images has the function of assisting clinical medical diagnosis. In order to improve the CNN model in medical image segmentation, such as small receptive fields and insensitive detail features, a multi-scale integrated attention U-shaped network architecture is proposed to improve medical image segmentation quality, which is based on multi-scale strategies and attention mechanisms. First, a new dual-path factorization multi-scale fusion block is proposed to expand the receptive field of image features and further extract the detailed information of image features. Secondly, integrate channel and space fusion self-attention blocks into the architecture, and use the characteristics of the attention mechanism to suppress irrelevant parts or background to highlight the spatial information of deep features. Finally, a multi-scale attention block is introduced. The module fuses feature information of multiple scales to highlight the most prominent feature maps in different scales to adapt to the size of the current segmentation object. In order to verify the reliability of the model, the proposed network model is applied to medical image segmentation tasks such as lungs, cell contours, and liver. Experimental results show that the proposed method is superior to the current mainstream methods for medical image segmentation in terms of accuracy, Dice coefficient, AUC, and sensitivity. It is the mainstream method currently used for medical image segmentation.