Shape Prior Enhanced Unet Model for Medical Image Segmentation

Medical image segmentation is a critical and challenging task in modern medicine. This task faces issues such as low contrast between the target structures and surrounding tissues, as well as uncertainties in the morphology and location of pathological tissues. These challenges often result in the suboptimal performance of traditional segmentation methods, especially when dealing with complex structures. To address these issues, we propose a UNet model integrated with shape priors. This model utilizes a Shape Prior Module (SPM) to generate and update shape knowledge, which is then integrated as feature maps into the skip connections of the U-shaped network. By combining global and local information, the SPM enhances the network's comprehensive understanding of the image. Additionally, this module can be seamlessly integrated with the convolutional neural network architecture, enhancing the model's practicality and generalization capabilities. Moreover, we design a comprehensive loss function that integrates Dice, CE, and BoundaryDoU metrics, significantly improving the efficiency of model parameter optimization and learning. Experimental results on two medical image datasets demonstrate that the proposed shape prior network model yields better outcomes in enhancing segmentation accuracy and improving classification detection precision.