Breast cancer is a leading cause of mortality among women worldwide, and accurate tumor detection in histopathology images is important for early diagnosis and effective treatment. Traditional manual annotation methods are time-consuming, subjective, and prone to variability, necessitating the development of automated and precise segmentation techniques. In this study, the Dual-Path Encoder-Decoder (DPED) model is proposed for tumor segmentation in breast histopathology images. The DPED architecture consists of a Detail Path, which captures fine-grained textures, and a Contextual Path, which extracts global semantic information. A Feature Fusion Module incorporating spatial and channel attention mechanisms enhances feature integration, while a Dual-Stage Decoder refines tumor boundaries with boundary-aware upsampling. The model is trained using AdamW and SGD with momentum, with dynamic learning rate scheduling via cosine decay and polynomial learning rate strategies to ensure stable convergence. The experiments held on BreCaHAD show that the method proposed has a segmentation accuracy of around 97%, with the performance surpassing state-of-the-art segmentation methods. These results indicate the advantage of DPED for improving precision and reliability in the automated diagnosis of breast cancer in histopathological image analysis.