Automated and accurate segmentation of medical images is important for facilitating clinical diagnosis and treatment. Currently, state-of-the-art(SOTA) diffusion based medical image segmentation methods are hampered by inherent randomness when generating diffusion model outcomes. Multiple generations are required to mitigate this randomness, which present a challenge for diffusion models. Due to the extended inference time required by diffusion models for multistep iterations, the process of obtaining final segmentation results by multiple generations is prolonged. As a result, this hampers the application of diffusion models in the medical field and limits the research potential of these models. In this paper, we present a medical image segmentation framework that concurrently predicts labels and noise. By leveraging label constraints within the diffusion model, we effectively suppress randomness, enabling the generation of segmentation maps with reduced errors in the initial stages and thereby suppress randomness. The performance of the proposed method is assessed using the ISIC2016 and Brats2018 datasets. Our approach necessitates just a single generation to produce effective segmentation results without the need for multiple steps to mitigate randomness and outperforms compared SOTA methods.