In autonomous driving scenarios, a single trajectory often fails to fully capture the complexity of human driving behavior. As a result, an increasing number of approaches have adopted diffusion models to generate multimodal future trajectories. However, these methods generally struggle with trajectory accuracy, neglect safety constraints in the environment, and lack interpretability. To address these limitations, we propose a novel trajectory planning framework that utilizes a transformer-based diffusion model to generate multimodal trajectories, guided by the topology information of road. Our framework integrates structural constraints to enhance safety, improve trajectory accuracy, and provide greater interpretability by ensuring alignment with high-level navigation goals.