DA-Diff: Combining Dual Attention with Diffusion Models for 3D Medical Image Segmentation

This paper to address the limitations of existing medical image segmentation methods in effectively extracting spatial and channel features while suppressing noise interference. We propose novel Dual-Attention Denoising Diffusion model (DA-Diff) employs a denoising diffusion process to mitigate image noise interference, utilizes spatial and channel attention mechanisms to collaboratively extract features from guided images, and adopts a skip connection with attention-gated fusion strategy for feature refinement. Experimental results on the multimodal brain tumor Magnetic Resonance Imaging (MRI) dataset and multi-organ Computed Tomography (CT) volumetric dataset demonstrate that DA-Diff achieves superior performance in both Dice score and HD95 metrics compared to existing state-of-the-art models, confirming that the synergistic design of dual attention mechanisms and diffusion models effectively reduces noise impact while addressing insufficient feature extraction in medical images, significantly enhancing the recognition accuracy of complex anatomical structures in medical image segmentation tasks.