Diff-UNet: A Diffusion Embedded Network for Volumetric Segmentation

Abstract

In recent years, Denoising Diffusion Models have demonstrated remarkable success in generating semantically valuable pixel-wise representations for image generative modeling. In this study, we propose a novel end-to-end framework, called Diff-UNet, for medical volumetric segmentation. Our approach integrates the diffusion model into a standard U-shaped architecture to extract semantic information from the input volume effectively, resulting in excellent pixel-level representations for medical volumetric segmentation. To enhance the robustness of the diffusion model's prediction results, we also introduce a Step-Uncertainty based Fusion (SUF) module during inference to combine the outputs of the diffusion models at each step. We evaluate our method on three datasets, including multimodal brain tumors in MRI, liver tumors, and multi-organ CT volumes, and demonstrate that Diff-UNet outperforms other state-of-the-art methods significantly. Our experimental results also indicate the universality and effectiveness of the proposed model. The proposed framework has the potential to facilitate the accurate diagnosis and treatment of medical conditions by enabling more precise segmentation of anatomical structures. The codes of Diff-UNet are available at https://github.com/ge-xing/Diff-UNet.

A novel LVPA-UNet network for target volume automatic delineation: An MRI case study of nasopharyngeal carcinoma

“…This study has several limitations. The Diffusion Probabilistic Models (DPMs) [ [45] , [46] , [47] , [48] ] have recently shown outstanding performance in the field of medical image segmentation. By simulating the diffusion process between image voxels, DPMs refine image features and capture contextual information, allowing them to implicitly recognize complex patterns and subtle changes within images.…”

DiffSwinTr: A diffusion model using 3D Swin Transformer for brain tumor segmentation

“…To improve the performance of MedSegDiff, Wu et al 35 proposed a conditional Transformer network framework and a SS-Former module to model the information interaction between the semantic features and Gaussian noise. Xing et al 38 proposed a medical volumetric image segmentation method, called Diff-UNet, which introduces the DDPM into the U-Net model to provide excellent pixel-level representation.…”

“…However, for 3D volumetric data such as MRI images, using 2D image slices will lose some of the spatial information. In addition, some diffusion models 34,38 based on U-Net are often unable to capture remote dependencies and extract global spatial feature information from MRI images. To address the above challenges, we design an algorithm based on diffusion model to achieve accurate brain tumor segmentation.…”

Mo1640: PREDICTION OF THE INCIDENCE OF INFECTIOUS DIARRHEA IN CHINA: A NOVEL VAR PREDICTION MODEL BASED ON BAIDU INDEX

“…Although SAM can be applied to each slice of a volume to get the final segmentation, it does not consider the correlation in the depth dimension. Many previous studies have shown that depth correlation is essential for 3D medical image segmentation [14,15,39]. To address this issue, we propose a novel adaptation method inspired by the image-to-video adaptation, with some modifications.…”