D3EU-Net: A Diffusion and DeepSeek-aided Dual Encoder-based U-Net Model for Microscopic Medical Image Segmentation

Abstract

Medical image segmentation plays a pivotal role in automated diagnosis and treatment planning. While deep learning models have shown strong performance in histopathological image analysis, they often struggle with noisy inputs and ambiguous structures. In this work, we present D3EU-Net, a multimodal segmentation framework that integrates diffusion-based denoising with semantic and spatial conditioning to address these challenges. The model employs a transformer-based text encoder to incorporate descriptive prompts generated by a large language model, and a lightweight convolutional spatial encoder to extract prior structural cues from binary masks. These embeddings are fused within a conditioned U-Net architecture through cross-attention modules, enabling context-aware and noise-robust segmentation. Our approach models noise through an iterative diffusion process, acting as a regularizer that improves generalization under diverse conditions. Experiments on TNBC, MoNuSeg and GlaS datasets yield that D3EU-Net achieves Dice scores of 88.46%, 83.16% and 98.67% respectively, surpassing several state-of-the-art (SOTA) results. These results suggest that multimodal guidance, combining textual and spatial priors with diffusion, can enhance segmentation quality and adaptability for microscopic medical images. Additional results are available at https://github.com/SaptarshiPani/D3EU-Net

More by Dmitry Kaplun

• Codewithsayanjib/slm-systematic-review: v1.1Zenodo · 2026
• Microscopy Image Dataset of pulmonary vessels for Quantitative assessment of fibrosisScientific Data · 2026
• Microscopy Image Dataset of pulmonary vessels for Deep Learning-Based Quantitative assessment of fibrosisFigshare · 2026
• MA2MU-Net++: A Mahalanobis distance and Ackley function based multi-modal U-Net++ model for breast nuclei segmentation in histopathology imagesBiomedical Signal Processing and Control · 2026