Semi-Supervised Semantic Segmentation of Cell Nuclei with Diffusion Model

Accurate segmentation of cell nuclei in microscopic images is important for disease diagnosis and tissue microenvironment analysis. Supervised methods have shown promise but need large annotated datasets. Semi-supervised strategies help by incorporating unlabeled data, though performance can remain limited when labeled data are scarce. We propose a diffusion model-based semi-supervised framework that can use diverse unlabeled data in an unsupervised manner. Our method, called DTSeg, employs a latent diffusion model (LDM) and a transformer-based decoder. The LDM learns from varied unlabeled images, while the decoder is trained in a supervised manner with limited labeled data. Experiments on three datasets show that our method achieves better segmentation performance than existing semi-supervised approaches.