A dual-stream feature extraction network based segmentation method for cell images using SAM and U-Net

Cell image segmentation is a critical task in biomedical studies, especially spatial transcriptomics analysis. Various deep learning-based methods have been applied to medical image segmentation. Among these methods, the U-Net architecture and the Segment Anything model demonstrate several advantages in the analysis process. However, noise and diversity of object shapes in cell images still pose research challenges. This study proposes a novel segmentation method for cell images, called CellSamUnet, which is based on a dual-stream feature extraction network that utilizes an effective fusion of SAM2 and U-Net encoders. Experimental results on well-known benchmark datasets demonstrate that the proposed method outperforms existing baseline methods. The source codes and datasets used in this work can be downloaded from our github page https://github.com/dinhnguyen1412hcm/CellSamUnet.