The research of cell segmentation recently heavily relies on supervised algorithms, which in turn depend on extensive manual annotations. However, due to the abundance and density of cells in pathological images, manual annotation tasks are both labor-intensive and inefficient. While previous efforts have successfully reduced the algorithm's reliance on labels, they still require cumbersome point annotations or rely heavily on staining conditions, leading to limited generalization. In this paper, a cell detection algorithm is proposed, which can be used to classify and segment cells in pathological images by using an unsupervised region segmentation algorithm to guide the generation of point prompts for the segmentation of the foundation visual model SAM. We evaluated this algorithm on the BCDataset, MoNuSeg and PanNuke datasets, demonstrating its ability to achieve cell segmentation without meticulous manual annotations. The results indicate that our algorithm can inspire further research avenues and facilitate the detection of a broader range of cell types.