Deep learning algorithms have demonstrated significant potential in segmenting diverse tissue morphologies within histopathological whole-slide images. However, the availability of these datasets is limited, and obtaining expert annotations is often time-consuming and difficult. Consequently, training precise segmentation algorithms poses a considerable challenge. To address this, we introduce Simulated Segmentation with Masked Patches (SSMP), a novel method for training segmentation models that relies on weak or incomplete labels. This approach can significantly reduce model development time from days to hours by bypassing the labor-intensive annotation process. In few-shot settings, our method has shown promising results on datasets such as Camelyon16, MSKCC Lymphnode, and BCNB, achieving fewer false positives compared to existing weakly-supervised methods. Moreover, on large lymph node metastatic datasets like Camelyon16 and MSKCC, SSMP surpasses MIL-based algorithms in whole-slide image classification tasks, particularly in scenarios with limited data.