Weakly supervised cell instance segmentation under various conditions

Cell instance segmentation is important in biomedical research. For living cell analysis, microscopy images are captured under various conditions (e.g., the type of and cell). Deep-learning-based methods can be used to perform if sufficient annotations individual boundaries prepared as training data. Generally, required for each condition, which very time-consuming labor-intensive. To reduce annotation cost, we propose a weakly supervised method that segment regions by only using rough centroid positions This dramatically reduces cost compared with standard segmentation. We demonstrated efficacy our on images; it outperformed several conventional weakly-supervised average. In addition, without any manual pairs phase contrast fluorescence nuclei stained