Image Segmentation of the Cell Nuclei Staining Using U-net and Segment Anything Model

Zifan Zuo
DOI: https://doi.org/10.1145/3638584.3638606
2023-12-08
Abstract:Cell nuclei examination plays a pivotal role in biological research due to their primary responsibility for housing genetic information, as well as providing a major landmark for identifying a cell. This study introduces an automated approach for studying cell nuclei morphology, leveraging machine learning models to identify their forms and yield insights for further investigation. The principal challenge addressed herein is the time-intensive requirement of manually labeling raw images in the training set. Our proposed solution employs Meta's recently developed segment-anything model (SAM) to expedite this process. The SAM model offers several auto-segmentation methodologies, including full-automatic segmentation and point-as-input segmentation. Our experiments indicate that the point selection mode offers the most promising results. Its accuracy matches human labeling and significantly reduces the time investment and presents scope for enhancement through modification specific to biomedical image processing. By adopting the SAM model to the biology domain, this work opens avenues for further automation in cell nuclei study with potential generalization capabilities.
Computer Science,Biology
What problem does this paper attempt to address?