Answers are grounded in this paper and cite their sources.
Foundation models have advanced microscopy image segmentation, but existing models are trained on two-dimensional data, struggle with complex structures in volumetric data, and address only a narrow range of analysis tasks. Here we present Segment Anything for Microscopy 2 (μSAM2), a foundation model that adapts SAM2 to joint 2D and 3D segmentation in light and electron microscopy. Trained on annotations of more than 23 million cells and nuclei and 1.6 million cellular structures, μSAM2 achieves state-of-the-art automatic and interactive segmentation and outperforms generalist and specialist methods on volumetric data. Dedicated models further enable accurate mitochondrion and nucleus segmentation in electron microscopy. Beyond segmentation, μSAM2 supports pixel and object classification from few annotations as well as cell tracking in 2D and 3D time-lapse data. All functionality is available in the user-friendly μSAM napari plugin. We demonstrated the capabilities of μSAM2 on challenging use cases spanning developmental biology, organoid research, cardiac physiology, and histopathology.
No takes yet. Share an insight, caveat, or question.
Archit et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: