Accurate segmentation of cells, organelles or other structures in microscopy images is a bottleneck for many researchers. While many powerful tools have been proposed for this analysis task, they often focus on a single segmentation task or require extensive training data, making it time-consuming to apply them to new data. Here, I will present Segment Anything for Microscopy (μSAM), our more versatile tool for segmentation and tracking in multidimensional microscopy data. It is based on Segment Anything, a vision foundation model for image segmentation. We have extended it by training generalist models for light and electron microscopy, which clearly improves segmentation quality across a wide range of imaging conditions. We have also implemented interactive and automatic segmentation in a napari plugin that can speed up diverse segmentation tasks and provides a unified solution for microscopy annotation across different microscopy modalities. I will especially show its capabilities for segmenting electron microscopy data and our efforts for building even better models for this imaging modality.
Constantin Pape studied physics in Heidelberg, where he also pursued his PhD under the supervision of Anna Kreshuk at EMBL and Fred Hamprecht at the University of Heidelberg. During his PhD he developed efficient image analysis methods for EM connectomics and volume EM. During this time, he spent a year as a visiting scientist at Janelia Research Campus. He started as an independent group leader at the University of Göttingen in 2022. His group is developing modern deep learning methods for image analysis, targeting applications ranging from high-content screening microscopy for clinical diagnostics to organelle analysis in volume EM and in-situ protein identification in cryogenic electron tomography.
This webinar is presented by Volume Imaging Australia, a special interest group of the Australian Microscopy and Microanalysis Society (AMMS), and Microscopy Australia.
Image from Archit, A., et al., Segment Anything for Microscopy. Nature Methods (2025). DOI: 10.1038/s41592-024-02580-4
