Identifying and delineating cellular structures in microscopy images is crucial to understand the complex processes of life. This task is called “segmentation” and allows a range of applications, such as the analysis of the reaction of cells to pharmacological treatments or comparing cellular structures in different genotypes. It was already possible to carry out the automatic segmentation of these biological structures, but the dedicated methods only worked in specific conditions and adapt to new conditions was expensive. An international research team led by the University of Göttingen has now developed a method to re -train the software segment based on the existing in more than 17,000 microscopy images with more than 2 million structures noted by hand.
Its new model is called segment anything for microscopy and can accurately segment the images of similar tissues, cells and structures in a wide range of configurations. To make it available to researchers and doctors, they have also created μsam, easy to use software to “segment anything” in microscopy images. His work was published in Methods of nature.
To adapt the existing software to the microscopy, the research team first evaluated it in a large set of open source data, which showed the potential of the model for microscopy segmentation. To improve quality, the equipment recovered it in a large microscopy data set. This drastically improved the model performance for the segmentation of cells, nuclei and small structures in cells known as organelles. Then, the team created its software, μsam, which allows researchers and doctors to analyze images without the need to first paint structures manually or train a specific AI model. The software is already widely used internationally, for example, to analyze nerve cells in the ear as part of an auditory restoration project, segment artificial tumor cells for cancer investigation or analyze images of electronic microscopy of volcanic rocks.
“Analyzing cells or other structures is one of the most challenging tasks for researchers working in microscopy and is an important task for both basic biology research and medical diagnosis,” says Professor Junior Constantin Pape at the Institute of Sciences of the computer of the University of Göttingen. “My group specializes in the creation of tools to automate such tasks and often the researchers ask us to help. Before the development of segment of anything for microscopy, we had to ask them to first write down many structures by hand, a difficult and difficult time – The consumption task.