Scientists used machine learning to create a label-free tool for predicting 3D fluorescence directly from transmitted-light images. Instead of fluorescence microscopy, the new method only uses black-and-white images from a bright-field microscope.
Cells viewed through a bright-field microscope appear to the naked eye as gray sacs. A trained scientist can locate the cell’s edges and nucleus but not much else. Using a convolutional neural network, the team trained computers to recognize details in bright-field images. The researchers tested their technique on 12 different cellular structures and demonstrated that it could be used to generate multi-structured, integrated images. For most of the structures, the model generated by the machine learning algorithm predicted images that matched the fluorescently labeled images.
According to the researchers, the new label-free method can also predict immunofluorescence (IF) from electron micrograph (EM) inputs, expanding the potential applications.
The tool could aid scientists in understanding what goes wrong in cells during disease. Cancer researchers could use the technique on archived tumor biopsy samples to learn more about how cellular structures change as cancers progress or respond to treatment. The algorithm could also help regeneration medicine by revealing how cells change in real-time as researchers try to grow organs or other new body structures in the lab.