Translating widefield microscopy images into the 3D using neural networks

Understanding the 3D structure of biological entities is crucial for gaining mechanistic biomedical knowledge. A confocal light microscope is a well-established tool used to obtain 3D data from biological specimens. Yet, it comes with the drawbacks of high equipment prices and heavy human labor. In this project, we introduce a 3D focal stacking solution using deep neural networks (DNN). Instead of restoring 3D models from confocal microscopes, our model produces in-focus images by inputting widefield microscope images, which may be obtained with significantly simpler equipment. This enables the translation from widefield microscope images into the 3D model by segmenting the in-focus pixels, allowing the image of 3D biological specimens in vivo.