Deep Learning Based Supervised Semantic Segmentation of Electron Cryo-Subtomograms

摘要: Cellular Electron Cryo-Tomography (CECT) is a powerful imaging technique for the 3D visualization of cellular structure and organization at submolecular resolution. It enables analyzing native structures macromolecular complexes their spatial inside single cells. However, due to high degree structural complexity practical limitations, systematic recovery CECT images remains challenging. Particularly, macromolecule likely be biased by its neighbor molecular crowding. To reduce bias, here we introduce novel convolutional neural network inspired Fully Convolutional Network Encoder-Decoder Architecture supervised segmentation macromolecules interest in subtomograms. The tests our models on realistically simulated data demonstrate that new approach has significantly improved performance compared baseline approach. Also, proposed model generalization ability segment do not exist training data.