A convolutional autoencoder approach for mining features in cellular electron cryo-tomograms and weakly supervised coarse segmentation
share
Cellular electron cryo-tomography enables the 3D visualization of cellular organization in a near-native state at submolecular resolution. However, the content of a cellular tomogram is often complex, making it difficult to automatically isolate different in situ cellular components. In this paper, we propose a convolutional autoencoder-based unsupervised approach to provide a coarse characterization of 3D patches extracted from tomograms. We demonstrate that the autoencoder can be used for the efficient and coarse characterizing of features that correspond to macromolecular complexes and surfaces, like membranes. In addition, it can be used to detect non-cellular features related to sample preparation and data collection like carbon edges from the grid, and tomogram boundaries. The autoencoder is also able to detect patterns that may indicate spatial interactions between cell components. Furthermore, we demonstrate that our autoencoder can be used for weakly supervised semantic segmentation of cellular components requiring very small amount of manual annotation.
READ FULL TEXT