Cell organelle segmentation in electron microscopy
HHMI Janelia Research Campus // Ashburn, USA
Q&A Session I // Wednesday, October 28 // 6.15 pm – 7 pm (CET)
Datasets of cells and tissue imaged using modern high resolution 3D electron microscopy protocols, such as Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) show a plethora of cellular organelles and sub-cellular structures. However, extracting this multitude of information using a purely manual segmentation is too time-consuming, even for a single cell. We are therefore working towards machine learning based solutions to extract a comprehensive map of all known cellular organelles across a variety of cell types and ultimately within different tissues from different species.
We trained deep neural networks to directly and simultaneously predict signed boundary distances of 36 classes of cellular substructures. Naive thresholding of these predictions at zero produces a promising initial segmentation of cellular substructures. This data enables us to automatically align correlative light – electron microscopy images as well as address a variety of biological questions.