Deep neural network automated segmentation of cellular structures in volume electron microscopy
Recent advances in automated segmentation using deep neural network models allow identification of intracellular structures. This study describes a new pipeline to train a convolutional neural network for rapid and efficient detection of structures of wide range in size and complexity. AbstractThree-dimensional electron microscopy is an important imaging modality in contemporary cell biology. Identification of intracellular structures is laborious and time-consuming, however, and impairs effective use of a potentially powerful tool. Resolving this bottleneck is therefore a critical next step in frontier biomedical imaging. We describe Automated Segmentation of intracellular substructures in Electron Microscopy (ASEM), a new pipeline to train a convolutional neural network to detect structures of wide range in size and complexity. We obtain for each structure a dedicated model based on a small number of sparsely annotated ground truth annotations from only one or two cells. To improve model generalization to different imaging conditions, we developed a rapid, computationally effective strategy to refine an already trained model by including a few additional annotations. We show the successful automated identification of mitochondria, Golgi apparatus, endoplasmic reticulum, nuclear pore complexes, caveolae, clathrin coated pits and coated vesicles in cells imaged by focused ion beam scanning electron microscopy with quasi-isotropic resolution. ASEM enabled us to uncover a wide range of membrane-nuclear pore diameters within a single cell and to derive morphological metrics from clathrin coated pits and vesicles at all stages of maturation consistent with the classical constant-growth assembly model.