Main Menu (Mobile)- Block

Main Menu - Block


janelia7_blocks-janelia7_secondary_menu | block
More in this Project Team Landing Page
custom_misc-custom_misc_lab_updates | block
node:field_content_summary | entity_field

With the increasing availability of large-volume, high-resolution datasets made possible via Focused Ion Beam-Scanning Electron Microscopy (FIB-SEM) there is also an expanding need to extract the multitude of information stored within these data-rich volumes. 

node:body | entity_field

In order to analyze these data, segmentation is first required. However, a purely manual segmentation is too time-consuming for large volumes. Thus, it is unfeasible to segment every substructure within an entire FIB-SEM volume, and an opportunity is missed.

We are developing tools for automated identification of all intracellular substructures within isotropic FIB-SEM data. We trained a deep neural network to directly and simultaneously predict signed boundary distances to the nearest object boundary of 36 classes of cellular substructures. Naive thresholding of these predictions at zero produces a promising initial segmentation of cellular substructures, while different thresholds form a component tree.  As a convenient side effect, boundary distance predictions also allow to immediately extract the distance between object instances, providing organelle-organelle contact sites for free. This project provides a platform for the automated segmentation and analysis of numerous FIB-SEM datasets across many cells types within different tissues and from different species.

Whole-cell FIB-SEM